OmicsFootPrint is a novel deep-learning framework that transforms multi-omics data into two-dimensional circular images and applies deep neural networks (DNN) for sample classification. The framework also integrates the SHapley Additive exPlanations (SHAP) algorithm to facilitate model interpretation. We applied OmicsFootPrint to The Cancer Genome Atlas (TCGA) and cancer cell lines datasets, achieving high accuracy in two-class and multi-class classification for lung and breast cancer subtypes. It also accurately predicted drug response phenotypes in cancer cell lines. Benchmarking against nine popular methods, including traditional and contemporary multi-omics algorithms, demonstrated OmicsFootPrint's superior performance. The transformation of multi-omics data into compact circular images significantly reduces memory requirements. OmicsFootPrint supports explainable AI models, advancing multi-omics data analysis, enhancing our understanding of disease mechanisms, and improving drug response predictions. This approach can significantly impact drug discovery and development, offering a powerful tool for precision medicine.
Learning Objectives:
Upon completion, participants will be able to understand the application of DNN models for multi-omics data integration and model interpretation.
Understand the role of OmicsFootPrint in integrating multi-omics data for cancer research.
Learn how deep learning models, such as EfficientNetV2, are applied to classify cancer subtypes using multi-omics data.
Explore how SHAP analysis is used to interpret model predictions and identify key genomic features.
Review the benchmarking results of OmicsFootPrint against other state-of-the-art multi-omics data integration methods.
