OmicsFootPrint: a framework to integrate and interpret multi-omics data using circular images and deep neural...

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Multi-omics data analysis is crucial for understanding complex diseases.
• Existing methods often lack interpretability and efficiency.
• Advanced computational frameworks are needed to integrate and interpret diverse omics datasets.

Purpose of the Study:

• To introduce OmicsFootPrint, a novel framework for multi-omics data analysis.
• To enhance the interpretability of complex disease models using deep learning and explainability algorithms.
• To validate the framework's performance in cancer subtype classification and drug response prediction.

Main Methods:

• Transformation of multi-omics data into intuitive 2D circular images.
• Application of deep neural networks for data analysis.
• Integration of the SHapley Additive exPlanations (SHAP) algorithm for model interpretability.

Main Results:

• High accuracy in classifying lung cancer subtypes (AUC 0.98 ± 0.02) and breast cancer subtypes (AUC 0.83 ± 0.07).
• Successful differentiation between invasive lobular and ductal breast carcinomas.
• Effective prediction of drug responses in cancer cell lines (median AUC 0.74), outperforming existing methods.
• Robust performance maintained even with reduced training data.

Conclusions:

• OmicsFootPrint provides an efficient and interpretable approach to multi-omics data analysis.
• The framework significantly advances the understanding of complex disease mechanisms.
• OmicsFootPrint represents a valuable tool for cancer research and personalized medicine.