Official Review · ICLR 2025

The paper focuses on understanding the predictability of neuronal activity using zebrafish as a model organism. The study introduces the Zebrafish Activity Prediction Benchmark (ZAPBench), a structured framework designed to forecast neuronal behavior at single-cell resolution. The primary research question is centered around how accurately future neuronal activity can be predicted based on prior observations. This investigation aims to uncover fundamental predictability limits in complex neural systems, providing a formal benchmark that can guide and standardize future research in this domain. Key contributions of the paper include: The creation of ZAPBench, a dataset tailored for predictive neuroscience, focusing on single-cell activity in zebrafish. A comprehensive methodology to assess prediction accuracy across various models. Foundational insights into the neural predictability that could enhance brain function understanding and improve modeling standards in neuroscience. The framework is intended to encourage reproducibility and provide a standard for benchmarking advancements in neuronal activity prediction methods. Originality: The paper demonstrates originality by addressing the challenge of predicting neuronal activity with a new focus on zebrafish, providing a dataset that enables prediction at single-cell resolution. By framing a well-defined benchmark (ZAPBench) and emphasizing zebrafish as a model organism, the study introduces a valuable structure for exploring neuronal predictability. This originality lies in applying prediction models to a high-resolution, single-cell level dataset, which pushes beyond the coarser frameworks commonly used in neural forecasting. Quality: The research is grounded in solid methodology, with a detailed approach to constructing the ZAPBench framework and rigorously evaluating prediction models. By providing an extensive dataset and clear benchmark criteria, the authors lay a high-quality foundation for reproducible research. The paper's methodology allows for systematic assessment and comparison of various forecasting methods, thereby contributing significantly to the reliability and comprehensiveness of the research. Clarity: The paper’s clarity is evident in its systematic presentation of the problem, methodology, and dataset. It carefully articulates the goals, including the potential limits of predictability in neuronal systems, and provides transparent descriptions of the benchmarks and data structures. Additionally, the focus on single-cell resolution is well-explained, helping readers grasp the significance of such granular predictability in the context of neuroscience. Significance: The significance of this work is substantial, given the pressing need for robust frameworks in neuroscience to predict and understand brain function. By creating ZAPBench, the authors lay the groundwork for a standardized method to evaluate neuronal predictability, likely stimulating future research and prac