DCSMAP: A Hybrid Linear-Pangenome Framework for Accurate and Efficient Read Alignment

Linear alignment, XGBoost-based selection of reads for remapping to a personalized pangenome graph, then local indel realignment

DCSMAP is a hybrid alignment workflow: reads are first aligned to the linear reference, then an XGBoost classifier trained on linear-alignment features selects reads for remapping to a personalized pangenome graph, followed by local indel realignment. The classifier achieved an AUC of 0.9831 on held-out test data and selected a mean of 14.80% of read pairs for graph remapping across 17 runs covering seven Genome in a Bottle samples, two sequencing platforms and two read lengths. Variant-calling errors were reduced by 46.66% relative to BWA-MEM2 and by 8.39% relative to vg personalized-pangenome alignment, with reference bias comparable to that of pangenome alignment. On HG003 DNBSEQ PE150 data, DCSMAP ran 2 to 2.5 times faster than the personalized-pangenome workflow on a standard CPU server.

Synthesis written by Geno'X. For the full original abstract, please refer to the source publication.

Preprint, not peer reviewed. The idea is pragmatic: pay the cost of the pangenome graph only for the fraction of reads whose linear placement is doubtful, making reference-bias reduction accessible without heavy infrastructure. Validation relies solely on GIAB reference samples; the effect on pathogenic variants in difficult regions and on diagnostic yield remains to be shown, and the abstract does not state code availability.

Analysis by Dr Thibaut Benquey