Utilization of Machine Learning Algorithms to Classify Copy Number Events and Predict Loss of Heterozygosity in Breast Cancer Patients

Abstract

Copy number events, specifically copy number aberrations, are occurrences that play an important part in the development of certain cancer types. However, elucidating the correct type of copy number aberration has always been a debated topic with the use of both in silico and in vitro techniques having their own disadvantages. Machine learning may prove to be an excellent avenue to explore as recent advancements have made it easier to build, study, and apply in the medical field. With it, determining a gene's specific copy number type may be possible to elucidate thus allowing better understanding for possible target therapy. Using pre-established and validated software, elucidated copy number segmentation value was inputted for a regiment of machine learning algorithm. The top five models were selected for hyperparameter tuning with cross validation with the end goal of Ensembl voting while genomics data was visualized to ensure better clarity for data interpretation. Analysis resulted in a clear pipeline for copy number analysis for future data entry to increase the trustworthiness of the model. However, future studies can look into the use of next generation sequencing data, which can offer more coverage of the genome at the cost of higher computational burden.