Classification of Cancer Primary Sites Using Machine Learning and Somatic Mutations

Area of Science:

• Oncology
• Bioinformatics
• Computational Biology

Background:

• Accurate human cancer classification, including primary site, is crucial for understanding and effective treatment.
• Big data in somatic mutations offers opportunities for machine learning-based cancer classification.

Purpose of the Study:

• To investigate cancer classification using machine learning and somatic mutation data.
• To explore the utility of gene features, somatic mutations, and chromosomal information for predicting primary tumor sites.

Main Methods:

• Utilized support vector machine (SVM) for multiclass classification.
• Analyzed 1,760,846 somatic mutations from 230,255 cancer patients.
• Used gene symbol, somatic mutation, chromosome, and gene functional pathway as predictors for 6,751 subjects across 17 tumor sites.

Main Results:

• Baseline accuracy using only gene features was 0.57, improving to 0.62 with mutation and chromosome data.
• Five primary sites (large intestine, liver, skin, pancreas, lung) achieved >0.70 F-measure.
• The large intestine model showed the highest performance with an 0.87 F-measure.

Conclusions:

• Somatic mutation data is valuable for predicting primary tumor sites using machine learning.
• This study pioneers the classification of primary tumor sites using machine learning and somatic mutation data.
• Findings highlight the potential of integrating genomic data with machine learning for cancer diagnostics.