SeekBrain's multi-agent framework automates neuroscience discovery

SeekBrain, developed by Jiamin Wu and 28 collaborators, is an autonomous multi-agent system designed to tackle the integration of multi-scale, multimodal neuroscience data. The framework dynamically constructs a repertoire of analysis recipes extracted from code-paper pairs, coupling this codified expertise with agentic planning and execution engines. This allows SeekBrain to scalably generate hypotheses and analytical pipelines on demand, overcoming the fragmented workflows that often constrain discoveries in modern neuroscience. The system is built for hierarchical planning and cross-modal data analysis, making it adaptable to highly heterogeneous datasets. Instead of forcing researchers to manually stitch together tools for every data type, SeekBrain acts as an intelligent lab assistant that proposes and executes analysis steps based on established code-paper knowledge.

Systematic evaluation on the expert-annotated BrainArena benchmark showed SeekBrain substantially outperforming state-of-the-art agent baselines across various analysis tasks. More importantly, in real-world research deployments, the framework successfully integrated behavioral, neural, and anatomical data to uncover structured, distributed neural representations of larval zebrafish behavior. It also revealed a shared axis of regional decoding strength across the brain in a mouse decision-making task. These results position SeekBrain as a scalable, practical tool for accelerating data-driven discovery in neuroscience, potentially enabling researchers to automate complex analyses, validate hypotheses faster, and extract insights from multimodal datasets that might otherwise remain hidden. For the broader AI community, SeekBrain demonstrates how multi-agent systems can move beyond chatbots into specialized scientific discovery.