Cebra

CEBRA is a machine learning tool that utilizes non-linear techniques to generate reliable and high-performing latent spaces from joint behavioural and neural data collected simultaneously. This tool enables the correlation between behavioural actions and neural activity to be mapped, leading to a deeper understanding of neural dynamics during adaptive behaviours and the identification of underlying behaviour-related factors. CEBRA produces neural latent embeddings that can be utilized for hypothesis testing and discovery-driven analysis. Its accuracy and effectiveness have been validated across various datasets, including calcium and electrophysiology data, as well as in sensory and motor tasks and both simple and complex behaviours across different species. It can be applied to single or multi-session datasets without the need for labels. CEBRA is capable of mapping and uncovering intricate kinematic features, generating consistent latent spaces across 2-photon and Neuropixels data, and achieving rapid and highly accurate decoding of natural movies from the visual cortex. The tool's source code is available on GitHub, and a pre-print version can be accessed on arxiv.org. Neuroscientists who aim to analyze and decode behavioural and neural data to unveil underlying neural representations will find CEBRA to be a valuable tool.