#32 · Metabolomics

Structural prominence in genomic models does not establish functional mechanism

The common thread is interrogating pretrained or structured biological models against specific functional readouts rather than treating structural features as evidence of mechanism. In a genomic foundation model, high-gain gated-FFN rows show that structural prominence enriches for functional importance but does not calibrate it. A glycomics study recasts glycan motifs as a containment-ordered DAG with empirical-Bayes variance priors, raising true positives from 300 to 442 across 45 glycomics datasets. A transcription factor atlas of 1,552 binding datasets across 10 cell types, using ARES, a multi-agent system, shows that motif readers often differ from canonical TFs and shape regulatory output. What is new is the separation of prominence from mechanistic claims and attention to motif readers distinct from canonical TFs. The items are heterogeneous preprints, so conclusions remain provisional.