Leveraging machine learning for classifying the pathogenicity of genetic variants in cancer genomics and precision medicine

Ashkan Bigdeli

doi:10.17918/00011034

The classification of genetic variants plays a crucial role in advancing precision medicine, particularly in oncology, where accurate variant interpretation can directly impact patient care. However, existing methods for classifying the pathogenicity of genetic variants remain largely manual, resource-intensive, and prone to inconsistencies across institutions. This thesis explores the integration of machine learning into cancer genomics, focusing on the development and application of the Azurify Model, a novel machine learning-based classifier for somatic variant interpretation. We first establish a strong foundation in cancer genomics, detailing the genetic and molecular mechanisms underlying cancer development and progression. Key bioinformatics tools and computational pipelines are reviewed, emphasizing their role in variant detection, annotation, and classification. We then present Azurify, a machine learning model trained on a diverse set of cancer sequencing data. Azurify is designed to classify genetic variants with high accuracy, independent of sequencing platforms and clinical assay variations. Performance assessments demonstrate that Azurify not only surpasses existing variant classification methods but also generalizes well across multiple independent datasets. The model successfully identifies pathogenic somatic mutations in leukemia and lymphoma subtypes, reinforcing its potential clinical utility. Further, the study examines the genetic determinants influencing the efficacy of CAR-T cell therapy, a promising immunotherapeutic approach for blood cancers. By leveraging sequencing data from CAR-T-treated lymphoma patients and the Azurify model, key genetic predictors of treatment response and toxicity are identified. The findings highlight the potential of integrating machine learning-based genomic analysis with clinical decision-making to effectively research personalized cancer treatment strategies. We also investigate the broader implications of machine learning in clinical oncology and variant classification. Expanding Azurify's capabilities, incorporating real-time variant databases, and improving deployment can all lead to a more robust model with increased utility in both research and clinical environments. Indeed, the integration of machine learning into genomic medicine represents a transformative step toward more efficient, accurate, and accessible cancer variant classification that can be applied for more effective research discovery regarding treatment planning.

Leveraging machine learning for classifying the pathogenicity of genetic variants in cancer genomics and precision medicine

Files and links (1)

Abstract

Metrics

Details

Leveraging machine learning for classifying the pathogenicity of genetic variants in cancer genomics and precision medicine

Files and links (1)

Abstract

Metrics

Details

Drexel University Social media