PathME: pathway based multi-modal sparse autoencoders for clustering of patient-level multi-omics data.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Multi-omics data integration is crucial for understanding complex diseases like cancer.
• Identifying molecular patient subgroups can lead to personalized treatment strategies.
• Existing methods face challenges with high dimensionality and feature magnitude in multi-omics data.

Purpose of the Study:

• To develop a robust framework for unsupervised clustering of multi-omics data.
• To leverage pathway information for effective dimensionality reduction.
• To identify biologically plausible patient subgroups and their molecular drivers.

Main Methods:

• A multi-modal sparse denoising autoencoder framework was combined with sparse non-negative matrix factorization.
• Pathway information was utilized to reduce omics data dimensionality into patient-specific pathway scores.
• Machine learning techniques were employed to disentangle the impact of omics features on pathway scores.

Main Results:

• The proposed method successfully clustered patients across multiple cancer datasets (gene expression, miRNA, methylation, CNVs).
• Biologically plausible disease subtypes characterized by specific molecular features were identified.
• The approach demonstrated competitive clustering performance compared to existing methods and provided interpretable results.

Conclusions:

• The multi-modal sparse denoising autoencoder effectively integrates multi-omics data at the pathway level.
• The framework addresses the high-dimensional nature of omics data for robust patient subgroup identification.
• Patient-specific pathway scores enable reliable discovery of disease subgroups for personalized medicine.