ECMarker

ECMarker predicts gene expression biomarkers and reveals gene regulatory mechanisms associated with disease phenotypes using interpretable machine learning.

Key Features:

• Interpretable and Scalable Model: Integrates semi- and discriminative-restricted Boltzmann machines with lateral connections at the input gene layer to enable biological interpretability and prioritize genes without prior feature selection.
• Predictive Accuracy: Predicts disease stages and biomarker genes with demonstrated high accuracy in non-small-cell lung cancer (NSCLC) datasets.
• Revealing Regulatory Networks: Models lateral connections at the input layer to infer potential gene networks and regulatory mechanisms implicated in disease progression.
• Clinical Interpretability: Identified biomarker genes predict survival rates of early lung cancer patients with statistical significance (P-value < 0.005).
• Drug Repurposing Potential: Links biomarker signatures to existing drugs, identifying candidates used for late-stage or other cancers that may impact early-stage lung cancer biomarkers.

Scientific Applications:

• Oncology research: Decoding molecular underpinnings of cancer progression and discovering early-stage biomarkers, demonstrated on NSCLC data.
• Prognosis and therapeutic hypothesis generation: Informing prognosis and suggesting drug repurposing candidates for early-stage lung cancer based on biomarker and network analyses.

Methodology:

Integrates semi- and discriminative-restricted Boltzmann machines with lateral connections at the input gene layer to support classification tasks, gene prioritization, and inference of input-layer gene connections without requiring prior feature selection.