Human to Mouse Cross-Species Transfer Learning forElectrophysiology-to-Transcriptomics Mapping in CorticalGABAergic Interneurons

Single-cell electrophysiological recordings provide a powerful window into neuronal functional diversityand offer an interpretable route for linking intrinsic physiology to transcriptomic identity. Here, wereplicate and extend the electrophysiology-to-transcriptomics framework introduced by Gouwenset al. [ 1] using publicly available Allen Institute Patch-seq datasets from mouse and human cortex.We focused on GABAergic inhibitory interneurons in an aligned subclass structure (Lamp5, Pvalb,Sst, Vip). After quality control, we analyzed 3,699 mouse visual cortex neurons and 506 humanneocortical neurons from neurosurgical resections. Using standardized electrophysiological featuresand sparse PCA, we reproduced major class-level separations reported in the original mouse study. Forsupervised prediction, a class-balanced random forest provided a strong feature-engineered baselinein mouse data and a reduced but still informative baseline in human data. We then developed anattention-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 cross-species transfer by pretraining on mouse data and fine-tuning on humandata for an aligned 4-class task, improving human macro-F1 versus human-only training. Together,these results support reproducibility of the Gouwens pipeline and show measurable transfer-learninggains for human subclass prediction.