EMLI-ICC

EMLI-ICC integrates genome-wide proteome and transcriptome data with an ensemble machine-learning framework to predict tumor metastasis and stratify prognostic risk in intrahepatic cholangiocarcinoma (ICC).

Key Features:

• Integration of Multi-Omics Data: Integrates 216 proteome and 244 transcriptome profiles to develop predictive models for ICC metastasis.
• Feature Selection and Biomarker Identification: Selects 132 feature genes as biomarkers that form the basis of the EMLI-Metastasis classifier.
• Ensemble Machine Learning Approach: Implements a weighted ensemble based on the k-Top Scoring Pairs (k-TSP) method that generates multiple binary expression rank-based classifiers via bootstrap aggregating.
• Weighted Voting Mechanism: Aggregates predictions from ten distinct binary metastasis classifiers using a weighted voting scheme and reports performance of 97.1% accuracy on proteome and 85.0% accuracy on transcriptome datasets.
• Risk Stratification Model: Includes the EMLI-Prognosis algorithm using 21 gene-pair signatures to stratify patients into high-risk and low-risk groups with significant differences in overall survival (P-value < 0.05).

Scientific Applications:

• Metastasis Prediction: Predicts metastatic potential of ICC tumors using integrated proteome and transcriptome classifiers.
• Prognostic Risk Stratification: Stratifies ICC patients into prognostic high-risk and low-risk groups using 21 gene-pair signatures (EMLI-Prognosis).
• Multi-cohort Machine-Learning Diagnostics: Demonstrates a machine-learning-based multi-cohort diagnostic approach applicable to diverse cancer datasets.

Methodology:

Integrates 216 proteome and 244 transcriptome profiles, selects 132 feature genes, applies bootstrap aggregating to generate multiple k-TSP binary expression rank-based classifiers, combines ten classifiers with weighted voting for EMLI-Metastasis, and uses EMLI-Prognosis with 21 gene-pair signatures for risk stratification.