UC-133 Biomedical Deep Learning - A staged approach using trustworthy deep learning for multi-omics data classification

UC-133 Biomedical Deep Learning - A staged approach using trustworthy deep learning for multi-omics data classification

https://www.kennesaw.edu/ccse/events/computing-showcase/fa24-cday-program.php

Genetic data such as mRNA, miRNA, and DNA methylation offer precious insights into the underlying causes variant diseases. These types of data provide various layers of information, simultaneously enhancing our understanding of the disease and improving diagnostic accuracy. Combining mRNA, miRNA, and DNA methylation data allows for a multi-dimensional approach to identifying biomarkers, potentially leading to earlier and more accurate diagnosis. However, integrating all modalities is not practical. The clinical cost increases significantly with every modality incorporated. In contrast to previous methods, our model uses partial modalities when possible. We will use subjective logic and trustworthy deep learning under the staged approach to perform disease risk prediction. During our research process, we explored effective modality combinations for single view and bi-view models, designed an optimized multi-perception layer architecture for single-view classification, and implemented methods to quantify and optimize uncertainty in incomplete multi-omics data integration.