EARN

EARN predicts driver genes in metastatic breast cancer by integrating Artificial Neural Networks, Random Forest, and non-linear Support Vector Machines to prioritize human protein-coding genes.

Key Features:

• Input data and source: Somatic mutation data from 450 metastatic breast tumor samples obtained from the cBio Cancer Genomics Portal.
• Ensemble learning: Combines Artificial Neural Networks (ANN), Random Forest (RF), and non-linear Support Vector Machines (SVM) for prediction.
• Feature extraction: Extracts features using four distinct software tools applied to the mutation data.
• Score aggregation decision strategy: Aggregates predicted scores from individual classifiers to prioritize candidate genes.
• Gene prioritization: Prioritizes human protein-coding genes annotated in the NCBI database.
• Gene set enrichment analysis: Performs gene set enrichment analysis to derive biological inferences from predicted drivers.
• Pathway enrichment analysis: Uses ReactomeFIVIz with a false discovery rate (FDR) threshold of < 0.03 to identify significant pathways and propose a gene set panel.
• Proposed gene panel: Identifies a novel MBCA gene set including HDAC3, ABAT, GRIN1, PLCB1, KPNA2, NCOR1, TBL1XR1, SIRT4, KRAS, CACNA1E, PRKCG, GPS2, SIN3A, ACTB, KDM6B, and PRMT1.
• Performance metrics: Reports Receiver Operating Characteristic Area Under the Curve (ROC-AUC) of 99.24% for MBCA and 99.79% for TCGA BRCA primary tumor samples.

Scientific Applications:

• Novel gene panel identification: Provides a candidate gene set for metastatic breast cancer diagnosis or study (genes listed above).
• Precision oncology and targeted panel design: Enables design of targeted genetic panels to reduce reliance on whole-genome or whole-exome sequencing.
• Functional interpretation: Supports biological interpretation of predicted driver genes through gene set and pathway enrichment analyses.

Methodology:

Analysis of somatic mutation data from 450 metastatic breast tumor samples from the cBio Cancer Genomics Portal; feature extraction using four distinct software tools; ensemble classification combining ANN, RF, and non-linear SVM with score aggregation to prioritize NCBI-annotated protein-coding genes; gene set enrichment analysis and pathway enrichment via ReactomeFIVIz (FDR < 0.03); statistical validation including ROC-AUC comparisons with TCGA BRCA primary tumor samples.