MAC-ErrorReads: machine learning-assisted classifier for filtering erroneous NGS reads

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Next-generation sequencing (NGS) generates vast biological data, but increased speed and affordability lead to more errors.
• Data quality assurance is shifting from hardware to software preprocessing stages.
• Effective error detection and filtration are crucial for reliable NGS data analysis.

Purpose of the Study:

• To introduce MAC-ErrorReads, a novel Machine learning-Assisted Classifier for filtering erroneous NGS reads.
• To evaluate the performance of machine learning algorithms in identifying and removing low-quality reads.
• To enhance the accuracy and reliability of genomic data.

Main Methods:

• Developed MAC-ErrorReads, a binary classification model using five supervised machine learning algorithms.
• Extracted features using Term Frequency-Inverse Document Frequency (TF_IDF) from diverse biological datasets.
• Trained and tested models on datasets including E. coli, S. aureus, H. Chr14, Arabidopsis thaliana Chr1, and M. zebra.

Main Results:

• Naive Bayes (NB) within MAC-ErrorReads demonstrated robust performance across datasets with high accuracy, precision, recall, and F1-scores.
• MAC-ErrorReads NB achieved a 38.69% alignment rate for S. aureus reads, outperforming many error correction tools.
• MAC-ErrorReads showed competitive alignment rates (e.g., >99% for H. Chr14, ~90% for A. thaliana Chr1, ~83% for M. zebra) compared to other tools.

Conclusions:

• Machine learning effectively filters NGS reads, significantly improving assembly quality and genomic coverage.
• Integrating artificial intelligence with genomics enhances NGS data quality and downstream analysis.
• This approach holds promise for advancing genetics, genomics, and personalized medicine research.