Shared-AE: Automatic Identification of Shared Subspaces in High-dimensional Neural and Behavioral Activity.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Machine Learning

Background:

• Understanding brain function requires linking neural activity to behavior.
• Analyzing complex behaviors like social interactions necessitates joint feature extraction from neural and behavioral data.
• Existing methods struggle with the joint analysis of intricate behaviors and neural signals.

Purpose of the Study:

• To develop a novel autoencoder (AE) framework, Shared-AE, for identifying shared and private features between neural activity and complex behaviors.
• To enhance the understanding of the alignment between neural representations and behavioral outputs.
• To provide a computational tool for analyzing large-scale neural recordings during diverse behavioral paradigms.

Main Methods:

• Implementation of an autoencoder (AE) framework named Shared-AE.
• Inclusion of a novel regularization term to identify shared features across modalities.
• Simultaneous capture of unique private features specific to neural activity and behavior.
• Application to large-scale neural activity data from mice during decision-making and social behaviors.

Main Results:

• Shared-AE successfully identified both shared features (across neural and behavioral activity) and private features (unique to each modality).
• The model demonstrated effectiveness in analyzing neural data from distinct behavioral contexts (decision-making and social interaction).
• Significant enhancement in understanding the neural-behavioral alignment was achieved.

Conclusions:

• The Shared-AE framework provides a powerful method for dissecting the relationship between neural dynamics and complex behaviors.
• This approach advances the joint analysis of multimodal neural and behavioral data.
• The publicly available code facilitates further research in computational neuroscience and brain-computer interfaces.