SALAI-Net

SALAI-Net performs species-agnostic high-resolution local ancestry inference (LAI) along DNA sequences to estimate population labels from haplotype data for historical reconstruction and ancestry-aware genomic analyses.

Key Features:

• Species-agnostic application: Operates without species-specific biological parameters, relying solely on haplotype data to infer local ancestry across different species and ancestries.
• High-resolution LAI: Provides fine-scale local ancestry inference along DNA sequences for segment-level population-label assignment.
• Reference-matching informed by identity-by-descent: Employs a reference matching approach inspired by identity-by-descent techniques to estimate population labels for each DNA segment.
• Interpretability: Produces ancestry labels that are directly attributable to reference matches, supporting interpretable segment-level assignments.
• Performance: Demonstrated superior balanced accuracy in benchmark tests on whole-genome human data and generalization to dog breeds when trained on human data.
• Computational efficiency: Reported to run up to two orders of magnitude faster and to use considerably less RAM than competing methods in benchmarks.
• Benchmark datasets: Evaluated using whole-genome datasets including the 1000 Genomes Project, Simons Genome Diversity Project, HapMap, Human Genome Diversity Project, and Canid genomes.

Scientific Applications:

• Historical and population inference: Reconstruction of human history and migration patterns using segment-level ancestry assignments.
• Ancestry-adjusted GWAS: Use in genome-wide association studies that require local ancestry information for adjustment.
• Polygenic risk scores (PRSs): Integration of local ancestry information into PRS calculations for ancestry-aware risk estimation.
• Cross-species and breed analyses: Application to non-human datasets such as dog breeds for comparative and evolutionary studies.

Methodology:

Applies a reference-matching approach inspired by identity-by-descent on haplotype data to estimate population labels per DNA segment and was benchmarked on whole-genome datasets including the 1000 Genomes Project, Simons Genome Diversity Project, HapMap, Human Genome Diversity Project, and Canid genomes.