DRAFT: graph-recurrent learning of dynamic gene regulatory networks along single-cell multi-omics trajectories
Year: 2026
Type: preprint
Abstract: Single-cell multi-omics data offer an opportunity to resolve how gene regulatory networks (GRNs) are rewired during cell-fate transitions, but most existing methods infer static snapshots, ignore temporal dependencies or assume linear regulation. Here we present DRAFT, a deep learning framework that combines graph attention with recurrent temporal memory to infer dynamic GRNs from time-resolved single-cell multi-omics data. On simulated data, DRAFT outperforms existing methods in accuracy and temporal smoothness. In mouse cellular reprogramming and Drosophila adult spermatogenesis, it recovers interactions supported by independent ChIP-seq, perturbation and Hi-C data, together with known stage-specific regulators. By inferring regulatory activity from dynamic network structure rather than expression alone, DRAFT identifies regulators whose activity precedes their transcriptional peak, including the candidate early germline regulator Mondo. Germline-specific Mondo knockdown alters germline stem cell number and expression of the predicted targets kuz and mamo. DRAFT also enables nonlinear alignment of co-developing lineages and reveals stage-specific cell-cell communication.
Institution: Heidelberg University
Language: English
Subfield: Molecular Biology
Domain: Life Sciences
Open Access status: green
Funder: Deutsche Forschungsgemeinschaft
OpenAlex ID: W7236097828
