PySNV for complex intra-host variation detection

TL;DR

The paper introduces PySNV, a new software for detecting complex genetic variations within a host, which outperforms existing tools in accuracy and consistency.

Contribution

PySNV is a novel software specifically designed to detect complex intra-host variants with high accuracy, including long indels and low frequency mutations.

Findings

PySNV detected 1863 out of 1869 simulated variants with an F1-score of 0.99.

PySNV showed higher consistency in SARS-CoV-2 replicate data compared to LoFreq, reporting 21% more variants.

PySNV maintains computational speed comparable to other methods while improving detection of low frequency and long indel variants.

Abstract

Intra-host variants refer to genetic variations or mutations that occur within an individual host organism. These variants are typically studied in the context of viruses, bacteria, or other pathogens to understand the evolution of pathogens. Moreover, intra-host variants are also explored in the field of tumor biology and mitochondrial biology to characterize somatic mutations and inherited heteroplasmic mutations. Intra-host variants can involve long insertions, deletions, and combinations of different mutation types, which poses challenges in their identification. The performance of current methods in detecting of complex intra-host variants is unknown.

First, we simulated a dataset comprising 10 samples with 1869 intra-host variants involving various mutation patterns and benchmarked current variant detection software. The results indicated that though current software can detect most variants with F1-scores between 0.76 and 0.97, their performance in detecting long indels and low frequency variants was limited. Thus, we developed a new software, PySNV, for the detection of complex intra-host variations. On the simulated dataset, PySNV successfully detected 1863 variant cases (F1-score: 0.99) and exhibited the highest Pearson correlation coefficient (PCC: 0.99) to the ground truth in predicting variant frequencies. The results demonstrated that PySNV delivered promising performance even for long indels and low frequency variants, while maintaining computational speed comparable to other methods. Finally, we tested its performance on SARS-CoV-2 replicate sequencing data and found that it reported 21% more variants compared to LoFreq, the best-performing benchmarked software, while showing higher consistency (62% over 54%) within replicates. The discrepancies mostly exist in low-depth regions and low frequency variants.

https://github.com/bnuLyndon/PySNV/.

Figures