Deep Learning Enhances Variant Calling Accuracy in Genomic Data

DOI

10.2991/978-94-6239-674-6_33How to use a DOI?

Keywords

Deep learning; Variant calling; Genomic data

Abstract

The accuracy of variant calling in next-generation sequencing (NGS) is critical for genomic research and clinical applications. Traditional variant callers, using diverse methodologies such as haplotype- based, position-based, and pattern growth approaches, often produce discordant results due to their inherent design and statistical methods. This study investigates the use of deep learning to integrate and optimize information from multiple variant callers. We developed a deep learning neural network designed to improve variant calling accuracy by leveraging features derived from base-specific information, sequencing biases, and quality metrics. The network was optimized through careful tuning of hyperparameters, including layer count, optimizer choice, learning rate, and sample balancing techniques. The final architecture included eight layers, utilized the Adam optimizer with a learning rate of 10 minus 5, and employed SMOTE for sample balancing. Benchmarking against both simulated and real datasets demonstrated that the neural network significantly outperforms traditional and concordance-based variant callers. In simulated datasets, the neural network achieved an F1 score of 0.980, surpassing the best single variant caller (0.888) and concordance-based caller (0.927). On a real genomic dataset (NA12878 Genome in a Bottle), the network outperformed existing methods with a precision of 0.859 and a recall of 0.911, leading to a notable reduction in false positives while maintaining high sensitivity. These results validate the efficacy of deep learning in variant calling and highlight its potential to improve precision and recall in genomic analyses, providing a robust tool for both research and clinical genomics.

Copyright

© 2026 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - Bhargava Rathod
PY  - 2026
DA  - 2026/05/28
TI  - Deep Learning Enhances Variant Calling Accuracy in Genomic Data
BT  - Proceedings of the International Conference on Sustainable Computing and Artificial Intelligence (ICSCAI 2025)
PB  - Atlantis Press
SP  - 391
EP  - 404
SN  - 2352-5401
UR  - https://doi.org/10.2991/978-94-6239-674-6_33
DO  - 10.2991/978-94-6239-674-6_33
ID  - Rathod2026
ER  -