Paper List

期刊: ArXiv Preprint
发布日期: 2026-03-11
BioinformaticsNeuroscience

Cross-Species Transfer Learning for Electrophysiology-to-Transcriptomics Mapping in Cortical GABAergic Interneurons

Unknown

Theo Schwider, Ramin Ramezani
Figure
Figure
Figure
Figure
Figure

30秒速读

IN SHORT: This paper addresses the challenge of predicting transcriptomic identity from electrophysiological recordings in human cortical interneurons, where limited labeled data and class imbalance hinder model performance.

核心创新

  • Methodology Developed an attention-based BiLSTM that operates directly on structured IPFX feature-family representation (12 families, 498 features), eliminating the need for sparse PCA preprocessing and providing interpretable attention weights over feature families.
  • Methodology Implemented a cross-species transfer learning framework with joint supervised training (shared encoder + two heads) followed by human-only fine-tuning, improving human macro-F1 by leveraging abundant mouse data (3,699 cells) to augment limited human data (506 cells).
  • Biology Demonstrated conserved electrophysiological-to-transcriptomic mapping across species for GABAergic interneuron subclasses (Lamp5, Pvalb, Sst, Vip), enabling meaningful cross-species transfer despite biological and experimental distribution shifts.

主要结论

  • Successfully replicated the Gouwens et al. (2020) baseline with random forest achieving 90.72% accuracy and 0.8728 macro-F1 on mouse data, confirming reproducibility of the electrophysiology-to-transcriptomics pipeline.
  • The attention-based BiLSTM with SMOTE and ArcFace achieved 0.8923 macro-F1 on mouse data, matching feature-engineered baselines while providing interpretable attention weights over 12 electrophysiological feature families.
  • Cross-species transfer learning (mouse pretraining + human fine-tuning) improved human macro-F1 compared to human-only training, demonstrating measurable gains despite distribution shifts and limited human sample size.
研究空白： Human Patch-seq datasets are typically smaller (506 cells vs. 3,699 mouse cells), more imbalanced (Pvalb: 57.9% vs. Lamp5: 9.9%), and suffer from experimental distribution shifts, making direct application of mouse-trained models challenging and limiting prediction accuracy.

摘要: Single-cell electrophysiological recordings provide a powerful window into neuronal functional diversity and offer an interpretable route for linking intrinsic physiology to transcriptomic identity. Here, we replicate and extend the electrophysiology-to-transcriptomics framework introduced by Gouwens et al. (2020) using publicly available Allen Institute Patch-seq datasets from both mouse and human cortex. We focus on GABAergic inhibitory interneurons to target a subclass structure (Lamp5, Pvalb, Sst, Vip) that is comparable and conserved across species. After quality control, we analyzed 3,699 mouse visual cortex neurons and 506 human neocortical neurons from neurosurgical resections. Using standardized electrophysiological features and sparse PCA, we reproduced the major class-level separations reported in the original mouse study. For supervised prediction, a class-balanced random forest provided a strong feature-engineered baseline in mouse data and a reduced but still informative baseline in human data. We then developed an attention-based BiLSTM that operates directly on the structured IPFX feature-family representation, avoiding sPCA and providing feature-family-level interpretability via learned attention weights. Finally, we evaluated a cross-species transfer setting in which the sequence model is pretrained on mouse data and fine-tuned on human data for an aligned 4-class task, improving human macro-F1 relative to a human-only training baseline. Together, these results confirm reproducibility of the Gouwens pipeline in mouse data, demonstrate that sequence models can match feature-engineered baselines, and show that mouse-to-human transfer learning can provide measurable gains for human subclass prediction.