NanoSNP

NanoSNP identifies single nucleotide polymorphisms (SNPs) from low-coverage Oxford Nanopore sequencing data to enable accurate variant detection and phasing in population-scale genomic studies.

Key Features:

• Deep Learning Approach: Employs Bi-long short-term memory (Bi-LSTM) deep learning networks to capture sequential dependencies for SNP detection on Nanopore reads.
• Haplotype-Aware Pipeline: Implements a multi-step, multi-scale haplotype-aware pipeline that divides low-coverage reads into distinct haplotypes for phasing SNP sites.
• Pileup Model: Utilizes naive pileup features to initially predict potential SNP sites.
• Haplotype Model: Extracts long-range haplotype features and short-range pileup features from each phased haplotype and combines them to improve genotype prediction accuracy.
• Performance Evaluation: Benchmarked against Clair, Clair3, Pepper-DeepVariant, and NanoCaller, showing superior SNP calling on ~16× Oxford Nanopore data, including difficult-to-map regions and the major histocompatibility complex (MHC).
• Cross-Genome Testing: Validated across six human genomes (HG002–HG007) to assess robustness across diverse genetic backgrounds.

Scientific Applications:

• Population-scale variant discovery: Enables SNP discovery and phasing in population-scale studies using low-coverage Oxford Nanopore sequencing.
• Disease association studies: Produces SNP genotypes from low-coverage data suitable for association analyses.
• Genetic diversity and evolutionary biology: Supports analyses of genetic diversity and evolutionary patterns from low-coverage Nanopore datasets.
• Analysis of challenging regions: Facilitates variant analysis in difficult-to-map genomic regions and the MHC region using low-coverage data.

Methodology:

Uses Bi-long short-term memory (Bi-LSTM) deep learning within a multi-step, multi-scale haplotype-aware pipeline that phases low-coverage reads; applies a naive pileup model to predict candidate SNPs and a haplotype model that extracts long-range haplotype and short-range pileup features from phased haplotypes which are combined for genotype prediction; performance was assessed by benchmarking against Clair, Clair3, Pepper-DeepVariant, and NanoCaller on ~16× Oxford Nanopore data across HG002–HG007 and in difficult-to-map and MHC regions.