Validation of a Convolutional Neural Network Model for Spike Transformation Using a Generalized Linear Model.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Neural Engineering

Background:

• Identifying causal relationships in neural activity is crucial for understanding brain function.
• Existing methods face challenges in accurately modeling complex neural dynamics.

Purpose of the Study:

• To introduce and validate a novel deep learning approach for modeling neural spike activity.
• To assess the ability of a convolutional neural network to predict output neural activity from input activity.

Main Methods:

• Utilized a convolutional neural network (CNN) to model neural spike activity.
• Trained and validated the CNN on data generated from a generalized linear model (GLM).
• Evaluated the correlation between predicted and true spiking probabilities and the recovery of model parameters (kernels).

Main Results:

• The CNN achieved high correlation between predicted and true probabilities of output neuron spiking.
• The CNN successfully recovered the true model variables (kernels) used in the GLM.
• Demonstrated the model's efficacy in a controlled GLM environment.

Conclusions:

• Deep learning, specifically CNNs, offers a powerful tool for identifying causal relationships in neural activity.
• The validated CNN model provides a foundation for future research with more complex, non-linear neural models.
• This work contributes to advancing predictive modeling in neuroscience and neural engineering.