DeepHapNet: a haplotype assembly method based on RetNet and deep spectral clustering

Area of Science:

• Genetics and Genomics
• Bioinformatics
• Computational Biology

Background:

• Gene polymorphism arises from single-nucleotide polymorphisms (SNPs), crucial for biogenetics.
• Haplotypes, sequences of SNP loci, contain more genetic information than individual SNPs.
• Haplotype assembly is vital for gene function understanding, disease diagnosis, and species gene identification.

Purpose of the Study:

• To introduce DeepHapNet, a novel deep learning method for accurate haplotype assembly.
• To leverage read pair correlations and clustering for improved haplotype reconstruction.

Main Methods:

• Utilized Retentive Network (RetNet) for multiscale feature extraction and learning global read relationships.
• Employed SpectralNet for read clustering based on learned features.
• Constructed haplotypes from identified read clusters.

Main Results:

• DeepHapNet demonstrated strong performance in haplotype assembly for both simulated and real datasets.
• The method proved effective for diploid and polyploid organisms.
• Successfully handled both long and short sequencing reads.

Conclusions:

• DeepHapNet offers a robust and versatile solution for haplotype assembly across various organisms and read types.
• The approach advances the field of biogenetics by improving the accuracy and efficiency of genetic information analysis.