Mouse to Human Cross-Species Transfer Learning for Electrophysiology-to-Transcriptomics Mapping in Cortical GABAergic...

Area of Science:

• Neuroscience
• Computational Biology
• Genomics

Background:

• Single-cell electrophysiology reveals neuronal functional diversity.
• Linking intrinsic physiology to transcriptomic identity is crucial for understanding brain function.
• Existing methods utilize electrophysiological feature organization for neuronal classification.

Purpose of the Study:

• To adapt and apply the Gouwens electrophysiological feature-family organization to mouse and human inhibitory interneurons.
• To develop and evaluate a novel attention-based BiLSTM model for neuronal classification.
• To assess the effectiveness of cross-species transfer learning for improving human neuronal classification.

Main Methods:

• Analysis of Allen Institute Patch-seq datasets from mouse and human cortex.
• Application of standardized electrophysiological features and sparse-PCA.
• Development of a class-balanced random forest and an attention-based BiLSTM model.
• Evaluation of cross-species transfer learning by pretraining on mouse data and fine-tuning on human data.

Main Results:

• Major class-level separations were recovered for mouse and human inhibitory interneurons.
• A random forest provided a strong baseline for mouse data and an informative baseline for human data.
• The attention-based BiLSTM model offered feature-family-level interpretability.
• Cross-species transfer learning improved human neuronal classification performance.

Conclusions:

• The Gouwens feature-family organization is useful for analyzing neuronal diversity across species.
• The developed neural sequence model provides an interpretable alternative to compressed baselines.
• Cross-species transfer learning enhances the classification of human neurons, aiding in comparative neuroscience research.