The landscape of molecular and functional patient information has been rapidly expanding across all fields of medicine. There has been increasing focus on incorporating individualized patient attributes into clinical care. Multiomics data are now available, including genomics, transcriptomics, proteomics, metabolomics, and others. New bioinformatics methods for analyzing the vast and complicated wealth of information are needed. Machine Learning (ML), a component of artificial intelligence, refers to the methodology by which computer systems can analyze and interpret complex datasets in a manner that will predict outcomes or correlate with disease status. These algorithms have the potential to provide critical prognostic data, optimize precision medicine approaches, and improve accuracy and efficiency of pathologic diagnosis.
Dr. Kun-Hsing Yu will discuss how ML can integrate data generated by various modalities to enhance oncology research and practice. Recent progress in digitized data acquisition, data-driven algorithms, and computing infrastructure, have empowered ML applications for cancer subtype identification and survival outcome prediction. In this talk, he will outline recent breakthroughs in ML technologies and their applications in data integration, highlight the advances in quantitative pathology analyses, and identify the challenges for further progress in ML systems.
Dr. Pamela Becker will discuss an ML algorithm called MERGE based on genomics data to correlate gene expression with drug sensitivity in patients with acute myeloid leukemia (AML). Using MERGE, ~40 genes were identified for which high expression was correlated with sensitivity or resistance to classes of drugs in AML. Dr. Becker is also working with colleagues on deep learning methods to track the morphology, phenotype, and viability of AML cells in real time by video microscopy after drug exposure. These predictive algorithms, in combination with molecular data and functional screening, can be combined to optimize treatment for individual patients.
Dr. Lee Cooper will discuss how ML applied to digital pathology imaging enables quantitative and scalable evaluation of histology samples. In hematologic malignancies, this technology has the potential to improve diagnostic reproducibility and to optimize diagnostic criteria. His talk will discuss the development of ML algorithms for bone marrow aspirate smears and diffuse large B-cell lymphomas. In addition, he will illustrate challenges and opportunities in applying ML and address the critical role of data collection and validation in system development.