#37 · ML / AI methods

Explicit probabilistic models drive genomic and biological inference tasks

The dominant thread is explicit probabilistic specification, through a likelihood, prior, or joint density, carrying inference over genetic and biological data. An approximate-copula family with explicit moments and a likelihood score is applied to longitudinal modeling and tri-variate GWAS in UK Biobank, extending dependence modeling beyond marginal structure. baycn applies approximate Bayesian inference with flexible edge-state priors to infer DAG edge directions in biological networks, including genomic target-gene data. An empirical Bayes f-modeling EM algorithm with causally structured score matching targets linear causal representations across domains. Prediction-powered inference contributes closed-form power formulas for ML-based predictions, validated by simulation. The newer elements are copula-based correlation structure, cross-domain causal representation, and directed network topology. Two of the four records are published articles and two are preprints.