HIPPIE: A Multimodal Deep Learning Model for Electrophysiological Classification of Neurons.

Area of Science:

• Neuroscience
• Computational Biology
• Machine Learning

Background:

• Extracellular electrophysiological recordings are crucial for neuroscience research but present significant computational challenges.
• Noise, technical variability, and batch effects hinder accurate neuronal classification and clustering.

Purpose of the Study:

• To introduce HIPPIE (High-dimensional Interpretation of Physiological Patterns In Extracellular recordings), a deep learning framework for robust neuronal classification.
• To develop a method that addresses noise and batch effects in electrophysiological data.

Main Methods:

• HIPPIE utilizes a deep learning framework with self-supervised pretraining and supervised fine-tuning.
• Conditional convolutional joint autoencoders are employed to learn representations of waveforms and spiking dynamics.
• The framework integrates unlabeled datasets for robust representation learning.

Main Results:

• HIPPIE demonstrated superior performance in cell-type discrimination compared to unsupervised methods on both in vivo and in vitro recordings.
• The model's latent space effectively organizes neurons along electrophysiological gradients.
• HIPPIE enabled batch and individual corrected alignment of recordings across diverse experimental systems.

Conclusions:

• HIPPIE provides a general framework for decoding neuronal diversity from extracellular recordings.
• The deep learning approach offers a robust solution for electrophysiological classification and clustering.
• This method facilitates systematic analysis across diverse biological cultures and technologies.