The Unified Evaluation App for DNA Data Storage Codecs

Springer Nature 2021 LATEX template

arXiv:2608.09673v1 [cs.ET] 10 Aug 2026

The Unified Evaluation App for DNA Data Storage Codecs Aleksandar Anžel1,2*† , Chisom Anyabolu1,2† , Leon Wimbes2 , Luca Staus2 , Ihsan Tri Heldian2 , David Sonnabend2 , Khawla Elhadri2 , Samuel Becker2 , Felix Klein2 , Raffael Schoen2 , Michael Schwarz2 , Marius Welzel3 , Bernd Freisleben2 , Dominik Heider3 and Georges Hattab1,4 1

Center for Artificial Intelligence in Public Health Research (ZKI-PH), Robert Koch Institute, Nordufer 20, Berlin, 13353, Germany. 2 Department of Mathematics & Computer Science, University of Marburg, Hans-Meerwein-Straße 6, Marburg, D-35032, Germany. 3 Institute of Medical Informatics, University of Münster, Albert-Schweitzer-Campus 1, Gebäude A11, Münster, 48149, Germany. 4 Department of Mathematics & Computer Science, Freie Universität, Arnimallee 14, Berlin, 14195, Germany.

*Corresponding author(s). E-mail(s): anzela@rki.de; † These authors contributed equally to this work. Abstract Background: Deoxyribonucleic acid (DNA) data storage is a paradigm with great potential for ultra-dense and durable information preservation. However, the rapid proliferation of coding schemes, or codecs, each with their own design constraints and reporting practices, has led to a fragmented landscape that lacks a standardized comparative assessment. Methods: We developed an open-source, modular benchmarking platform that systematically integrates and evaluates state-of-the-art DNA storage encoding and decoding methods (codecs). Our approach uses a curated, diverse set of baseline data and applies multidimensional assessment criteria that are aligned with the consensus standard of the

1

Springer Nature 2021 LATEX template 2

The Unified Evaluation App for DNA Data Storage Codecs DNA Data Storage Alliance. These criteria include encoding/decoding throughput, computational efficiency, error correction performance across substitutions, insertions, and deletions, and cost efficiency. Results: The developed platform integrates standardized wrapper functions for encoding and decoding, allows for the integration of new methods, and automates reproducible evaluations with comprehensive visual and tabular reporting. Benchmarking both contemporary and classical codecs using their default parameters and multiple metrics demonstrates that no single algorithm is optimal across all evaluated dimensions. The trade-offs between information density, success rate, runtime, and cost are quantified and shown to be critical factors in the design of future-proof formats. Conclusions: Our work establishes a rigorously standardized, open-source evaluation framework that enables reproducible benchmarking, supports evidence-based codec selection, and provides the necessary foundation for translating DNA data storage from experimental research into deployable archival systems. Keywords: DNA data storage, Coding schemes, Codecs, Benchmarking, Comparative analysis, Performance evaluation, Unified framework, Multi-algorithm comparison, Error correction, Information density, Digital data archiving

1 Introduction In the rapidly evolving field of data storage, deoxyribonucleic acid (DNA) data storage has emerged as a promising solution to meet the global demand for information preservation. By 2030, the amount of data generated globally each year is projected to reach approximately 1 yottabyte (1024 bytes ≈ 1015 GB). To accommodate this growth, global general storage capacity is expected to increase roughly tenfold relative to 2020, reaching approximately 3.7 × 1013 GB. This surge is driven primarily by the exponential growth of artificial intelligence-related data, which is projected to constitute over 60% of enterprise data access and processing by that time [1]. Due to its unparalleled storage density, long shelf-life, potential low maintenance cost, and chemical modification capabilities, DNA offers a viable alternative for the long-term archival of large amounts of data [2, 3]. In addition, DNA is omnipresent and biologically significant. Consequently, as long as life exists, there will be a need to understand and manipulate DNA, ensuring its resilience and preventing it from becoming obsolete. The DNA molecule is made up of four fundamental building blocks of nucleic acids called nucleotides: adenine (A), cytosine (C), guanine (G), and thymine (T). A single strand of DNA, often called an oligonucleotide, represents an ordered sequence of these nucleotides. In effect, it forms a string through the alphabet — {A, C, G, T}. The ability to chemically synthesize

Springer Nature 2021 LATEX template The Unified Evaluation App for DNA Data Storage Codecs

3

any conceivable sequence of nucleotides underpins the potential to store digital information within DNA strands. By encoding binary data in nucleotide sequences, DNA can be used as a high-density storage medium, providing a novel approach to data storage. In this work, we focus specifically on in vitro DNA-based storage systems, rather than in vivo methods that involve storing data within living organisms. As shown in Figure 1, the initial step in the in vitro-based storage process is the conversion of digital files into binary data, which is then encoded into DNA sequences. Additional sequences, including an index, error correction codes, and primers for DNA amplification, are appended to or integrated within the oligonucleotides (oligos). The synthesis of oligonucleotides is typically achieved through chemical or enzymatic processes, resulting in the production of single-stranded DNA. This synthetic DNA is frequently stored in its single-stranded form; alternatively, a complementary strand is synthesized enzymatically, producing double-stranded DNA for storage. The selective extraction or amplification of specific oligos from the pool is achieved through polymerase chain reaction (PCR). Once extracted, the oligos are sequenced using a DNA sequencer, and the obtained nucleotide sequences are subsequently decoded to retrieve the original binary data. Despite its potential, DNA data storage faces significant challenges that impede its widespread adoption [3]. Current methods of writing (DNA synthesis) and reading (DNA sequencing) are notably slower than conventional storage technologies. Furthermore, unlike with sequencing, synthesis costs remain too high for widespread and scalable adoption. Another critical issue is DNA preservation, which requires ensuring the long-term stability and integrity of the encoded information. The use of DNA as a storage medium is limited by appropriate temperature, humidity, pH, and radiation protection, among other factors, which makes this storage solution significantly more complicated than its competitors. Moreover, the processes of DNA synthesis, storage, and sequencing are error-prone, which further complicates the situation. Due to the nature of DNA, these errors fall into one of three categories: insertion, deletion, or substitution of nucleotides (DNA bases). DNA degradation is closely tied to the chemical nature of the four nucleotides. For instance, when the GC content is lower than 40% or higher than 60%, the probability of synthesis and sequencing errors increases. This motivates encoding digital data for long-term DNA archival in a way that satisfies various conditions and constraints, thereby optimizing the DNA data storage medium. As technological hurdles in DNA-based digital data storage are gradually overcome, the field is reaching a pivotal point. Advances in computational power, reduced costs, and accelerated DNA synthesis are establishing DNA storage as a viable alternative to traditional storage media [4]. Several researchers have demonstrated the feasibility of storing and successfully retrieving digital information using DNA molecules. Early works by Church et al. [5] and Goldman et al. [6] were pivotal in bringing DNA data storage to the limelight. With advancements in DNA synthesis and sequencing technology, the end-to-end workflow for DNA-based data storage has been well

Springer Nature 2021 LATEX template 4

The Unified Evaluation App for DNA Data Storage Codecs

established [7]. However, current technology still imposes specific constraints on DNA sequences, such as GC content and the length of homopolymers (a sequence of identical nucleotides repeated multiple times). Addressing these constraints while achieving high storage density is vital for the efficient implementation of DNA data storage systems. Based on these constraints, newer methodologies for using DNA as a storage medium continue to emerge, pushing DNA data storage toward practical applications. These methodologies promise not only larger storage capacity and longer lifespan but also address various information and coding theoretic aspects, such as storage channel capacity and the design of error correction codes (ECC) tailored to specific errors encountered in DNA data storage. ECC development efforts have also targeted constrained coding [8–10], random access in DNA data storage systems [11–13], and image storage solutions [14], enhancing the robustness and versatility of DNA data storage systems. Despite the rapid development of DNA coding schemes (codecs) and researchers’ dedicated efforts to address the challenges of DNA data storage, a fair and unified comparison of these methods is lacking. The only notable exceptions are the works of Ping et al. [15] and Gimpel et al. [16], the former introducing a modular Python library implementing five DNA codecs, and the latter providing a wet-lab evaluation of six DNA codecs. Nevertheless, these works are still limited by different factors. On the one hand, the study by Ping et al. [15] evaluates only the encoding and decoding runtimes of each method, without considering additional performance metrics (e.g., cost, size, etc.). Moreover, the library supports only limited error simulation: substitutions are modeled, while insertions and deletions are not. On the other hand, Gimpel et al.’s benchmarking approach [16] is constrained by a fixed selection of codecs, workflows, and harmonized parameters, limiting generalizability as new algorithms and experimental techniques emerge, while also making it difficult for the community to expand or customize benchmarks as field requirements change. Additionally, their solution does not consider specialized sequence constraints or novel performance metrics, provides limited reproducibility and transparency for users lacking significant technical expertise, and does not include visual analytics or an accessible dashboard interface for rapid inspection and interpretation of benchmarking results. Beyond these studies, newly developed DNA codecs are usually evaluated using a limited set of metrics and compared with a selection of current state-of-the-art approaches. This practice is common in nearly all previously published works in this field, and it persists in even the most recent methods such as DNA-Aeon [17] and DNAformer [18]. The diverse landscape of these codecs, which spans various programming languages, software architectures, and a plethora of parameters of individual codecs, poses challenges for new and existing researchers. These researchers seek to accurately assess the current research landscape and effectively evaluate their newly developed methods. Furthermore, many existing codecs have suboptimal software architecture. This hinders effective program-based

Springer Nature 2021 LATEX template The Unified Evaluation App for DNA Data Storage Codecs

5

validation and impedes efforts to improve these methods. The absence of welldesigned, modular architectures complicates enhancing individual methods and obstructs developing standardized benchmarking tools for the field as a whole. As outlined by the DNA Data Storage Alliance, the establishment of a unifying evaluation framework is of paramount importance [19]. The Alliance’s work emphasizes the necessity for standardized methodologies to systematically assess and integrate DNA codecs. Such standardization enables rigorous comparison across critical parameters including storage density, cost-efficiency, write and read throughput, long-term data stability, encoding and decoding performance, scalability, interoperability with extant technologies, and adaptability to diverse data formats. Adherence to these standardized evaluation criteria is instrumental in advancing the field by fostering consistency, reproducibility, and the effective translation of research innovations into practical DNA data storage solutions. This endeavor seeks to establish a modular, open-source, and standardized user-friendly web app that facilitates an objective comparison of various codecs. The primary objective is to guide further advancements in the field. The proposed app identifies the most promising DNA data storage solutions by addressing known challenges and incorporating various evaluation criteria, promoting their practical implementation. The unified app makes three key contributions to the field: a comprehensive library of existing codecs; a robust benchmarking system that evaluates integrated codecs using various criteria; and an interactive testing module that allows users to simulate using DNA as a data storage medium with custom files and a selection of integrated codecs. Our main goal is to speed up innovation in DNA data storage by working together openly and with the community. By encouraging contributions from researchers and developers, we aim to cultivate a dynamic ecosystem that evolves alongside the field and promotes standardization and reproducibility in DNA data storage research.

2 Methods Our work proposes a unified evaluation app for DNA data storage codecs. The app is primarily designed to evaluate the characteristics of different encoding/decoding strategies based on the same baseline data set. To facilitate evaluation and user access, we transformed the original idea into a web application. The benchmark page and file upload page are two integral parts of the web application and core functionalities of the unified app. The former implements the evaluation of integrated codecs using a baseline data set. The latter enables users to evaluate integrated codecs using a file they uploaded, with the added option of simulating the long-term feasibility of storing DNA data. This section provides an in-depth look at the web app’s user interface, workflow, integrated codecs, evaluation system, and file upload functionality.

Springer Nature 2021 LATEX template 6

The Unified Evaluation App for DNA Data Storage Codecs Chemical

primer index

Payload

ECCs primer

Encoding into DNA

DNA synthesis

...ATCGAGTCT... Enzymatic 01101000111 10101010101 11001010100

Digital les

Original data

DNA Storage

Binary data DNA pool

Sequencing & Decoding

Data Retrieval

Illumina AGT

Selective PCR

GA T AT C

Nanopore

Fig. 1: Overview of the processes involved in a DNA data storage system. The DNA data storage process involves reading digital files in their binary format, encoding them into DNA sequences, synthesizing the DNA, storing it in a single- or double-stranded form, sequencing the DNA, and decoding it to retrieve the original binary data. Created in Biorender https: //www.biorender.com/.

2.1 User Interface and Workflow The unified application is presented through a web interface that is interactive, dynamic, and user-friendly. This interface is designed to guide users seamlessly through the stages of using DNA as a data storage medium. The interface consists of five pages: Home, Benchmark, File Upload, About, and Contact Us. The Home and Contact Us pages provide essential information about the app, including ways to contribute and sources of support. The About page contains the information about the benchmark data, evaluation metrics, and supported codecs. The Benchmark and File Upload pages encapsulate the core functionalities of the application. The following text focuses on these two functionalities and describes the user interface and workflow within these two pages.

2.1.1 Benchmark Page The benchmark page presents the evaluation results of integrated codecs defined in subsection 2.2, using the baseline data set and evaluation criteria outlined in subsection 2.3. The results presented on this page are calculated using the procedure outlined in subsubsection 2.1.2. Upon accessing the page, users are presented with a summary benchmark data set given in a tabular format, containing aggregated evaluation results across multiple runs. An interactive expander element containing the detailed benchmark results in a tabular format is located immediately below the summary benchmark.

Springer Nature 2021 LATEX template The Unified Evaluation App for DNA Data Storage Codecs

7

The purpose of the expander element is to reduce cognitive load by omitting the larger detailed benchmark data set from immediate view. By following the well-known visualization mantra “Overview first, [. . . ], then details-ondemand” by Schneiderman [20], users are not overwhelmed by the details but can still inspect them closely if necessary. In addition, users can interact with the tables by clicking the cells containing the column names to easily sort the tables in either ascending or descending order based on the values in that column. Below both benchmark tables are buttons that allow users to easily download them in CSV format. Below the expander containing a detailed benchmark table lies the first visualization, which presents the data information density of each codec across all runs and all data. This ratio is visualized using a whisker chart to capture the variance of information density for each codec. The chart includes interactive elements such as linked zooming and tooltips, encouraging deeper exploration of the results. The whisker representation mirrors the volatility plots proposed in the DNA Data Storage Alliance’s white paper for throughput benchmarking, whereby inter-quartile ranges reveal violations of the “encode faster than synthesis” heuristic [19]. The final section of the benchmark page provides a multi-select element, a chart type selector, and two small-multiple charts — one visualizing the runtime and the other visualizing the costs of each codec during both the encoding and decoding process across the selected benchmark files. The multiselect element is by default loaded with all files that are part of the benchmark data set; however, users can selectively analyze only some of the files using this widget. Immediately below the multi-select element lies a chart type selector that allows users to choose between whisker and bar charts. The selection is then reflected in both small-multiple charts below. Each version of a small-multiple chart depicts a different file from the benchmark data set. Within each small multiple, codecs are placed on the xaxis, while either time or cost is placed on the y-axis. For each codec, colors of whiskers or bars indicate either an encoding or decoding process. Additionally, users can easily download all chart in either vector or raster image formats. The benchmark page, with all elements mentioned earlier, is shown in Figure 2.

2.1.2 File Upload Page While the benchmark page showcases precalculated evaluation results derived from the workflow accessible on the file upload page, the file upload page enables users to interactively explore the processes of DNA data encoding, storage, and decoding in real-time. Therefore, we structure the current section into three distinct parts, each addressing a different segment of the pipeline. The pipeline encompassing both benchmark and file upload functionalities is depicted in Figure 4 and given as a supplementary file. The file upload page, with all elements described in subsequent sections, can be seen in Figure 3.

Springer Nature 2021 LATEX template 8

The Unified Evaluation App for DNA Data Storage Codecs ︱Base Error Method DNA_Aeon

︱Benchmark

LTCode

OnlineCode

RaptorCode

RepetitionCode

SimpleCode

The2DCode

Filename=The_Picture_of_Dorian_Gray

100

XORBasedCode

Filename=Vue_de_la_Buena_Vista

)%( etaR sseccuS

50

0 0

0.02

0.04

0.06

0.08

0

0.02

0.04

Filename=all_black

100

0.06

Filename=bbc_clock

)%( etaR sseccuS

50

0 0

0.02

)%( etaR sseccuS

download Download Benchmark Data

0.04

0.06

0.08

0

0.02

Filename=nature

100

0.04

0.06

Filename=Around_the_World_in_Eighty_Days

50

0

search_insights Expand Detailed Information

0

0.02

0.04

0.06

0.08

0

0.02

0.04

Base_error_ratio

0.06

Base_error_ratio

︱Detailed Charts per File and Error Multiplier

︱Information Density

Choose error multiplier value or leave it 0 to skip 0.0 12

)tn/b( ytisneD_noitamrofnI

10

help What is error multiplier? expand_more

8

Select the files to review (This will affect charts below)

6

The_Picture_of_Do…

4

nature

Vue_de_la_Buena_…

Around_the_World…

Zero_File

all_black

bbc_clock

egypt

bbc_peregrine_falcon

2 0 DNA_Aeon

LTCode

OnlineCode

RaptorCode

RepetitionCode

Success Rate

Error Multiplier: 0.0

Method

︱Success Rate

100 80

)%( etaR sseccuS

100 80

)%( etaR sseccuS

60

60 40 20

55.97%

0 DNA_Aeon

40 20

OnlineCode

RaptorCode

RepetitionCode

5.33%

0 LTCode

OnlineCode

Method

7.69% DNA_Aeon

LTCode

27.94%

19.50%

RaptorCode

RepetitionCode

Method

Filename=The_Picture_of_Dorian_Gray

Filename=Vue_de_la_Buena_Vista

100 80

Fig. 2: Overview of the Benchmark page. The benchmark page contains performance evaluation of various DNA codecs. The results are presented in both tabular and graphical formats, allowing users to compare encoding efficiency, speed, and other relevant metrics comprehensively. DNA Encoding Process Upon accessing the file upload page, users are prompted to upload a file of their choice, which can be in any format. Upon upload, the file is automatically converted into a NumPy [22] boolean array, which emulates the file’s binary representation. This approach, while inherently less efficient in terms of speed and memory usage compared to utilizing a true binary format, is deliberately chosen to ensure compatibility across all codecs. By standardizing this representation, we establish a universal benchmarking framework that facilitates fair and consistent evaluation of diverse coding techniques. Following the upload and binary transformation, users are presented with a drop-down menu that contains a list of DNA codecs. Users can choose one or more DNA codecs integrated into the app and described in subsection 2.2, allowing them to easily explore and evaluate different coding strategies for the uploaded file. The selection of codecs initiates the encoding process. The binary representation of the original uploaded file (defined in a NumPy array) is parsed as input to the selected DNA codecs. The process runtime depends greatly on the size of the uploaded file and the number and type of selected DNA codecs. As a result, each DNA codec outputs a list of DNA sequences, with the number of sequences and their structure varying significantly between codecs. Each DNA sequence is represented as a string where each character is an element of the alphabet {A, C, G, T }. Moreover, for each codec, this step in

Springer Nature 2021 LATEX template The Unified Evaluation App for DNA Data Storage Codecs

9

︱File Upload Upload a file

 Upload

200MB per file

 Your data is automatically deleted afer changing the tab or closing the window.

︱DNA Encoding, Storage, and Decoding Select preferred DNA Coding Schemes Choose DNA Coding Schemes

The length of processing time depends on chosen methods

Error Simulation Choose length of storage (in months) or leave it empty to skip (uses MESA API)

Choose error multiplier value or leave it 0 to skip 0.00

Aleksandar Anžel, Chisom Anyabolu, Leon Wimbes, Luca Staus, Ihsan Tri Heldian, David Sonnabend, Khawla Elhadri, Samuel Becker, Felix Klein, Raffael Schön, Marius Welzel, Michael Schwarz, Bernd Freisleben, Dominik Heider, and Georges Hattab (2026) The Unified Evaluation Tool for DNA Data Storage Codecs Journal Name. [DOI LINK].

Fig. 3: Overview of the Upload page. Screenshot of the upload page where users can submit their own files to benchmark across multiple DNA codecs. The interface supports simulation of DNA degradation using MESA [21], enabling users to observe the impact on coding performance and error resilience with real-time benchmark results. the pipeline outputs the runtime, information density, coding cost, and the size of the intermediate file that contains the list of DNA sequences. DNA Data Storage To better compare DNA codecs and simulate DNA data degradation during storage, we employed two different strategies. First, for the creation of the

Springer Nature 2021 LATEX template 10

The Unified Evaluation App for DNA Data Storage Codecs

LEGEND START

Benchmark Calculation [BC] User (File) Upload [UU] Shared Process

Benchmark Calculation (BC) or User Upload (UU)?

UU

BC

[BC] Traverse Integrated Encoding Methods and Benchmark Files

[UU] Upload File (User Action)

YES

[BC] Benchmark Results

[UU] Select Encoding Methods (User Action) [BC] Encoding Method Already Benchmarked Against the File? Encode the File into a Binary NumPy Array

Encode Binary Data into DNA Sequences using the Selected Method [UU] Storage Length Simulation? NO

NO

[BC] Simulate Errors (deletions, insertions, and substitutions) in Varying Quantities

Decode DNA Sequences into Binary NumPy Array YES

[BC] Save Results [UU] Input Storage Length (User Action)

[UU] Send Sequences to the MESA API for Degradation Simulation

Compare with Original Array

Output Results: Runtime, Cost, etc.

Whisker or Bar

[BC] Specify the Chart Type

END

Fig. 4: Flowchart depicting the workflow of our web app. The chart outlines the encoding, storage, and decoding processes. Key user interactions and decision points are highlighted to enhance understanding of the methodology.

Springer Nature 2021 LATEX template The Unified Evaluation App for DNA Data Storage Codecs

11

benchmark data set and each run, our platform incorporated DNA degradation using error rates observed by the HEDGES [23] authors with a high mutagenesis kit, with a substitution rate of 0.0238, a deletion rate of 0.0082, and an insertion rate of 0.0039 as the baseline. Starting with a multiplier of 0 (no degradation) and gradually increasing it in steps of 0.5 up to 2.5, we tested the robustness of all codecs. The second strategy, used on the file upload page in addition to the first, relies on the DNA data storage simulator MESA [21]. In addition to selecting the codec, users can optionally specify the DNA storage length in months. This integer value, along with the list of DNA sequences outputted by each of the selected DNA codecs, is parsed to the MESA’s API [21]. MESA enables users to simulate DNA sequence degradation due to various storage properties such as environmental temperature, synthesis and sequencing methods, storage hosts, etc. Given the extensive range of simulation parameters available in MESA, we fixed almost all of them to their default values and provided only the storage length as a user-selectable option. This decision was made to streamline the user experience and reduce overhead, allowing users to focus on the primary factors that significantly impact DNA degradation. By limiting user input to essential parameters, we minimize the risk of errors that could arise from misconfiguration while ensuring that results remain consistent and reproducible across different analyses. Moreover, MESA is only used on the upload page because its detailed error simulation can be timeconsuming, whereas the first strategy is much faster and more suitable for benchmarking purposes. This approach not only enhances usability for those who may lack deep expertise in DNA storage conditions but also facilitates more straightforward comparisons of results across different experiments. The output of this process provides a list of DNA sequences for each selected codec. Importantly, both strategies introduces slight changes (degradation) to each sequence during simulation. Users may choose to omit the storage length value (in the case of the second strategy) or set the degradation multiplier to 0 (in the case of the first strategy) if they do not wish to simulate DNA data storage degradation (aging, physical mishandling, suboptimal storage conditions, enzymatic activity). This allows them to focus solely on evaluating the robustness of each DNA codec against file size and file content. Clustering and Multistrand Alignment Considerations In DNA data storage workflows, recovering the original data requires preprocessing of sequencing reads that typically involves clustering similar strands followed by multistrand alignment to generate a consensus sequence for decoding [19]. Clustering groups noisy or erroneous copies of oligos, usually using Levenshtein distance or similar metrics, which is computationally expensive with complexity at least quadratic in the number of reads [19, 24]. Multistrand alignment further refines sequences and corrects indels and substitutions through exact or heuristic algorithms whose computational costs vary from exponential to polynomial but remain substantial [19]. Because clustering

Springer Nature 2021 LATEX template 12

The Unified Evaluation App for DNA Data Storage Codecs

and alignment algorithms are closely intertwined with the error-correction and decoding logic, and differ substantially among codecs, our benchmarking framework does not isolate these steps as a separate, uniform preprocessing stage. Instead, each integrated codec internally implements or omits clustering and alignment according to its design principles, error model assumptions, and algorithmic strategies. This design choice preserves the authenticity of each codec’s performance, respects methodological heterogeneity, and aligns with best practices for evaluation standardization in DNA storage systems [19]. DNA Decoding Process The resulting list of DNA sequences from the DNA data storage step of the pipeline is used as input for decoding by the selected DNA codecs. Similar to the encoding process, the runtime of this procedure is significantly influenced by the size of the uploaded file and the number and type of selected codecs. After each list of DNA sequences is decoded back into a boolean NumPy array by the chosen DNA codecs, a comparison with the original array is performed to ensure data integrity. The outcome of this operation is a boolean value that indicates whether the selected DNA codecs successfully retrieved the decoded information as it was originally present in the file. Additionally, similar to the encoding process, the output includes the runtime, information density, decoding cost, and size of the intermediate file that contains a boolean NumPy array for each codec. The results collected during the encoding and decoding processes are presented in both tabular and visual formats. The tabular representation mirrors that of the benchmark page, first showcasing the summarized results and then providing detailed evaluation outcomes. Both tables can be easily downloaded in CSV format using the button located immediately below each table. Below the tables, four vertically stacked charts display different results for both the encoding and decoding processes: success rate, information density, runtime, and cost. The success rate chart illustrates the similarity between the original file and the file that has undergone encoding and decoding; the information density chart shows the difference in size between the encoded file and its original counterpart; while the cost chart visualizes the associated costs for each process. Lastly, the runtime chart shows the runtime of the encoding and decoding processes.

2.2 Coding Schemes (Codecs) The study integrates eight codecs: Repetition codec [6], Simple codec [5], 2D codec [25], XOR-based codec [26], DNA-Aeon codec [17], and three near-optimal rateless erasure codecs (NORECs), namely NOREC4DNA Raptor-based codec [27, 28], NOREC4DNA Online-based codec [27, 29], and NOREC4DNA LT-based codec [27, 30]. To enhance the readability, maintainability, and extensibility of these codecs and the overall DNA data storage process, we systematically restructured or adapted all original codecs. This

Springer Nature 2021 LATEX template The Unified Evaluation App for DNA Data Storage Codecs

13

was achieved by following two key principles. First, if a codec was implemented in an outdated or less commonly used programming language, such as Pascal [31] or Perl [32], we reimplemented it in Python [33]. Second, for codecs developed in modern programming languages that outperform Python in speed (e.g., C++ [34]), we created a wrapper script in Python to ensure uniform usage of all codecs across our app. Since each codec can employ different parameters that directly influence its performance and error-correction efficiency, we selected default parameter settings based on the respective studies. In cases where multiple parameters were possible, we carefully chose the most representative configuration as the default to ensure meaningful comparisons across all schemes, while also selecting parameters that provide each codec with a flexible configuration capable of encoding and decoding a wide variety of files, even if this flexibility may come at the cost of reduced efficiency. Below is a list of the specific parameters of each codec we used in our platform. Additional, less relevant parameters are available in the GitHub repository. • Repetition codec: No parameters. • Simple codec: Default parameters — a segment length of 12 bytes, with the maximum number of “oligos” set to 219 = 524, 288. • 2D codec: Default parameters — (a) In the inner encoder, the Galois field size (p) is 47, the message length is 39, the payload length is 33, the symbol size is 1, and the p-factor is 2. (b) In the outer encoder, the Galois field size (p) is 47, the message length is 713, the payload length is 594, the symbol size is 30, and the p-factor is 7. • XOR-based codec: Default parameters — 12 nucleotides per primer, a maximum of 6 nucleotides per byte, the minimum strand length follows the equation 27 + (10 × MAX NUCLEOTIDES PER BYTE), the maximum payload index gap and the minimum number of larger payloads are 10, and the minimum XOR size and step are 2 and 1, respectively. • NOREC4DNA Raptor-based codec: Overhead value is 0.9, headers are included, the error detection algorithm is Reed-Solomon, the upper bound threshold for dropping packets is 1, chunk size is 75, seed value is 2, and 7 repair symbols are used. • NOREC4DNA Online-based codec: Same parameters as Raptor-based codec, with additional parameters: epsilon calculated per file and the quality value at 8. • NOREC4DNA LT-based codec: Same as Raptor-based codec, but with the upper bound threshold for dropping packets changed to 1.0, 8 repair symbols, and the implicit mode set to enabled. • DNA-Aeon codec: Default parameters — chunk size is 50, the error detection algorithm is “crc” with the header length of 8, and the package redundancy value set to 0.4.

Springer Nature 2021 LATEX template 14

The Unified Evaluation App for DNA Data Storage Codecs

The restructuring effort, promoted by this study, aims to create a robust and flexible framework that simplifies the integration of new codecs. Central to this framework are two key wrapper functions: encode(file path, bool array) and decode(file path, list of dna strings), implemented for each codec. The encode function accepts either the path to a file or a corresponding NumPy boolean array that represents the binary data of the file. It processes this data using the selected codec and returns a list of DNA strings that represent the encoded sequence. Each DNA string in the list corresponds to a word where each character is an element of the alphabet {A, C, G, T }. In contrast, the decode function takes either the path to a file or a list of DNA strings and applies the chosen decoding algorithm to reconstruct the original binary data, returning it as a NumPy boolean array. Each encoding or decoding operation is executed as a separate process to ensure platform robustness and prevent crashes, as each codec handles inputs differently, particularly in cases where the data may contain degradation or noise. These wrapper functions provide a flexible foundation for the system, enabling seamless integration of future codecs. By standardizing input and output formats through these functions, researchers can easily integrate new algorithms or adapt existing ones without extensive modifications to the overall codebase. This modular approach facilitates rapid development and testing of new methods while ensuring that the framework remains adaptable to the evolving landscape of DNA data storage technologies.

2.3 Evaluation System To evaluate all integrated codecs uniformly, we used a carefully curated baseline data set alongside multiple evaluation criteria. This baseline data set serves as a reference point, providing a consistent foundation for comparing various codecs. It was meticulously selected to accurately represent the types of data the codecs will encounter in practical applications. Building on the criteria proposed by the DNA Data Storage Alliance for DNA codecs [19], we structured our benchmarking suite around five orthogonal dimensions: (1) scalability, (2) computational performance, (3) error-detection and correction efficacy, (4) compatibility and specification transparency, and (5) biochemical constraint adherence. These axes map one-to-one to the DNA Data Storage Alliance recommendations of sub-quadratic algorithmic complexity, throughput-per-watt targets, explicit success probabilities for substitutions/insertions/deletions, Rosetta-Stone style codec format descriptions, and GC/homopolymer constraints, respectively. In the following sections, we will describe both the baseline data set and the evaluation criteria in detail. We will outline the selection process for the baseline data set, highlighting the importance of using diverse data types to test the robustness of each codec. Additionally, we will discuss how each evaluation criterion was determined and the rationale behind its inclusion in our assessment framework.

Springer Nature 2021 LATEX template 15

The Unified Evaluation App for DNA Data Storage Codecs

2.3.1 Baseline (Benchmark) Data Set To thoroughly evaluate the performance of the DNA data storage codecs, this study utilized a diverse set of benchmark test files. The set consists of two text, three image, two audio, one video, and one document file, carefully selected to represent a broad spectrum of file types and contents commonly encountered in digital storage. This diverse data set allowed us to examine each codec on a variety of data modalities, revealing whether the method’s performance relied on the underlying semantics or structure of the data, or was more influenced by the noise characteristics of the encoding (synthesis) and decoding (sequencing) processes. Each file in the data set was encoded, underwent error simulation, and decoded five times for each codec. This comprehensive assessment ensured a thorough evaluation of the capabilities of the codecs. Overall sizes of the files contained within the benchmark data set can be seen in Figure 5. In addition, the byte value distribution of each file is presented in histograms as part of Figure 6. We give a brief description of each benchmark file below.

Size (bytes)

bbc_peregrine_falcon.wav

Entropy (shannons)

751800

0.652

Zero_File.pdf 7523

7.40

Around_the_World_in_Eighty_Days.txt

403521

4.71

The_Picture_of_Dorian_Gray.txt

465592

4.62

bbc_clock.mp3

121753

0.578

0.527

1511673 70296 185727 0

0.996

4.21

egypt.mp4

nature.gif

0.926 0.589

7.97

all_black.png 2778

Vue_de_la_Buena_Vista.jpg

Normalized entropy

5.21

0.5M

8.00

1.000

7.97

0.996

7.64 1M

1.5M

0

2

4

6

0.955 8

0

0.5

1

Fig. 5: Overview of benchmark data set characteristics for various file types. Left: Distribution of file sizes in bytes across all tested formats. Middle: Measured Shannon entropy values for each file, indicating the degree of redundancy and randomness present in the data. Right: Normalized entropy values for each file, illustrating the uniformity of byte distributions. First, the two text files that are part of the data set are books “Around the world in eighty days” by Jules Verne [35] and “The Picture of Dorian Gray” by Oscar Wilde [36]. Inclusion of the two natural text files allowed for the efficient assessment of the codecs in handling simple, low entropy, ASCII-based data and their scalability when processing larger volumes of textual information. Second, the three image files in standard formats were included to evaluate how well the codecs manage binary data with higher entropy, typical of visual media. Besides two images with higher entropy in JPEG and GIF formats, caused by their high compression rates, we created a lower entropy PNG image that contained only black pixels. The images were selected to represent

Springer Nature 2021 LATEX template 16

The Unified Evaluation App for DNA Data Storage Codecs

different resolutions, color depths, and entropies providing insights into how effectively the methods compressed and preserved visual information. Third, the two audio files in MP3 [37] and WAV [38] formats are incorporated into the benchmark tests to further challenge the codecs. These files tested the capability of the system to handle continuous, non-textual data. Due to different compression algorithms and formats the two files have completely different byte-value profiles — one has a uniform and flat while other has highly skewed histogram of byte values, as seen in Figure 6. The audio files presented an intricate test for maintaining fidelity in waveform-based data, which is more intricate than text or image data. Fourth, a video file in MP4 format was also included to assess the codecs’ performance with multimedia data that combines both audio and visual components. This file represents a more intricate data type, requiring the codecs to effectively manage large volumes of information while maintaining synchronization between audio and video streams. The choice of an MP4 file is particularly relevant, as it is the biggest in size, contains the highest Shannon entropy of 8 bits per byte, and the highest normalized entropy of 1, reflecting a uniform byte distribution across all possible values. The uniform byte distribution, as illustrated in Figure 6, is particularly significant because it can cause the presence of homopolymers following the encoding process, which can present challenges for various DNA codecs. Finally, we incorporated a PDF document formatted in A4 size that contained only repeated instances of the character 0 (zero). This specific choice was made to evaluate how well the codecs handle uniform data, low in size but high in entropy, as evidenced by Figure 5 and Figure 6.

2.3.2 Evaluation Criteria Consistent with both DNA Data Storage Alliance codec guidance [19] and recent comparative studies of fountain, HEDGES [23], and arithmetic-based methods [17, 18], we evaluate each codec along four quantitative metrics — information density, success rate (data integrity), runtime, and cost. First, Information Density refers to the amount of meaningful or useful information contained within a given unit of data or space. In data coding and compression contexts, information density is typically calculated as the ratio between the size of the original data and the size of the encoded or compressed data. For example, if an input file is 300 bits and it is encoded into a DNA strand of 100 nucleotides (bases), the information density is 3 (300 bits / 100 bases). This ratio indicates how efficiently the codec stores information: a higher value means more data is packed into each unit of the storage medium. Moreover, some codecs report information density based solely on the payload, which can distort the true information density of the coded sequence. However, we argue that this metric should encompass all sources of redundancy, including both inner code overhead and the supplementary sequences introduced by outer error correction codes. Therefore, our approach computes information

Springer Nature 2021 LATEX template The Unified Evaluation App for DNA Data Storage Codecs

ycneuqerF dezilamroN

0.15

filename=Around_the_World_in_Eighty_Days.txt

filename=The_Picture_of_Dorian_Gray.txt 0.15

0.1

filename=Vue_de_la_Buena_Vista.jpg 0.008 0.006

0.1

0.004

0.05

0.05

0

0 filename=Zero_File.pdf

ycneuqerF dezilamroN

0.06

0.002 0 filename=all_black.png

0.3

filename=bbc_clock.mp3 0.01

0.25

0.05

0.008

0.04

0.2

0.03

0.15

0.006

0.02

0.1

0.004

0.01

0.05

0.002

0

0

0

filename=bbc_peregrine_falcon.wav

ycneuqerF dezilamroN

0.25

filename=egypt.mp4 0.006 0.005

0.02

0.15

0.004

0.015

0.003

0.1

0.01

0.002

0.05

0.005

0.001 0

32

64

96 128 160 192 224 256

Byte Value (0-255)

0

filename=nature.gif

0.025

0.2

0

17

0

32

64

96 128 160 192 224 256

Byte Value (0-255)

0

0

32

64

96 128 160 192 224 256

Byte Value (0-255)

Fig. 6: Histograms displaying the byte value distributions for representative files from the benchmark data set. Each histogram visualizes the frequency of byte values across diverse file types (PDF, TXT, WAV, MP3, MP4, PNG, GIF, JPG), revealing structural patterns, regions of redundancy, or randomness that influence algorithmic performance. density using the entire binary or DNA sequence, explicitly including headers and any redundant data. Second, Success Rate is evaluated by comparing the original binary data with the information retrieved after decoding the DNA sequences. This metric reflects the robustness of each codec against DNA degradation. When assessed across different base error ratios, it illustrates the codec’s resilience to varying levels of degradation. Third, Runtime refers to the time taken in seconds to convert digital files into DNA sequences and subsequently back into their original binary format. The encoding time reflects the algorithm’s efficiency in transforming binary data into DNA sequences, while the decoding time evaluates how quickly and accurately the original data can be reconstructed from the encoded DNA. Fourth and last, Cost pertains to the estimated expenses associated with chemically or enzymatically synthesizing and sequencing DNA. This measure is influenced by factors such as the length of the oligonucleotides and the average costs associated with sequencing and synthesis processes. Currently, storage costs are not included in this estimate. This mapping ensures that our app produces indicators directly actionable for system architects. For instance, the information density metric can be combined with DNA Data Storage Alliance’s scalability clause to expose codecs

Springer Nature 2021 LATEX template 18

The Unified Evaluation App for DNA Data Storage Codecs

whose density gains are achieved only at super-quadratic computational cost. Collectively, these evaluation criteria form a comprehensive framework that facilitates a thorough analysis of the strengths and weaknesses of the DNA data storage codecs assessed in this study.

3 Results We organize our findings into two distinct sections. The first section details the benchmark results of various DNA codecs using default parameters, highlighting their performance metrics and comparative advantages. The second section introduces our app, outlining its key features, functionalities, and inherent limitations.

3.1 Benchmark The benchmark results provided several critical insights into the performance of various DNA codecs integrated within the app. We emphasize that all performance measures examined in this study are strongly influenced by the full set of relevant codec parameters. Below we systematically present the key findings derived from our analysis. This includes a detailed examination of specific performance metrics, such as performance variability, base error, computational efficiency, and cost efficiency. Additionally, we conduct a comparative analysis of the codecs, highlighting their respective advantages and disadvantages based on empirical data. Performance Variability The benchmark results demonstrated significant variability in information density across the integrated codecs and benchmark data sets. In the absence of error, defined herein as an error multiplier of 0.0, all eight codecs successfully completed the benchmark and produced measurable information-density values, allowing direct comparison across the entire set of codecs as seen in Figure 7. NOREC4DNA LT-based and Raptor-based codecs produced consistently high information densities across most files, reaching approximately 0.9 b/nt. In the no-error setting, NOREC4DNA Online-based codec demonstrated a similar pattern, although with slightly lower values. DNAAeon produced a lower but stable information density, while Repetition and XOR-based codecs occupied the lower end of the information density range, reflecting the redundancy introduced by these codecs. The most significant file-dependent variation was observed for Simple and 2D codecs. In a subset of files, as shown in Figure 8, these codecs exhibited significantly higher information-density values compared to the remaining codecs. Conversely, in other files, their information densities were closer to the overall range observed across the benchmark set. This finding suggests that the information density was not solely determined by the choice of codec; rather, it was influenced by the interplay between the coding scheme and the input

Springer Nature 2021 LATEX template 19

The Unified Evaluation App for DNA Data Storage Codecs

file’s structure. This is consistent with the benchmark design, which intentionally included files with different formats, sizes, byte-value distributions, and entropy profiles. Across a range of increasing error multipliers, the information density values for codecs that were successfully completed remained largely comparable to those in the no-error condition. This is expected since the information density is computed from the original file size and the length of the encoded DNA sequences, whereas the error multiplier affects the degradation and decoding stages. Therefore, changes observed in error-multiplier charts should be interpreted primarily in relation to the presence or absence of completed codec runs. Missing entries indicate codec–file–error combinations that did not complete successfully under the default configuration we used in this study. These results show that information density alone is insufficient for evaluating the performance of codecs. NOREC4DNA LT-based and Raptor-based codecs demonstrated the most consistent information density profile across the benchmark set, while Simple and 2D codecs achieved higher information densities in selected cases but with stronger file dependence. DNA-Aeon showed a lower but more stable information density profile. These patterns underscore the need for codec comparison tools that evaluate storage efficiency alongside robustness and runtime, rather than ranking codecs solely by information density.

) t n / b( y ti s n e D _ n oi t a m r o f nI

1 0.8 0.6 0.4

DN

A _A

eon

LTC

od

e

On

lin e

Co

Rap de

to r

Co

Rep de

e t it

io n

S im Co

de

p le

Co

The de

2D

Co

XO de

RB

ase

dC

od

e

Method

Fig. 7: Overall information density of the evaluated codecs in the no-error condition. The plot summarizes the information density achieved by each codec before degradation is introduced, allowing direct comparison across all eight integrated codecs.

Springer Nature 2021 LATEX template 20

The Unified Evaluation App for DNA Data Storage Codecs

Filename=The_Picture_of_Dorian_Gray

Filename=Vue_de_la_Buena_Vista

Filename=Zero_File

12

) t n / b( y ti s n e D _ n oi t a m r o f nI

10 8 6 3.61

4 2

0.92

0.91 LTC

od

e

Rap

1.12

0.34

0 to r

Co

Rep de

e t it

0.91

0.92

0.34

0.60

1.12

0.55

0.89

0.90

0.34

0.60

0.72

0.54

S im The DN LTC Rap R Si The DN LTC Rap R Si The DN A _A A _A A _A od od 2 2D 2D p t o r e p e t it m p le t o r e p e t it m p le e e Co Co Co Co Co Co eon eon eon io n le C o d D C o d io n io n de de de de de de Co Co Co e e de de de

Filename=all_black

Filename=bbc_clock

Filename=egypt 11.73

12

) t n / b( y ti s n e D _ n oi t a m r o f nI

10 8 6 4 2

0.87

0.86 LTC

od

e

Rap

0.60

0.34

0 to r

Co

Rep de

e t it

0.27

0.53

0.91

0.92

0.34

0.94

1.06

0.55

0.91

0.92

0.34

1.13

0.55

S im The DN LTC Rap R Si The DN LTC Rap R Si The DN A _A A _A A _A od od 2 2D 2D p t o r e p e t it m p le t o r e p e t it m p le e e Co Co Co Co Co Co eon eon eon io n le C o d D C o d io n io n de de de de de de Co Co Co e e de de de

Filename=nature

Filename=Around_the_World_in_Eighty_Days

Filename=bbc_peregrine_falcon

12

) t n / b( y ti s n e D _ n oi t a m r o f nI

10 8 6 4 2

1.44

0.92

0.91

0.34

0 LTC

od

e

Rap

to r

Co

Rep de

e t it

1.11

0.91

0.92

0.55

0.91

0.92

0.55

S im The DN LTC Rap R Si The DN LTC Rap R Si The DN A _A A _A A _A od od 2 2D 2D p t o r e p e t it m p le t o r e p e t it m p le e e Co Co Co Co Co Co eon eon eon io n le C o d D C o d io n io n de de de de de de Co Co Co e e de de de

Method

Method

Method

Fig. 8: File-level information density across codecs at error multiplier 0.5. The plots show how information density varies across benchmark files and codecs. Missing codec entries indicate runs that did not complete successfully under the default configuration used in this study. Base Error and Data Integrity The results of the success-rate analysis showed that the robustness of the codecs differed significantly across the integrated methods. Unlike information density, which remained relatively stable across completed runs, the success rate showed a high degree of sensitivity to the introduction of base

Springer Nature 2021 LATEX template The Unified Evaluation App for DNA Data Storage Codecs

21

errors. As shown in Figure 9, across the comprehensive benchmark, DNA-Aeon achieved the highest overall success rate at 55.97%, followed by NOREC4DNA Raptor-based codec at 27.94% and NOREC4DNA LT-based codec at 19.50%. NOREC4DNA Online-based and XOR-based codecs each achieved 7.69%, Repetition codec achieved 5.33%, 2D codec achieved 2.67%, and Simple codec did not successfully recover any tested file under the evaluated conditions. This approach directly aligns with the success-probability and error-tolerance criteria recommended for DNA codec evaluation [19]. The base-error analysis further showed that most codecs experienced a sharp decline in success rate once errors were introduced as illustrated in Figure 10. In the absence of errors, several codecs successfully recovered the original data. However, as the base-error ratio increased, successful recovery became increasingly concentrated in a smaller subset of methods. DNA-Aeon showed the most consistent recovery profile across the evaluated base-error range, while NOREC4DNA Raptor-based LT-based codecs demonstrated successful recovery only under more limited conditions. The remaining codecs showed low or inconsistent recovery once degradation was introduced. These results demonstrate that high information density does not necessarily imply high data integrity. Some codecs that achieved competitive density values showed limited recovery under degraded conditions, while DNA-Aeon achieved the strongest success-rate profile despite its lower information density. Therefore, the benchmark indicates a clear trade-off between storage efficiency and robustness, reinforcing the need to evaluate DNA storage codecs using both density and data-recovery metrics. Computational Efficiency The runtime results showed substantial differences in computational performance across codecs, files, encoding and decoding stages. Runtime was measured separately for encoding and decoding, and the results are presented on a logarithmic scale because execution times varied by several orders of magnitude. As illustrated in Figure 11a, in the absence of errors, smaller and lower-intricacy files generally required shorter execution times. In contrast, larger or more intricate data types, such as the egypt video and audio files, exhibited significantly extended runtime across several codecs. The separation between encoding and decoding further showed that computational cost was not evenly distributed across the two stages. Some codecs completed encoding relatively quickly but required substantially longer decoding time, while others showed the opposite pattern. NOREC4DNA LT-based, Online-based, and Raptor-based codecs generally required longer runtimes for larger files, reflecting the additional computational overhead associated with rateless erasure coding and reconstruction. DNA-Aeon also showed increased decoding time under degraded conditions, particularly for larger files and files with more complex byte-value distributions. In contrast, Repetition, Simple, and XOR-based codecs frequently exhibited superior processing speeds upon

Springer Nature 2021 LATEX template 22

The Unified Evaluation App for DNA Data Storage Codecs

100

) %( et a R s s e c c u S

80 60

55.97%

40 20 0

27.94%

19.50% 7.69%

DN

A _A

eon

LTC

od

e

On

lin e

5.33%

Co

Rap de

to r

Co

Rep de

e t it

io n

S im Co

de

p le

7.69%

2.67%

0.00%

Co

The de

2D

Co

XO de

RB

ase

dC

od

e

Method

Fig. 9: Overall success rate across evaluated codecs. Success rate was calculated by comparing the original binary data with the decoded output after degradation and decoding. DNA-Aeon achieved the highest overall success rate, followed by NOREC4DNA Raptor-based and LT-based codecs. completion. However, their lower success-rate profile indicates that runtime efficiency alone does not imply practical robustness. As the error multiplier increased, the runtime results became more uneven and more sparse, as illustrated in Figure 11b. This pattern reflects both the additional computational burden introduced by degraded sequences and the fact that some codec–file–error combinations did not complete under the default configuration used in this study. Consequently, runtime must be interpreted jointly with completion and success-rate results. A codec that completes quickly but fails to recover the original data offers limited practical value, while a codec with higher runtime may be preferable when it provides stronger data integrity under error. Overall, the runtime analysis shows that computational efficiency is a major source of variation among DNA storage codecs. The benchmark therefore highlights a three-way trade-off between information density, data integrity, and runtime. NOREC4DNA Raptor-based and LT-based codecs provided comparatively strong information density but required greater computational time for larger files. DNA-Aeon achieved the strongest success-rate profile but could become computationally expensive during decoding under degraded conditions. Simpler codecs were faster in completed runs, but this advantage was offset by weaker recovery performance.

Springer Nature 2021 LATEX template The Unified Evaluation App for DNA Data Storage Codecs

23

Cost Efficiency The cost analysis showed that estimated storage cost varied substantially across codecs, benchmark files, encoding and decoding stages. As with runtime, cost values were presented on a logarithmic scale because the estimates spanned several orders of magnitude. As illustrated in Figure 12a, across the benchmark set, larger files exhibited substantially higher estimated costs; for example, the egypt video file and the larger audio and text files were the most expensive. In contrast, smaller or highly uniform files, such as all black and Zero File, generally required lower estimated costs. Across codecs, encoding costs were consistently higher than decoding costs. This pattern reflects the fact that the encoded DNA sequence, including payload, headers, and redundancy, directly affects the amount of DNA that must be synthesized. Codecs that introduced more redundancy or generated longer encoded sequences therefore produced higher cost estimates, while more compact sequences generally reduced the cost burden. However, the relationship between cost and information density was not uniform across all files, indicating that cost was also influenced by file size, codec-specific overhead, and the number and length of generated DNA sequences. The cost results also show that low cost cannot be interpreted independently of data recovery. While some codecs produced comparatively lower costs for specific files, these advantages were not always accompanied by high success rates under degraded conditions. Conversely, DNA-Aeon showed stronger recovery performance but did not always produce the lowest cost estimate. Therefore, cost efficiency must be interpreted alongside information density and success rate, rather than as a standalone measure of codec performance. As the error multiplier increased, the cost results became more sparse, as illustrated in Figure 12b, reflecting the same pattern observed in the information-density and runtime results. Missing entries indicate codec–file– error combinations that did not complete successfully under the default configuration used in this study. Overall, the cost analysis highlights a practical trade-off in DNA data storage: codecs that improve robustness or introduce additional redundancy may increase synthesis-related cost, while lower-cost configurations may provide weaker protection against degradation. Impact of Codec Choices The combination of these results shows that codec choice has a direct effect on storage efficiency, data integrity, computational performance, and estimated cost. No single codec dominated across all evaluation criteria. NOREC4DNA LT-based and NOREC4DNA Raptor-based codecs achieved consistently high information densities, but their performance must be interpreted alongside their runtime and recovery behaviour. DNA-Aeon achieved the strongest overall success-rate profile, despite producing lower information density than the rateless erasure codes. In contrast, less complex or more compact schemes were sometimes computationally efficient or cost-efficient for specific files. However, these methods showed weaker recovery under degraded conditions.

Springer Nature 2021 LATEX template 24

The Unified Evaluation App for DNA Data Storage Codecs

It is also important to note that some codecs appear in figures summarizing multiple metrics (e.g., encoding/decoding runtime or cost) despite not achieving successful end-to-end reconstruction. For instance, the Simple codec is included because it was able to perform encoding and decoding under certain error regimes and for certain files. However, the decoded outputs never matched the original input files, as evidenced by Figure 9. Consequently, although measurable results are reported, this codec did not achieve a successful encode–decode cycle, and its results should therefore be interpreted with caution. Similar behavior was also observed for other codecs, indicating that this limitation is not unique to a single method. These findings suggest that codec selection is application-dependent. For use cases where storage density is the primary constraint, NOREC4DNA LT-based and Raptor-based codecs may be preferable. For scenarios where recovery under degradation is more important, DNA-Aeon provides a more robust profile under the evaluated configuration. For rapid testing or lowcomplexity files, simpler codecs may still be useful, provided that their lower error tolerance is acceptable. Therefore, the app’s comparative framework is valuable because it allows users to evaluate codecs across multiple dimensions rather than selecting a method based on a single metric. Overall, the benchmark demonstrates that codec evaluation requires a multi-criteria approach in which information density, success rate, runtime, and cost are interpreted collectively.

3.2 Web Application First, the application’s modular nature, abstracted implementations, and use of modern programming languages enable easier adoption by experts in the field. We also utilized Docker, which automates app deployment within portable containers. By doing so, our work encourages reproducibility and collaboration by ensuring consistent behavior across different environments. Second, even though the app’s architecture is designed and finished, improving the app’s utility remains a work in progress. This is due to the rapidly evolving field of DNA data storage, which leads to constant advancements in codecs. Our plan is to continuously update the app with new DNA codecs while also ensuring smooth collaborations with other scientists. To achieve this, we have open-sourced the web app and its data, written extensive documentation, and developed the app to adhere to commonly used Python conventions and standards (e.g., Python Enhancement Proposal (PEP) 8, PEP 20, PEP 257, PEP 287 [39], etc.). Third, we have clearly outlined the process for integrating currently unavailable DNA codecs into our app. By leveraging a version control system (i.e., git) and a GitHub repository, along with a detailed step-by-step guide, we aim to encourage other researchers and practitioners to actively contribute to and enhance the app. Fourth and last, we developed a Python script that automates the recalculation of the benchmark data set whenever a new codec is successfully

Springer Nature 2021 LATEX template The Unified Evaluation App for DNA Data Storage Codecs

25

integrated into the web app. This script features a user-friendly dashboard that allows users to easily access both previously calculated benchmark results and the results generated by the newly implemented codecs. Users have the option to update the pre-calculated benchmark results with the new data, and seamlessly push these changes to the app’s main repository. To further support extensive benchmarking efforts, we also provide an additional script designed to run the pre-calculation on an High-Performance Computing (HPC) Slurm environment [40], which is particularly beneficial for benchmarks that may require substantial computing time or resources. This automation not only enhances efficiency but also streamlines the integration of new codecs into the existing framework. By simplifying this evaluation process, we aim to encourage frequent updates and enhancements to the app, keeping it at the forefront of DNA data storage technology.

4 Discussion The interdisciplinary nature of DNA data storage requires rigorous evaluation frameworks that bridge the gap between theoretical foundations and practical implementation challenges. Our unified evaluation app fills a critical gap in the field by providing standardized benchmarking capabilities while maintaining the flexibility essential for ongoing methodological innovation. Furthermore, our platform fosters collaboration among researchers and serves as a guidance or reference app for various storage needs. The study optimizes a series of established DNA data storage codecs at the software engineering level, offering a universal application for DNA data storage. The app ensures compatibility with existing technologies and stability for long-term storage by establishing an evaluation and analysis system based on widely accepted key parameters in the field. The study uses different file types and formats to evaluate classical DNA data storage codecs, enabling systematic analysis and comparison. However, such evaluations are complicated by the fact that many DNA codecs have different configuration options that can significantly impact performance. This variability poses a challenge for benchmarking because direct comparisons may be confounded by differences in parameter settings. To address this issue, the study carefully selects optimal or default configurations for each codec, prioritizing settings that demonstrate consistent and robust performance across a wide range of benchmark files. This strategy ensures a fair, standardized evaluation framework that remains practically relevant, facilitating meaningful comparisons while preserving the diversity inherent to each codec. Therefore, future additions to the benchmark could incorporate parameter sweeps, adaptive optimization strategies, or contextdependent tuning (e.g., error profiles, synthesis, and sequencing constraints). This would provide a more comprehensive evaluation of each codec’s potential performance under diverse scenarios.

Springer Nature 2021 LATEX template 26

The Unified Evaluation App for DNA Data Storage Codecs

It is important to emphasize that the benchmark results presented in this study, while encompassing multiple codecs and diverse file types, should not be interpreted as definitive indicators of the absolute quality or performance of the individual codecs. Due to the large configuration space and file-dependent behavior of DNA storage methods, we did not exhaustively optimize parameters for every codec–file combination. Instead, we employed a set of well-justified, “sane” default configurations for each codec, as described earlier, to ensure consistency and practical feasibility. Consequently, the primary contribution of this work lies not in ranking codecs, but in demonstrating the capabilities of the proposed app as a unifying tool and platform that enables standardized, reproducible, and extensible evaluation across heterogeneous methods, thereby supporting more transparent and comparable benchmarking practices in the field. This study aimed to solve the challenge of unifying and standardizing methods and their evaluation without stifling innovation. Although standardization is crucial for consistently evaluating different codecs, maintaining flexibility is equally important for exploring new solutions and identifying emerging problems. Maintaining this balance is critical to ensuring the field continues to evolve and adapt to new discoveries. The structured design of the wrapper functions used in this study supports scalability and ensures that new methods can be incorporated with minimal disruption. As the field of DNA data storage advances, this flexible framework will allow the research community to efficiently explore and implement innovative encoding and decoding strategies, thereby advancing the capabilities of DNA as a data storage medium. Integrating the DNA Data Storage Alliance’s codec taxonomy enabled us to reposition our platform as not only a performance web application, but also a compliance verification framework. By scoring each method against codified community criteria, we create a living baseline that can adapt as consensus standards (e.g., ISO/IEC JTC1 SC31) evolve. This mitigates the risk of future format obsolescence, as discussed in Landsman and Strauss [41]. The benchmark results further demonstrate why such a standardized framework is necessary. Codec performance was not governed by a single metric: methods that achieved high information density did not always achieve high recovery success, low runtime, or low estimated cost. NOREC4DNA LTbased and Raptor-based codecs showed consistently high information-density profiles, while DNA-Aeon achieved the strongest overall success-rate performance under the evaluated degradation settings. These differences indicate that codec choice is inherently application-dependent and should be guided by the intended balance between storage efficiency, data integrity, computational feasibility, and cost. Thus, the main contribution of the app is not to identify a universally best codec, but to make these trade-offs visible under a common evaluation framework. However, while comprehensive within its scope, our evaluation framework faces several inherent limitations that warrant consideration. First, although

Springer Nature 2021 LATEX template The Unified Evaluation App for DNA Data Storage Codecs

27

reliance on MESA’s default parameters for degradation simulation is justified by the need for methodological consistency, such an approach may not accurately reflect the full diversity of storage conditions encountered across various deployment scenarios. In future work, alternative solutions, such as DNA Storalator [42], will be explored, with the goal of providing the end user with additional DNA storage simulation options and yielding a more comprehensive picture of the DNA storage landscape. Second, the comparison relies on selected default or representative codec configurations. This was necessary because the parameter space of several codecs is too large for exhaustive exploration, but it also means that some failed or missing runs may have succeeded under alternative settings. In particular, some codec–file–error combinations did not complete under the evaluated configuration, and several implementations returned limited diagnostic information when failures occurred. Future work should therefore include more systematic parameter sweeps and more explicit failure classification, distinguishing between decoding failure, timeout, unsupported error types, file-size limitations, and implementation-level errors. Third, the binary data representation through NumPy arrays, chosen for compatibility across codecs, introduces computational overhead that obscures the true performance characteristics of optimized implementations. Fourth and finally, although we attempted to implement as many DNA codecs as possible, we were unable to include several others that remain highly relevant to the field. Some noteworthy examples are Gungnir [43] and DNAformer [18], among others. The former uses proof-of-work idea to address substitution, insertion, and deletion errors in a DNA sequence, while the latter combines deep neural networks, specifically a transformer model with error-correcting codes, to significantly enhance information retrieval from DNA-stored data. Our ongoing efforts aim to integrate more such methods into our framework to further strengthen the comprehensiveness and utility of our work.

5 Conclusion The development of the proposed unified evaluation app is a foundational step toward standardizing the assessment of DNA data storage while preserving the methodological flexibility essential for continued innovation. Our platform implements the five-dimensional evaluation framework proposed by the DNA Data Storage Alliance, providing the research community with a reproducible benchmarking system that directly addresses real-world deployment considerations. Our analysis of eight codecs using default parameters and across various data types shows that no single algorithm can optimize all performance dimensions simultaneously. This finding underscores the importance of selecting application-specific methods and highlights the value of comparative evaluation frameworks. The observed performance patterns, particularly regarding

Springer Nature 2021 LATEX template 28

The Unified Evaluation App for DNA Data Storage Codecs

scalability limitations and cost-efficiency trade-offs, provide actionable insights for algorithm developers and system architects alike. The app’s modular architecture and standardized wrapper functions facilitate the integration of emerging codecs while maintaining evaluation consistency. This design philosophy enables the platform to adapt to technological advances while preserving historical benchmarking data for longitudinal analysis. Aligning with the DNA Data Storage Alliance criteria positions our framework to support the field’s transition from experimental demonstrations to commercial implementations. Looking ahead, successfully deploying DNA storage technologies critically depends on standardized evaluation methodologies that can guide research priorities and industrial adoption decisions. Our contribution provides the community with a foundation for evidence-based method selection and a platform for the collaborative advancement of the field. As synthesis costs decline and sequencing throughput increases, standardized evaluation frameworks will be essential for realizing DNA storage’s transformative potential. The open-source nature of our implementation encourages community participation and ensures that the platform evolves alongside technological developments. By establishing this evaluation infrastructure, we enable the systematic assessment necessary to transform DNA data storage from a promising laboratory technique into a reliable, standardized technology for long-term information preservation. Future research should expand the evaluation framework to incorporate emerging synthesis technologies and novel error-correction schemes, while maintaining backward compatibility with existing benchmarking data. Additionally, integrating machine learning techniques optimized for graphics processing unit (GPU)-accelerated encoding and decoding would enhance the platform’s performance and adaptability. This would enable more efficient and accurate analysis of increasingly complex DNA data storage methods. Code availability. Source code, help, and documentation are available at https://github.com/AAnzel/UNACORM/. The repository also contains the source code necessary to reproduce all examples shown in this work. It is licensed under the GNU General Public License, Version 3.0, and can be manipulated, improved, and extended freely by any user. Availability of data and materials. The data used in this study can be found at https://github.com/AAnzel/UNACORM/tree/master/Data/ Original. Authors’ contributions. C.A. and A.A. wrote the manuscript, designed the app, and evaluated the results. C.A., A.A., and I.T.H. conducted the experiments and validated the findings. I.T.H. and A.A. developed the main app. C.A., A.A., and G.H. designed the user experience. L.W., L.S., D.S., K.E., S.B., F.K., R.S., M.W., and M.S. reimplemented or adapted individual codecs that are part of the main app. C.A. and A.A. supervised the reimplementation. D.H. and G.H. supervised the project. C.A, A.A, M.W, M.S., K.E., R.S., L.W.,

Springer Nature 2021 LATEX template The Unified Evaluation App for DNA Data Storage Codecs

29

D.H., B.F., and G.H. proofread, and revised the manuscript. All authors read and approved the final manuscript. Funding. This work was partially financially supported by the LOEWE program of the State of Hesse (Germany) in the MOSLA research cluster. Conflict of interest. The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

References [1] HUAWEI TECHNOLOGIES CO., LTD. Data storage 2030: Building a fully connected, intelligent world. Tech. Rep., HUAWEI TECHNOLOGIES CO., LTD., Huawei Industrial Base, Bantian Longgang, Shenzhen 518129, P. R. China (2024). URL https://www-file.huawei.com/-/media/ corp2020/pdf/giv/2024/data storage whitepaper 2030 en.pdf. Accessed August 2025. [2] Anžel, A., Heider, D. & Hattab, G. The visual story of data storage: From storage properties to user interfaces. Computational and Structural Biotechnology Journal 19, 4904–4918 (2021). URL https: //www.sciencedirect.com/science/article/pii/S2001037021003627. https: //doi.org/https://doi.org/10.1016/j.csbj.2021.08.031 . [3] Bar-Lev, D., Sabary, O. & Yaakobi, E. The zettabyte era is in our dna. Nature Computational Science 4 (11), 813–817 (2024). URL https://doi.org/10.1038/s43588-024-00717-1. https://doi.org/10.1038/ s43588-024-00717-1 . [4] Hoose, A., Vellacott, R., Storch, M., Freemont, P. S. & Ryadnov, M. G. Dna synthesis technologies to close the gene writing gap. Nature Reviews Chemistry 7 (3), 144–161 (2023). URL https://doi.org/10.1038/ s41570-022-00456-9. https://doi.org/10.1038/s41570-022-00456-9 . [5] Church, G. M., Gao, Y. & Kosuri, S. Next-generation digital information storage in dna. Science 337 (6102), 1628–1628 (2012). URL https://www.science.org/doi/abs/10. 1126/science.1226355. https://doi.org/10.1126/science.1226355, https://www.science.org/doi/pdf/10.1126/science.1226355 . [6] Goldman, N. et al. Towards practical, high-capacity, low-maintenance information storage in synthesized dna. Nature 494 (7435), 77–80 (2013). URL https://doi.org/10.1038/nature11875. https://doi.org/10. 1038/nature11875 . [7] Takahashi, C. N., Nguyen, B. H., Strauss, K. & Ceze, L. Demonstration of end-to-end automation of dna data storage. Scientific Reports 9 (1),

Springer Nature 2021 LATEX template 30

The Unified Evaluation App for DNA Data Storage Codecs 4998 (2019). URL https://doi.org/10.1038/s41598-019-41228-8. https: //doi.org/10.1038/s41598-019-41228-8 .

[8] Löchel, H. F., Welzel, M., Hattab, G., Hauschild, A.-C. & Heider, D. Fractal construction of constrained code words for dna storage systems. Nucleic Acids Research 50 (5), e30–e30 (2021). URL https://doi.org/10.1093/nar/gkab1209. https://doi. org/10.1093/nar/gkab1209, https://academic.oup.com/nar/articlepdf/50/5/e30/42917805/gkab1209.pdf . [9] Yin, Q., Zheng, Y., Wang, B. & Zhang, Q. Design of constraint coding sets for archive dna storage. IEEE/ACM Transactions on Computational Biology and Bioinformatics 19 (6), 3384–3394 (2022). https://doi.org/ 10.1109/TCBB.2021.3127271 . [10] Limbachiya, D., Gupta, M. K. & Aggarwal, V. Family of constrained codes for archival dna data storage. IEEE Communications Letters 22 (10), 1972–1975 (2018). https://doi.org/10.1109/LCOMM.2018. 2861867 . [11] Tabatabaei Yazdi, S. M. H., Yuan, Y., Ma, J., Zhao, H. & Milenkovic, O. A rewritable, random-access dna-based storage system. Scientific Reports 5 (1), 14138 (2015). URL https://doi.org/10.1038/srep14138. https://doi.org/10.1038/srep14138 . [12] Organick, L. et al. Random access in large-scale dna data storage. Nature Biotechnology 36 (3), 242–248 (2018). URL https://doi.org/10.1038/nbt. 4079. https://doi.org/10.1038/nbt.4079 . [13] Cao, B. et al. Achieving handle-level random access in an encrypted dna archival storage system via frequency dictionary mapping coding. Patterns 6 (9) (2025). URL https://doi.org/10.1016/j.patter.2025.101288. https://doi.org/10.1016/j.patter.2025.101288 . [14] Ping, Z. et al. Towards practical and robust dna-based data archiving using the yin–yang codec system. Nature Computational Science 2 (4), 234–242 (2022). URL https://doi.org/10.1038/s43588-022-00231-2. https://doi.org/10.1038/s43588-022-00231-2 . [15] Ping, Z. et al. Chamaeleo: a robust library for dna storage coding schemes. bioRxiv (2020). URL https://www.biorxiv.org/content/early/2020/ 03/19/2020.01.02.892588. https://doi.org/10.1101/2020.01.02.892588, https://www.biorxiv.org/content/early/2020/03/19/2020.01.02.892588.full.pdf . [16] Gimpel, A. L., Grass, R. N.

Remschak, A., Stark, W. J., Heckel, R. & Comparison of state-of-the-art error-correction

Springer Nature 2021 LATEX template The Unified Evaluation App for DNA Data Storage Codecs

31

coding for sequence-based dna data storage. bioRxiv (2025). URL https://www.biorxiv.org/content/early/2025/07/16/2025. 07.11.664297. https://doi.org/10.1101/2025.07.11.664297, https://www.biorxiv.org/content/early/2025/07/16/2025.07.11.664297.full.pdf . [17] Welzel, M. et al. Dna-aeon provides flexible arithmetic coding for constraint adherence and error correction in dna storage. Nature Communications 14 (1), 628 (2023). URL https://doi.org/10.1038/ s41467-023-36297-3. https://doi.org/10.1038/s41467-023-36297-3 . [18] Bar-Lev, D., Orr, I., Sabary, O., Etzion, T. & Yaakobi, E. Scalable and robust dna-based storage via coding theory and deep learning. Nature Machine Intelligence 7 (4), 639–649 (2025). URL https://doi.org/10. 1038/s42256-025-01003-z. https://doi.org/10.1038/s42256-025-01003-z . [19] DNA Data Storage Alliance. Dna data storage codecs — examples, requirements, and metrics. Tech. Rep., The Storage Networking Industry Association (SNIA) (2025). URL https://www.snia.org/educational-library/ dna-data-storage-codecs-examples-requirements-and-metrics-2025. [20] Shneiderman, B. The eyes have it: a task by data type taxonomy for information visualizations. Proceedings 1996 IEEE Symposium on Visual Languages 336–343 (1996). https://doi.org/10.1109/VL.1996.545307 . [21] Schwarz, M. et al. MESA: automated assessment of synthetic DNA fragments and simulation of DNA synthesis, storage, sequencing and PCR errors. Bioinformatics 36 (11), 3322–3326 (2020). URL https://doi.org/10.1093/ bioinformatics/btaa140. https://doi.org/10.1093/bioinformatics/ btaa140, https://academic.oup.com/bioinformatics/articlepdf/36/11/3322/50670740/bioinformatics 36 11 3322.pdf . [22] Harris, C. R. et al. Array programming with NumPy. Nature 585 (7825), 357–362 (2020). URL https://doi.org/10.1038/s41586-020-2649-2. https: //doi.org/10.1038/s41586-020-2649-2 . [23] Press, W. H., Hawkins, J. A., Jones, S. K., Schaub, J. M. & Finkelstein, I. J. Hedges error-correcting code for dna storage corrects indels and allows sequence constraints. Proceedings of the National Academy of Sciences 117 (31), 18489–18496 (2020). URL https://www.pnas. org/doi/abs/10.1073/pnas.2004821117. https://doi.org/10.1073/pnas. 2004821117, https://www.pnas.org/doi/pdf/10.1073/pnas.2004821117 . [24] Shomorony, I. & Heckel, R. Information-Theoretic Foundations of DNA Data Storage (Now Foundations and Trends, 2022). URL https://

Springer Nature 2021 LATEX template 32

The Unified Evaluation App for DNA Data Storage Codecs ieeexplore.ieee.org/document/9721712.

[25] Grass, R. N., Heckel, R., Puddu, M., Paunescu, D. & Stark, W. J. Robust chemical preservation of digital information on dna in silica with errorcorrecting codes. Angewandte Chemie International Edition 54 (8), 2552– 2555 (2015). URL https://onlinelibrary.wiley.com/doi/abs/10.1002/anie. 201411378. https://doi.org/https://doi.org/10.1002/anie.201411378, https://onlinelibrary.wiley.com/doi/pdf/10.1002/anie.201411378 . [26] Bornholt, J. et al. A dna-based archival storage system. SIGPLAN Not. 51 (4), 637–649 (2016). URL https://doi.org/10.1145/2954679.2872397. https://doi.org/10.1145/2954679.2872397 . [27] Schwarz, P. M. & Freisleben, B. Norec4dna: using near-optimal rateless erasure codes for dna storage. BMC Bioinformatics 22 (1), 406 (2021). URL https://doi.org/10.1186/s12859-021-04318-x. https://doi. org/10.1186/s12859-021-04318-x . [28] Shokrollahi, A. Raptor codes. IEEE Transactions on Information Theory 52 (6), 2551–2567 (2006). https://doi.org/10.1109/TIT.2006.874390 . [29] Maymounkov, P. Online codes. Tech. Rep., Computer Science Department at New York University, Warren Weaver Hall, Room 305, 251 Mercer Street, New York, NY 10012 (2002). URL https://pdos.csail.mit.edu/ ∼petar/papers/maymounkov-online.pdf. [30] Luby, M. Lt codes. The 43rd Annual IEEE Symposium on Foundations of Computer Science, 2002. Proceedings. 271–280 (2002). https://doi.org/ 10.1109/SFCS.2002.1181950 . [31] Jensen, K. et al. PASCAL user manual and report (Springer-Verlag, Berlin, Heidelberg, 1974). URL http://pascal.hansotten.com/uploads/ books/Pascal User Manual and Report Fourth Edition.pdf. [32] Christiansen, T., Foy, B. D., Wall, L. & Orwant, J. Programming Perl: Unmatched power for text processing and scripting (O’Reilly Media, Inc., 2012). [33] Van Rossum, G. & Drake, F. L. Python 3 Reference Manual (CreateSpace, Scotts Valley, CA, 2009). [34] Stroustrup, B. The C++ Programming Language, Third Edition 3rd edn (Addison-Wesley Longman Publishing Co., Inc., USA, 1997). [35] Verne, J. Around the World in Eighty Days By Jules Verne (BookRix, 1994). URL https://www.gutenberg.org/ebooks/103.

Springer Nature 2021 LATEX template The Unified Evaluation App for DNA Data Storage Codecs

33

[36] Wilde, O. The Picture of Dorian Gray Collection of British Authors: Tauchnitz Edition (Bernhard Tauchnitz, 1908). URL https://www. gutenberg.org/ebooks/174. [37] BBC Archive. BBC Sound Effects — sound-effects.bbcrewind.co.uk. https://sound-effects.bbcrewind.co.uk/search?q=07002136. [Accessed 01-11-2024]. [38] BBC Archive. BBC Sound Effects — sound-effects.bbcrewind.co.uk. https://sound-effects.bbcrewind.co.uk/search?q=NHU05015024. [Accessed 01-11-2024]. [39] van Rossum, G. & de Boer, J. Interactively testing remote servers using the python programming language. CWI Quarterly 4 (4), 283–304 (1991). URL https://ir.cwi.nl/pub/18204/18204A.pdf . [40] Yoo, A. B., Jette, M. A. & Grondona, M. Feitelson, D., Rudolph, L. & Schwiegelshohn, U. (eds) Slurm: Simple linux utility for resource management. (eds Feitelson, D., Rudolph, L. & Schwiegelshohn, U.) Job Scheduling Strategies for Parallel Processing, 44–60 (Springer Berlin Heidelberg, Berlin, Heidelberg, 2003). [41] Landsman, D. & Strauss, K. The dna data storage model. Computer 56 (7), 78–85 (2023). https://doi.org/10.1109/MC.2023.3272188 . [42] Chaykin, G., Sabary, O., Furman, N., Shabat, D. B. & Yaakobi, E. Dna-storalator: a computational simulator for dna data storage. BMC Bioinformatics 26 (1), 204 (2025). URL https://doi.org/10.1186/ s12859-025-06222-0. https://doi.org/10.1186/s12859-025-06222-0 . [43] Zhang, J. et al. Gungnir codec enabling high error-tolerance and lowredundancy dna storage through substantial computing power. Nature Communications 17 (1), 4828 (2026). URL https://doi.org/10.1038/ s41467-026-71485-x. https://doi.org/10.1038/s41467-026-71485-x .

Springer Nature 2021 LATEX template 34

The Unified Evaluation App for DNA Data Storage Codecs

Method

DNA_Aeon XORBasedCode

LTCode

OnlineCode

Filename=The_Picture_of_Dorian_Gray

RaptorCode

RepetitionCode

Filename=Vue_de_la_Buena_Vista

SimpleCode

The2DCode

Filename=Zero_File

100

) %( et a R s s e c c u S

80 60 40 20 0 0

0.05

0

Filename=all_black

0.05

0

Filename=bbc_clock

0.05

Filename=egypt

100

) %( et a R s s e c c u S

80 60 40 20 0 0

0.05

Filename=nature

0

0.05

0

Filename=Around_the_World_in_Eighty_Days

0.05

Filename=bbc_peregrine_falcon

100

) %( et a R s s e c c u S

80 60 40 20 0 0

0.05

Base_error_ratio

0

0.05

Base_error_ratio

0

0.05

Base_error_ratio

Fig. 10: Effect of base-error ratio on codec success rate across benchmark files. The plots show how data recovery changes as base-error ratio increases for each file and codec. Most codecs show a sharp reduction in success rate after errors are introduced, while DNA-Aeon maintains the most consistent recovery profile across the evaluated conditions.

Springer Nature 2021 LATEX template 35

The Unified Evaluation App for DNA Data Storage Codecs

Filename=The_Picture_of_Dorian_Gray

Filename=Vue_de_la_Buena_Vista

Filename=Zero_File

Transformation

1000

Decoding Encoding

100

) s( e mi T

10 1 0.1 0.01 L T C O n R a p X O D N R e p S im T h e L O R X D R S T L O R X D R S T o d lin e t o r R B a A _ A e t i p le 2 D T C o d n lin e a p t o r O R B a N A _ A e p e t i im p le h e 2 D T C o d n lin e a p t o r O R B a N A _ A e p e t i im p le h e 2 D e e e C o C o s e d e o t io n C o C o d C o C o s e d e o t io n C o C o d C o C o s e d e o t io n C o C o d de de de de de de Co de Co de Co de Co n Co n Co n e e e de de de de de de

Filename=all_black

Filename=bbc_clock

Filename=egypt

1000 100

) s( e mi T

10 1 0.1 0.01 L T C O n R a p X O D N R e p S im T h e L O R X D R S T L O R X D R S T o d lin e t o r R B a A _ A e t i p le 2 D T C o d n lin e a p t o r O R B a N A _ A e p e t i im p le h e 2 D T C o d n lin e a p t o r O R B a N A _ A e p e t i im p le h e 2 D e e e C o C o s e d e o t io n C o C o d C o C o s e d e o t io n C o C o d C o C o s e d e o t io n C o C o d de de de de de de Co de Co de Co de Co n Co n Co n e e e de de de de de de

Filename=nature

Filename=Around_the_World_in_Eighty_Days

Filename=bbc_peregrine_falcon

1000 100

) s( e mi T

10 1 0.1 0.01 L T C O n R a p X O D N R e p S im T h e L O R X D R S T L O R X D R S T o d lin e t o r R B a A _ A e t i p le 2 D T C o d n lin e a p t o r O R B a N A _ A e p e t i im p le h e 2 D T C o d n lin e a p t o r O R B a N A _ A e p e t i im p le h e 2 D e e e C o C o s e d e o t io n C o C o d C o C o s e d e o t io n C o C o d C o C o s e d e o t io n C o C o d de de de de de de Co de Co de Co de Co n Co n Co n e e e de de de de de de

Method

Method

(a) Error multiplier 0.0.

Method

Springer Nature 2021 LATEX template 36

The Unified Evaluation App for DNA Data Storage Codecs

Filename=The_Picture_of_Dorian_Gray

Filename=Vue_de_la_Buena_Vista

Filename=Zero_File

Transformation Decoding Encoding

10k

) s( e mi T

100

1

0.01 DN

L O R R S T X D N L T C O n R a p R e p S im T h e X O D N L T C O n R a p R e p S im T h e X O R R li li A _ A T C o d n lin e a p t o e p e t im p le h e 2 D O R B a A A o o 2 2 p p t t e e C o r C o it io n C o C o d s e d _ A e o d e n e C o o r C o t it io n le C o D C o d B a s e d _ A e o d e n e C o o r C o t it io n le C o D C o d B a s e d eon e de de de de de de n n Co de Co de Co de Co Co Co e e e de de de de de de

Filename=all_black

Filename=bbc_clock

Filename=egypt

10k

) s( e mi T

100

1

0.01 DN

L O R R S T X D N L T C O n R a p R e p S im T h e X O D N L T C O n R a p R e p S im T h e X O R R li li A _ A T C o d n lin e a p t o e p e t im p le h e 2 D O R B a A A o o 2 2 p p t t e e C o r C o it io n C o C o d s e d _ A e o d e n e C o o r C o t it io n le C o D C o d B a s e d _ A e o d e n e C o o r C o t it io n le C o D C o d B a s e d eon e de de de de de de n n Co de Co de Co de Co Co Co e e e de de de de de de

Filename=nature

Filename=Around_the_World_in_Eighty_Days

Filename=bbc_peregrine_falcon

10k

) s( e mi T

100

1

0.01 DN

L O R R S T X D N L T C O n R a p R e p S im T h e X O D N L T C O n R a p R e p S im T h e X O R R li li A _ A T C o d n lin e a p t o e p e t im p le h e 2 D O R B a A A o o 2 2 p p t t e e C o r C o it io n C o C o d s e d _ A e o d e n e C o o r C o t it io n le C o D C o d B a s e d _ A e o d e n e C o o r C o t it io n le C o D C o d B a s e d eon e de de de de de de n n Co de Co de Co de Co Co Co e e e de de de de de de

Method

Method

Method

(b) Error multiplier 0.5.

Fig. 11: Encoding and decoding runtime across codecs under different error conditions. Runtime is shown on a logarithmic scale because execution times varied by several orders of magnitude across codecs and benchmark files. Figure 11a shows the no-error condition, while Figure 11b shows the corresponding runtime profile at error multiplier 0.5. Missing entries indicate codec–file–error combinations that did not complete successfully under the default configuration used in this study.

Springer Nature 2021 LATEX template 37

The Unified Evaluation App for DNA Data Storage Codecs

Filename=The_Picture_of_Dorian_Gray

Filename=Vue_de_la_Buena_Vista

Filename=Zero_File

Transformation Decoding Encoding

1M 100k

) $(t s o C

10k 1000 100 L T C O n R a p X O D N R e p S im T h e L O R X D R S T L O R X D R S T o d lin e t o r R B a A _ A e t i p le 2 D T C o d n lin e a p t o r O R B a N A _ A e p e t i im p le h e 2 D T C o d n lin e a p t o r O R B a N A _ A e p e t i im p le h e 2 D e e e C o C o s e d e o t io n C o C o d C o C o s e d e o t io n C o C o d C o C o s e d e o t io n C o C o d de de de de de de Co de Co de Co de Co n Co n Co n e e e de de de de de de

Filename=all_black

Filename=bbc_clock

Filename=egypt

1M 100k

) $(t s o C

10k 1000 100 L T C O n R a p X O D N R e p S im T h e L O R X D R S T L O R X D R S T o d lin e t o r R B a A _ A e t i p le 2 D T C o d n lin e a p t o r O R B a N A _ A e p e t i im p le h e 2 D T C o d n lin e a p t o r O R B a N A _ A e p e t i im p le h e 2 D e e e C o C o s e d e o t io n C o C o d C o C o s e d e o t io n C o C o d C o C o s e d e o t io n C o C o d de de de de de de Co de Co de Co de Co n Co n Co n e e e de de de de de de

Filename=nature

Filename=Around_the_World_in_Eighty_Days

Filename=bbc_peregrine_falcon

1M 100k

) $(t s o C

10k 1000 100 L T C O n R a p X O D N R e p S im T h e L O R X D R S T L O R X D R S T o d lin e t o r R B a A _ A e t i p le 2 D T C o d n lin e a p t o r O R B a N A _ A e p e t i im p le h e 2 D T C o d n lin e a p t o r O R B a N A _ A e p e t i im p le h e 2 D e e e C o C o s e d e o t io n C o C o d C o C o s e d e o t io n C o C o d C o C o s e d e o t io n C o C o d de de de de de de Co de Co de Co de Co n Co n Co n e e e de de de de de de

Method

Method

(a) Error multiplier 0.0.

Method

Springer Nature 2021 LATEX template 38

The Unified Evaluation App for DNA Data Storage Codecs

Filename=The_Picture_of_Dorian_Gray

Filename=Vue_de_la_Buena_Vista

Filename=Zero_File

Transformation Decoding Encoding

1M 100k

) $(t s o C

10k 1000 100 DN

L O R R S T X D N L T C O n R a p R e p S im T h e X O D N L T C O n R a p R e p S im T h e X O R R li li A _ A T C o d n lin e a p t o e p e t im p le h e 2 D O R B a A A o o 2 2 p p t t e e C o r C o it io n C o C o d s e d _ A e o d e n e C o o r C o t it io n le C o D C o d B a s e d _ A e o d e n e C o o r C o t it io n le C o D C o d B a s e d eon e de de de de de de n n Co de Co de Co de Co Co Co e e e de de de de de de

Filename=all_black

Filename=bbc_clock

Filename=egypt

1M 100k

) $(t s o C

10k 1000 100 DN

L O R R S T X D N L T C O n R a p R e p S im T h e X O D N L T C O n R a p R e p S im T h e X O R R li li A _ A T C o d n lin e a p t o e p e t im p le h e 2 D O R B a A A o o 2 2 p p t t e e C o r C o it io n C o C o d s e d _ A e o d e n e C o o r C o t it io n le C o D C o d B a s e d _ A e o d e n e C o o r C o t it io n le C o D C o d B a s e d eon e de de de de de de n n Co de Co de Co de Co Co Co e e e de de de de de de

Filename=nature

Filename=Around_the_World_in_Eighty_Days

Filename=bbc_peregrine_falcon

1M 100k

) $(t s o C

10k 1000 100 DN

L O R R S T X D N L T C O n R a p R e p S im T h e X O D N L T C O n R a p R e p S im T h e X O R R li li A _ A T C o d n lin e a p t o e p e t im p le h e 2 D O R B a A A o o 2 2 p p t t e e C o r C o it io n C o C o d s e d _ A e o d e n e C o o r C o t it io n le C o D C o d B a s e d _ A e o d e n e C o o r C o t it io n le C o D C o d B a s e d eon e de de de de de de n n Co de Co de Co de Co Co Co e e e de de de de de de

Method

Method

Method

(b) Error multiplier 0.5.

Fig. 12: Estimated encoding and decoding cost across codecs under different error conditions. Costs are shown on a logarithmic scale because estimates varied by several orders of magnitude across codecs and benchmark files. Figure 12a shows the no-error condition, while Figure 12b shows the cost profile at error multiplier 0.5. Missing entries indicate codec–file–error combinations that did not complete successfully under the default configuration used in this study.

Related documents

• Choosing and Using a Plant DNA Barcode#335761
• Biological identifications through DNA barcodes#6793
• Learning-Enhanced Composite DNA Data Storage Under Sampling Randomness and IDS Errors#986200
• Comparison of state-of-the-art error-correction coding for sequence-based DNA data storage.#150258
• Learning-Enhanced Composite DNA Data Storage Under Sampling Randomness and IDS Errors#1079303