Using generative adversarial networks for genome variant calling from low depth ONT sequencing data.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Genome variant calling is crucial for genetic studies but typically requires high-depth sequencing.
• Existing methods show reduced performance on low-depth data, limiting accessibility and increasing costs.
• Low-depth sequencing presents a significant challenge for accurate variant identification.

Purpose of the Study:

• To develop a novel method for accurate genome variant calling using low-depth sequencing data.
• To reduce the cost and increase the accessibility of genome-wide sequencing examinations.
• To demonstrate the applicability of the developed method across different species.

Main Methods:

• Trained a generative adversarial network (GAN) using public Oxford Nanopore (ONT) data from the Genome in a Bottle (GIAB) Consortium.
• Developed LDV-Caller, a method designed to project high-depth sequencing information from low-depth data.
• Validated the LDV-Caller model on 157 public Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) samples.

Main Results:

• LDV-Caller achieved a 94.25% F1 score on low-depth human data, comparable to state-of-the-art methods using double the depth.
• The method demonstrated strong performance on SARS-CoV-2 data, yielding a 92.77% F1 score at only 22x sequencing depth.
• The results indicate that LDV-Caller can effectively analyze different species with low-depth sequencing data.

Conclusions:

• LDV-Caller offers a cost-effective solution for accurate genome variant calling from low-depth sequencing data.
• The method shows promise for broader applications in genomics and pathogen surveillance.
• This advancement can significantly lower the financial barrier for genome-wide sequencing studies.