Jumping across biomedical contexts using compressive data fusion.

Area of Science:

• Bioinformatics
• Computational Biology
• Systems Biology

Background:

• Biological data is rapidly growing, encompassing genes, chemicals, diseases, and pathways.
• Interactions between biological objects have diverse semantic meanings, often overlooked by traditional methods.
• Existing approaches fail to fully leverage the semantic richness of heterogeneous biological datasets.

Purpose of the Study:

• To introduce Medusa, a novel approach for detecting significant size-k modules of biological objects.
• To explicitly distinguish and utilize diverse data semantics in biological data modeling.
• To advance the analysis of complex biological systems by incorporating semantic information.

Main Methods:

• Medusa employs collective matrix factorization to derive distinct data semantics.
• Module detection is formulated as a submodular optimization program.
• The approach allows flexible selection and combination of semantic meanings with theoretical quality guarantees.

Main Results:

• Medusa effectively associates genes with diseases and identifies disease modules across 310 complex diseases.
• The method shows superior performance in predicting gene-disease associations compared to methods ignoring semantic diversity.
• Combining different semantics generally leads to more accurate disease module recovery.

Conclusions:

• Medusa offers a powerful and flexible framework for analyzing heterogeneous biological data by leveraging semantic information.
• The approach demonstrates significant improvements in gene-disease association and disease module detection.
• The utility of specific semantics is disease-dependent, highlighting the benefit of combining multiple semantic interpretations.