Extended Data Fig. 3: Transfer learning adapts GET to new platforms and cell types.

a. GET trained on fetal cell types generalizes to adult cell types without retraining, outperforming the most correlated cell type baseline. X axis shows R² score between GET prediction in adult cell types and observed expression in the most similar fetal cell types. Y axis shows R² score between GET prediction and observed expression in the adult cell type. b. Schematic illustration of transferring GET to a lymph node 10x multiome dataset. c. Finetuned GET accurately predicts expression in training and leave-out evaluation lymph node cell types. d. Schematic showing the application of GET in the zero-shot setting to predict gene expression from glioblastoma (GBM) patient samples (top) and in the one-shot setting after being finetuned on a single GBM patient sample and used to predict gene expression for an extended cohort of GBM patients. e. Pearson correlation scores for GET expression prediction on GBM cells (n = 16 samples) comparing tumor cells, macrophages, and oligodendrocytes for zero-shot and one-shot (finetuned) settings. f. Radar plot showing leave-one-chromosome-out finetuning performance (R², Pearson correlation, Spearman correlation) of GET in one GBM tumor sample. Schematics in b and d created using BioRender (https://biorender.com).