Grant: Discovery Grants Program - Individual

Cancer genomics identifies all genes playing critical roles in carcinogenesis. With that information, doctors can develop an individualized treatment plan for cancer patients and provide precision medicine. The state-of-the-art cancer genomics profiling characterizes many clinically and biologically relevant patterns that are not resolvable by morphology nor distinguishable under the microscope for cancer diagnosis. This research program will develop techniques for a novel recommender (Explainable Artificial Intelligence, XAI) for cancer genomics by providing explainable advice to guide clinicians in their decision-making process for cancer patient prognosis, diagnosis, and treatment options. DNA-methylation profiling has been established as a useful research tool for cancer classification. Several methylation-based classifiers were developed using supervised or unsupervised approaches for certain tumor types. However, the existing cancer classifiers cannot provide automatic explainable advice for clinicians, and they lack clarity and reproducibility in clinical settings due to some technical challenges. Feature extraction from genomic data has been subject to manual arbitrary cut-offs that lead to variabilities of clinical tests. The proposed research program will overcome those challenges to provide a reliable and reproducible recommender (XAI) system by improving classification accuracy and providing AI-based interpretable results. The specific objectives of the research program include the following: developing algorithms for genomic feature extraction automatically based on the deep learning techniques using DNA-methylation (Objective 1); building multiclass multi-level tumor type and subtype classifiers across different tumor entities (Objective 2); creating a novel recommender system (XAI) for patient risk stratification by providing AI-based interpretable results (Objective 3); developing an online software architecture that can be incorporated into a practical classification tool (Objective 4). This research program will bring a novel recommender (XAI) strategy to large patient groups and guide clinicians to make an Explainable patient-specific decision while tailoring the provided recommendations based on each patient's characteristics. This research program will also contribute to developing automated feature extraction and XAI strategies that can be generalized to other fields beyond cancer genomics to solve challenging complex problems with a large feature space. This research program is innovative as it will bring a novel recommendation strategy to clinicians to provide reasoning behind the prediction to increase trust and transparency. Several proposed innovations are potentially patentable and will facilitate technology transfer. This research program will help Canada to establish a strong leading position in XAI-based clinical research, generate more interdisciplinary XAI activities, and eventually reduce the economic burdens.