Proof-of-concept study on an integrated multiplex pyrosequencing system and CLPSQ-Net algorithm for multiplex microhaplotypes genotyping.

Microhaplotypes (MHs), characterized by short fragment lengths and high polymorphism, hold great promise for forensic applications. Here, we present a streamlined multiplex MHs detection system based on pyrosequencing (PSQ), incorporating several key innovations: (1) selection of highly polymorphic short-fragment MHs from the 1000 Genomes Project (1KGP); (2) algorithmic optimization of nucleotide dispensation orders for accurate haplotyping; (3) utilization of peak-height simulation to augment the dataset, overcoming the limitation of scarce empirical data for machine learning; and (4) CLPSQ-Net, a deep contrastive learning framework for deconvoluting multiplex PSQ signals. In this feasibility study, the system demonstrated simultaneous genotyping of four MH loci (6-7 haplotypes per locus) and an input sensitivity of 100 pg of DNA in a preliminary evaluation. Simulated and experimental signals showed high concordance (cosine similarity >0.98 for uniplex, >0.99 for multiplex). CLPSQ-Net achieved 89.7 % classification accuracy and an F1-score of 88.4 % on experimental data, outperforming traditional regression methods (which exhibited accuracies below 38 % on the PSQ-8 dataset) by a substantial margin. This proof-of-concept study establishes a scalable framework for multiplex MH genotyping via PSQ. Our method-development study offers substantial improvements over conventional workflows: superior base-calling accuracy, streamlined efficiency with automated dispensation ordering, and expanded utility for complex loci profiling.