#33 · ML / AI methods

Causal evaluation and graph learning for gene regulatory and disease networks

The dominant thread is network-based causal modeling of biological systems rather than any single method. Practical Causal Evaluation Metrics for Biological Networks proposes sign-augmented Structural Intervention Distance (sSID) and a weighted variant to score inferred gene regulatory networks against qualitative prior knowledge. Causal-GNN couples causal effect estimation with graph neural network propensity scoring over gene regulatory networks to identify stable biomarkers through feature selection. The longitudinal blood-data framework converts time series into correlation-based causal disease networks, with explicit thresholding strategies for network construction and subsequent intervention analysis. What is new across these entries is an effort to make causal claims operational: evaluation metrics that incorporate sign and prior information, and network-level causal estimation applied to biomarker discovery and clinical longitudinal data.