Machine Learning Techniques for Classifying the Mutagenic Origins of Point Mutations.

Area of Science:

• Genetics
• Bioinformatics
• Computational Biology

Background:

• Distinguishing mutagenic mechanisms is crucial for cancer research and population genetics.
• Current methods often assume distinct sequence context relationships for different mutation types, but evidence is limited.
• Identifying the origin of point mutations is challenging due to overlapping mutation spectra.

Purpose of the Study:

• To evaluate if sequence context alone can resolve the mechanistic origin of point mutations.
• To differentiate between spontaneous mutations in the mouse germline and N-ethyl-N-nitrosourea (ENU)-induced mutations.
• To develop a machine learning tool for classifying mutation mechanisms.

Main Methods:

• Contrasted single nucleotide variants from spontaneous and ENU-induced mutagenesis in mice.
• Employed a novel log-linear modeling approach to analyze sequence context.
• Utilized a logistic regression classifier to discriminate between mutation classes.

Main Results:

• Neighboring bases contain significant information about point mutation direction, differing between ENU-induced and spontaneous variants.
• The logistic regression classifier achieved high performance in discriminating between mutation classes.
• Classifier features align with information content analyses, suggesting generalizability.

Conclusions:

• Machine learning, using sequence context, can effectively classify individual genetic variant mutation mechanisms.
• This approach offers a practical tool for applications in cancer genomics and population mutagenesis.
• The developed software is available as an open-source tool.