Using deep learning to identify recent positive selection in malaria parasite sequence data.

Area of Science:

• Genomics
• Computational Biology
• Parasitology

Background:

• Malaria, caused by Plasmodium parasites, presents a significant global health challenge.
• Understanding malaria pathogenesis and drug resistance requires timely detection of genetic mutations.
• Whole-genome sequencing (WGS) generates vast amounts of Plasmodium DNA data.

Purpose of the Study:

• To evaluate the potential of deep learning models for detecting genetic loci under recent positive selection in malaria parasites.
• To develop and validate a deep learning approach for identifying signals of drug resistance.

Main Methods:

• Developed DeepSweep, a deep learning-based approach trained on genetic regions with known sweeps.
• Applied DeepSweep to whole-genome sequencing data from Plasmodium falciparum and Plasmodium vivax.
• Compared DeepSweep's results with established extended haplotype homozygosity methods (iHS and Rsb).

Main Results:

• DeepSweep demonstrated high predictive accuracy (AUC > 0.95) in detecting recent genetic sweeps using simulated data.
• Identified known drug resistance loci in P. falciparum (pfcrt, pfdhps, pfmdr1) and P. vivax (pvmrp1).
• Achieved significant overlap (60-75%) with established methods for detecting positive selection.

Conclusions:

• The deep learning approach effectively detects positive selection signatures in malaria parasite WGS data.
• DeepSweep is generalizable and can be trained to identify other types of genetic selection.
• This machine learning tool can enhance parasite genome-based surveillance and support malaria control decision-making.