#39 · ML / AI methods

Alignment-free k-mer and distribution methods infer and score human centromere structure

The dominant thread is computational characterization of human centromeres in T2T data, where structure is assessed through sequence distributions and k-mer signatures rather than read or assembly alignment. A mathematical framework scores T2T genome assembly accuracy by comparing centromeric inter-motif distances via KL divergence, without alignment. HORoSCOPE applies k-mer signatures to infer centromere architecture and length from short reads, trained on 11,836 T2T centromeres. What is new is the shift toward distribution-level, alignment-independent descriptors for both evaluating and recovering centromere structure. Two remaining preprints are not centromere-specific: Hyper3D-lite audits count-preserving representations for long-read multi-contact genome data, and topology-aware thermodynamic scores address DNA probe specificity under fixed stringency.