#15 · ML / AI methods

Multi-omics frameworks integrate single-cell, methylation, and chromatin readouts

The common thread is computational frameworks that combine multiple data modalities, models, or learned representations to infer regulatory and epigenomic states. scTranslation benchmarks translation between single-cell multi-omics modalities, and TopOmics applies topic modelling across any combination of omics types with benchmarks against existing methods. Ensemble strategies recur: fourSynergy weights existing 4C-seq interaction callers by gradient-free optimization, and the enhancer-promoter framework integrates multi-cell-line features with an XGBoost meta-classifier. Deep models target chromatin and methylation directly, with THC-net predicting A/B compartments from six histone modifications and MethyNano using supervised contrastive pretraining for nanopore 5mC detection. Brieflow adds an end-to-end pipeline for optical pooled screening data. What is new is the emphasis on multimodal integration within unified, benchmarked frameworks, although the items remain methodologically heterogeneous.