cs.LGMar 11, 2026

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

Authors: Theo Schwider, Ramin Ramezani

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

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.

Figures & tables

Feature familyDescription
first_ap_vVoltage features of the first action potential, such as threshold, peak, amplitude, and afterhyperpolarization.
first_ap_dvDerivatives of the membrane potential around the first spike (e.g., dV/dtdV/dt at threshold and peak).
isi_shapeInter-spike interval metrics, including adaptation indices and ISI variability.
inst_freqInstantaneous firing rate measures as a function of time and stimulus amplitude.
spiking_threshold_vSpike threshold estimates across sweeps.
spiking_peak_vSpike peak voltages across sweeps.
Table 2: Summary of the 12 IPFX electrophysiological feature families used in this study, following the feature-family taxonomy described by Gouwens et al. (2020) and extracted using the Allen Institute IPFX pipeline (Allen Institute for Brain Science, 2025; Gouwens et al., 2020).
Mouse data classCount (%)
Lamp5402 (10.9%)
Pvalb745 (20.1%)
Sncg198 (5.4%)
Sst1663 (45.0%)
Vip691 (18.7%)
Table 3: Class distributions after QC (as reported in the draft).
Figure 1: UMAP comparison used to validate replication of the baseline feature pipeline.
Figure 2: Random forest confusion matrix on mouse test split (5-class).
Figure 3: Random forest confusion matrix on human evaluation (4-class; 5-fold stratified cross-validation).
Figure 4: Attention-weight summaries for interpretability in mouse (mean attention weights by feature family and class).
Figure 5: Attention-weight summaries for interpretability in human (mean attention weights by feature family and class).
Model variantMacro-F1 (k-fold avg)Accuracy (k-fold avg)
Baseline BiLSTM0.6580 ±\pm 0.02860.7710 ±\pm 0.0172
Dual pretrained transfer learning0.6795 ±\pm 0.01200.7905 ±\pm 0.0096
Table 6: Transfer-learning results on the aligned 4-class task (human evaluated with 5-fold stratified cross-validation).
Figure 6: Recap figure (reproduced for convenience). UMAP embedding of mouse inhibitory interneurons in the electrophysiology-derived low-dimensional space (IPFX features with the same dimensionality-reduction pipeline described in Materials and Methods). This panel is repeated here as a compact visual reference for the Discussion and does not present additional analyses.
Figure 7: Recap figure (reproduced for convenience). Representative confusion matrices (mouse left; human right) for the feature-based baseline classifier. These panels are repeated here as an at-a-glance summary of dominant error modes and class-imbalance effects; no new model fits are introduced in this figure.
More in this topic

Classification

Classification maps observations to discrete labels, including familiar, unknown, hierarchical, or multiple categories.

See all 408 papers

Explore similar work

CardsList

  1. Language as the Interface: Foundation-Model Contrastive Learning Links Transcriptomes and Electrophysiology

    Sep 29, 2026Junbo Shen, Jinying Gao, Bo LeiCross-Modal AlignmentMultimodal Foundation Models

    Department of Computer Science and Engineering, The Chinese University of Hong Kong · State Key Laboratory of Brain Cognition and Brain-inspired Intelligence Technology, Institute of Automation, Chinese Academy of Sciences, Beijing, China · Beijing Academy of Artificial Intelligence, Beijing, China +1

  2. Cross-species representation learning aligns mouse and human neural dynamics and tracks clinical drug efficacy

    Oct 8, 2026Marko Tvrdic, Justin Richmond Domingo, Jae Ann Buenaluz +4Representation LearningContrastive Learning

    Exin Therapeutics, Inc., 2261 Market Street STE 22614, San Francisco, California 94114, United States

  3. Towards a Pseudo-Labeling Workflow for Celltype-Classification from Explanted Brain Slice Recordings

    Jun 15, 2026Cora Jostock, Jonas Ort, Henner Koch +2Unsupervised ClusteringPseudo-Labeling

    University of Duisburg-Essen, Intelligent Embedded Systems Lab, Duisburg, Germany · RWTH University Hospital, Department of Neurosurgery, Aachen, Germany · RWTH University Hospital, Department of Epileptology, Aachen, Germany