Network-based multi-task learning models for biomarker selection and cancer outcome prediction.

Area of Science:

• Computational biology
• Genomics
• Cancer research

Background:

• Understanding cancer progression requires analyzing gene expression and transcriptome changes.
• Previous studies focused on single cancer types, overlooking universal or specific molecular mechanisms.
• Identifying commonalities and differences across tumor types is crucial for comprehensive cancer research.

Purpose of the Study:

• To develop network-based multi-task learning frameworks (NetML and NetSML) for discovering shared and specific differentially expressed genes across diverse cancer types.
• To address the limitations of single-cancer analyses by leveraging knowledge across multiple tumor datasets.
• To improve the understanding of molecular mechanisms driving carcinogenesis and cancer progression.

Main Methods:

• Introduced two network-based multi-task learning frameworks: NetML and NetSML.
• Utilized common latent gene co-expression modules and gene-sample biclusters across multiple cancer datasets.
• Applied frameworks to The Cancer Genome Atlas (TCGA) ovarian, breast, and prostate cancer datasets.

Main Results:

• NetML and NetSML demonstrated superior sample classification performance compared to models without cross-cancer knowledge sharing.
• Identified common and cancer-specific molecular signatures that correlate with known cancer marker genes.
• Detected signatures enriched in cancer-relevant Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways and Gene Ontology (GO) terms.

Conclusions:

• Network-based multi-task learning effectively identifies shared and specific gene expression patterns across different cancer types.
• The developed frameworks enhance cancer classification and provide insights into cancer-specific and universal tumorigenic mechanisms.
• Findings contribute to a deeper understanding of cancer biology and the discovery of novel biomarkers.