Work

TDASSM: A Multi-Scale Topological Data Analysis and Selective State Space Model Framework for Copy Number Variation Detection

Year: 2026
Type: conference-paper
Abstract: Copy Number Variations (CNVs) are large scale genomic structural variations linked to different human diseases and genetic variation. The identification of CNVs based on next generation sequencing (NGS) data is still a challenge, because of sequencing noise, GC-content bias and uneven mapping of reads. We suggest a hybrid approach called TDASSM which consists of Topological Data Analysis (TDA) and a Selective State Space Model (SSM) to detect CNVs in whole genome sequencing data. First, normalized genomic signals are transformed to multi-scale topological representations on topologies built with Persistent Homology of signals on genomic windows of 20, 50 and 100 bases, resulting in a 432-dimensional signal representation for each genomic bin. The topology-aware features are then mapped through a memory efficient Selective SSM that can model genomic dependencies without the computational constraints of a transformer-based architecture. The model also uses a dualtask learning approach for classification of CNV and continual estimation of copy number. As a proof-of-concept analysis of a short sample set of humans from the 1000 Genomes Project with relatively high coverage of chromosome 22, the model successfully recovered several regions of the genome that were also supported by the DGV.
Language: English
Subfield: Genetics
Open Access status: gold
OpenAlex ID: W7223655893
FWCI:
Cites: 21
Cited by:
Related to: 10