PIMO: pathway-based interpretable multiomics interactions for multiomics integration.

Area of Science:

• Computational biology
• Genomics
• Systems biology

Background:

• Modeling interomics interactions is crucial for understanding complex diseases.
• Epigenomic and structural alterations like DNA methylation and copy number alterations (CNAs) impact gene expression, disease progression, and survival.
• Current deep learning multiomics methods often overlook gene-level interactions due to complexity, limiting mechanistic insights.

Purpose of the Study:

• To propose PIMO, a pathway-based interpretable deep learning model for multiomics interaction analysis.
• To explicitly capture regulatory effects across different omics layers.
• To improve interpretability and mechanistic understanding of complex diseases through interomics modeling.

Main Methods:

• Developed a pathway-based interpretable deep learning multiomics interaction model (PIMO).
• The model explicitly captures regulatory effects across omics layers.
• Utilized multiple TCGA cancer datasets for model evaluation.

Main Results:

• PIMO consistently outperformed state-of-the-art baselines in cancer survival analysis.
• Achieved up to a 13% increase in the C-index compared to existing methods.
• Provided biologically interpretable analyses, identifying key pathways, genes, and interomics interactions involving DNA methylation and CNAs.

Conclusions:

• PIMO offers a novel approach to modeling interomics interactions for improved cancer survival prediction.
• The model's interpretability facilitates the identification of key biological drivers in complex diseases.
• PIMO demonstrates the potential of deep learning in uncovering gene-level interactions across multiple omics layers.