GSET: Gene Signals from Embedding Trajectories select compact contexts for mechanistic GRN inference

GSET improves mechanistic GRN inference from temporal transcriptomics by selecting compact, program-informed gene sets that recover more regulatory links than static selections at equal gene budget.

• Why it matters: Static gene selection can miss dynamically informative genes, making it hard to recover transcriptional relationships that change over time while keeping computation feasible at transcriptome scale.
• What they did: GSET retrieves genes from temporal gene embeddings using an optimal transport residual and a KL divergence over local co-expression neighborhoods, then integrates them with CardamomOT to form compact gene sets.
• The result: In synthetic benchmarks, the retrieved sets recover more regulatory links than static selections of the same size, and edge inference within the set is comparable to the full-gene model on the same gene pairs.

AI summary of the abstract · may contain errors · not medical advice · read the paper.