#21 · ML / AI methods

Similarity-driven methods advance biomolecular sequence and trajectory analysis

The common thread across these items is similarity-driven partitioning of biomolecular data, where pairwise comparison over large datasets is the principal constraint. FAMSA2 builds guide trees by medoid clustering with an LCS-based dissimilarity to align multiple protein sequences at scale, matching the accuracy of top tools while remaining fast. DOMUS applies average-linkage hierarchical clustering to an integrated similarity measure that blends multiple data views. iClust targets biological sequences with per-cluster prototypes and adaptive radii, followed by a consolidation step that keeps assignments interpretable. What is new is the pairing of similarity formulations suited to sequence and multi-view data with prototype-based, interpretable assignment. The items differ in scope, so this thread describes the dominant pattern rather than a method shared by all three.