#30 · ML / AI methods

Encoding biological priors into graph, clock, and metabolic models

The dominant thread is the explicit encoding of biological structure into computational models to predict regulation, aging, and metabolic behavior. rwMagLap places rhythmic genes on a Hermitian magnetic Laplacian to rank candidate circadian regulators of arrhythmic pituitary hormone genes, while TFMethyl Clock links epigenetic clock CpGs to transcription factor binding sites to predict chronological age. tDamAge quantifies transcriptomic damage, defined as aberrant transcript structures from RNA-seq, within a machine-learning aging clock. VirEvo combines AlphaGenome-based virtual dual-luciferase scoring with a genetic algorithm to compress the p16INK4a promoter. MitoCore and iHME are curated, downsized human metabolic models; iHME recalls 84.6% of essential genes in cancer metabolic dependency prediction, and MitoCore supports protein-constrained modeling with proteomics. These items are heterogeneous. Only rwMagLap and TFMethyl Clock are representation-style models over biological graphs or sites, while the metabolic reconstructions are mechanistic rather than learned. What is new is the specific prior attached to each task: a magnetic graph operator, TF-binding annotation, a transcript-structure measure, or a curated network.