NetML

NetML applies network-based multi-task learning to identify shared and cancer-specific differentially expressed genes and molecular signatures for biomarker discovery and cancer outcome prediction across multiple cancer datasets.

Key Features:

• Multi-Task Learning Frameworks: Employs two frameworks—NetML and its variant NetSML—to jointly learn from multiple cancer types and discover shared and cancer-specific molecular signatures.
• Network-Based Approach: Models latent gene co-expression modules and gene-sample biclusters to capture complex biological interactions for biomarker selection.
• Cross-Cancer Knowledge Sharing: Shares information across different cancer types to enhance sample classification performance compared to single-cancer models.
• Validation on Large-Scale Datasets: Validated on simulated data and real high-throughput datasets from The Cancer Genome Atlas (TCGA), including ovarian, breast, and prostate cancer data.
• Biological Relevance: Detected common and specific molecular signatures are enriched for Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways and gene ontology terms relevant to cancer.

Scientific Applications:

• Biomarker Discovery: Identifies both universal and cancer-specific biomarkers across multiple cancer types.
• Cancer Outcome Prediction: Uses discovered molecular signatures to support prediction of cancer outcomes.
• Cross-Cancer Signature Analysis: Enables comparative analysis of shared tumorigenic mechanisms across ovarian, breast, and prostate cancers.
• Gene Expression Investigation: Facilitates exploration of gene expression changes that drive tumorigenic processes.
• Translational Insights: Supports research aimed at developing targeted therapies and improving diagnostic accuracy through identified signatures.

Methodology:

Processes input expression files (e.g., sample_breast_expression.pk, sample_ov_expression.pk) with Python scripts (NetTL_three_domains.py, NetSTL_three_domains.py) to execute network-based multi-task learning models that leverage latent gene co-expression modules and gene-sample biclusters and export results in text format.