Exploring the Consistency of the Quality Scores with Machine Learning for Next-Generation Sequencing Experiments

Area of Science:

• Genomics
• Bioinformatics
• Machine Learning

Background:

• Next-generation sequencing (NGS) offers cost-effective, high-throughput data generation.
• Assessing the reliability of variant calls in NGS data is crucial for accurate downstream analysis.
• Current methods lack pre-evaluation of variant quality scores.

Purpose of the Study:

• To develop and evaluate a machine learning approach for predicting variant call quality scores.
• To identify informative annotations for predicting variant quality.
• To enhance the reliability assessment of variant calls in NGS data.

Main Methods:

• A machine learning model was developed using Random Forest Regressor (RFR), Multiple Linear Regression (MLR), and Nearest Neighbor Regression (NNR).
• The model utilized GATK Annotation Modules as features to predict variant quality scores.
• Performance was evaluated using simulated and real human genome sequencing data with high coverage (30x).

Main Results:

• RFR demonstrated superior performance in predicting variant quality scores compared to MLR and NNR.
• High predictability (R2 up to 96.7% in simulated, 97.8% in real data) was achieved using informative features.
• The developed models showed consistent robustness across both simulated and real-world genomic datasets.

Conclusions:

• Machine learning, particularly RFR, effectively predicts variant call quality scores from NGS data.
• The approach enhances the reliability of variant calling, facilitating more accurate genomic analyses.
• This method provides a valuable tool for pre-evaluating variant quality in bioinformatics pipelines.