PrognosiT

PrognosiT predicts tumor volume from gene expression data while extracting pathway- and gene-set-level molecular insights to address analysis of high-dimensional, highly correlated genomic datasets.

Key Features:

• Multiple Kernel Learning Algorithm: Employs multiple kernel learning techniques to enhance predictive accuracy.
• Pathway/Gene Set Integration: Incorporates prior knowledge from pathway and gene sets directly into the learning process to identify relevant molecular mechanisms.
• Efficient Feature Utilization: Utilizes fewer gene expression features—reported as less than one-tenth—compared with Random Forest (RF) and Support Vector Regression (SVR).
• Dynamic Prior Incorporation: Integrates pathway and gene-set information during model training rather than selecting a subset of genomic features beforehand.
• Differential Regulation Identification: Identifies up- and down-regulated genes in tumor and normal tissues to aid biomarker discovery.

Scientific Applications:

• Predictive Performance: Predicts tumor volume in thyroid carcinoma patients with performance comparable to or exceeding RF and SVR in reported studies.
• Molecular Mechanism Insights: Extracts pathway- and gene-set-level signals to reveal mechanisms underlying cancer progression and aggressiveness and to support biomarker discovery.

Methodology:

Uses a multiple kernel learning framework that directly integrates pathway and gene-set priors into the prediction process rather than preselecting genomic features, enabling identification of up- and down-regulated genes in tumor and normal tissues and achieving feature usage below one-tenth of that required by RF and SVR.