BrainWideBench: Benchmarking large-scale pretraining and across-animal transfer in multi-region neural recordings
Abstract
Advances in large-scale neural recording have made it possible to collect data across many animals and distributed brain regions, raising the question of whether this scale can be exploited to learn general-purpose neural representations transferable across diverse downstream tasks. Yet, progress toward this goal has been limited by fragmented evaluation protocols and a narrow focus on individual task domains. Here, we present BrainWideBench, a benchmark for evaluating across-animal transfer on multi-region neural recordings, built on the International Brain Laboratory Brainwide Map dataset of neural and behavioral recordings spanning 276 brain regions from 139 mice performing a sensory-guided decision-making task. The benchmark is organized around three complementary task suites that evaluate whether learned representations support downstream decoding of behavior, can predict masked or future neural activity, and can recover biologically meaningful anatomical organization. With this benchmark, we systematically evaluate pretraining methods across transfer settings, including finetuning on downstream objectives and zero-shot generalization to unseen animals. Our results confirm pretraining improves performance over matched single-session baselines, but we show current methods exhibit heterogeneity in transfer capabilities: gains depend strongly on the alignment between pretraining objectives and downstream tasks. No single approach performs uniformly well across all three suites, and most methods are designed to only address a subset of them. Together, these findings suggest that learning representations that jointly generalize across behavior, dynamics, and anatomy remains an open challenge. By providing a unified and reproducible evaluation suite, BrainWideBench establishes a framework for measuring progress toward general-purpose models of the mouse brain.
Related
- Creator 1.0: Phylogeny-Guided Stagewise Pretraining for Plant Genome Foundation Models across 1,000 Species
- OncoSpot enables benchmarking of cancer spatial transcriptomics methods across tumor evolutionary modes
- Benchmarking cartilage regeneration across ten chondrogenic cell sources using native tissue references
- Benchmarking Peptide-Protein Affinity Prediction Across Peptide and Target Shifts
- Top2All: Benchmarking sparse-to-transcriptome reconstruction from high-expression profiles across sex- and sex-steroid-defined conditions
BibTeX
@misc{andre2026brainwidebench,
title = {{BrainWideBench: Benchmarking large-scale pretraining and across-animal transfer in multi-region neural recordings}},
author = {Alexandre Andre and Shivashriganesh P. Mahato and Vinam Arora and Keshav Balaji and Divyansha Lachi and Nanda H. Krishna and Jingyun Xiao and Yizi Zhang and Ximeng Mao and Wenrui Ma and Han Yu and International Brain Laboratory and Daniel Birman and Niccolò Bonacchi and Gaelle A. Chapuis and Joana A. Catarino and Felicia Davatolhagh and Mayo Faulkner and Laura Freitas-Silva and Fei Hu and Julia M. Huntenburg and Anup Khanal and Inês Laranjeira and Petrina Lau and Guido T. Meijer and Nathaniel J. Miska and Jean-Paul Noel and Alejandro Pan-Vazquez and Georg Raiser and Cyrille Rossant and Karolina Z. Socha and Anne E. Urai and Miles J. Wells and Steven J. West and Olivier Winter and Blake Richards and Guillaume Lajoie and Cole Hurwitz and Mehdi Azabou and Matthew R. Whiteway and Liam Paninski and Eva L. Dyer},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2609.22064},
eprint = {2609.22064},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.22064},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - BrainWideBench: Benchmarking large-scale pretraining and across-animal transfer in multi-region neural recordings
AU - Alexandre Andre
AU - Shivashriganesh P. Mahato
AU - Vinam Arora
AU - Keshav Balaji
AU - Divyansha Lachi
AU - Nanda H. Krishna
AU - Jingyun Xiao
AU - Yizi Zhang
AU - Ximeng Mao
AU - Wenrui Ma
AU - Han Yu
AU - International Brain Laboratory
AU - Daniel Birman
AU - Niccolò Bonacchi
AU - Gaelle A. Chapuis
AU - Joana A. Catarino
AU - Felicia Davatolhagh
AU - Mayo Faulkner
AU - Laura Freitas-Silva
AU - Fei Hu
AU - Julia M. Huntenburg
AU - Anup Khanal
AU - Inês Laranjeira
AU - Petrina Lau
AU - Guido T. Meijer
AU - Nathaniel J. Miska
AU - Jean-Paul Noel
AU - Alejandro Pan-Vazquez
AU - Georg Raiser
AU - Cyrille Rossant
AU - Karolina Z. Socha
AU - Anne E. Urai
AU - Miles J. Wells
AU - Steven J. West
AU - Olivier Winter
AU - Blake Richards
AU - Guillaume Lajoie
AU - Cole Hurwitz
AU - Mehdi Azabou
AU - Matthew R. Whiteway
AU - Liam Paninski
AU - Eva L. Dyer
PY - 2026
DA - 2026/09/18
PB - arXiv
DO - 10.48550/arxiv.2609.22064
UR - https://arxiv.org/abs/2609.22064
AB - Advances in large-scale neural recording have made it possible to collect data across many animals and distributed brain regions, raising the question of whether this scale can be exploited to learn general-purpose neural representations transferable across diverse downstream tasks. Yet, progress toward this goal has been limited by fragmented evaluation protocols and a narrow focus on individual task domains. Here, we present BrainWideBench, a benchmark for evaluating across-animal transfer on multi-region neural recordings, built on the International Brain Laboratory Brainwide Map dataset of neural and behavioral recordings spanning 276 brain regions from 139 mice performing a sensory-guided decision-making task. The benchmark is organized around three complementary task suites that evaluate whether learned representations support downstream decoding of behavior, can predict masked or future neural activity, and can recover biologically meaningful anatomical organization. With this benchmark, we systematically evaluate pretraining methods across transfer settings, including finetuning on downstream objectives and zero-shot generalization to unseen animals. Our results confirm pretraining improves performance over matched single-session baselines, but we show current methods exhibit heterogeneity in transfer capabilities: gains depend strongly on the alignment between pretraining objectives and downstream tasks. No single approach performs uniformly well across all three suites, and most methods are designed to only address a subset of them. Together, these findings suggest that learning representations that jointly generalize across behavior, dynamics, and anatomy remains an open challenge. By providing a unified and reproducible evaluation suite, BrainWideBench establishes a framework for measuring progress toward general-purpose models of the mouse brain.
ER -