DeMoS

DeMoS detects dense co-expressed gene modules and derives prognostic gene signatures from cervical cancer gene expression data using Quasi-Clique detection and empirical statistical methods.

Key Features:

• Framework: Integrates Quasi-Clique detection algorithms with empirical statistical methods to identify dense co-expressed modules within gene expression data.
• Empirical Bayes Testing (Limma): Uses Limma Empirical Bayes testing to identify dysregulated genes and miRNAs from cervical cancer datasets.
• Quasi-Clique Detection: Applies Quasi-Clique algorithms to detect dense co-expressed modules evaluated by average correlation coefficients, selecting the highest correlated module as the gene signature.
• Gene and miRNA Identification: Identifies a core gene set (FGF9, FGF18, PPP1R9A, ERBB4, DCDC2, TOX3, ARMC3, DNALI1, RGL3, ENPP3) and dysregulated miRNAs such as hsa-mir-34c, with emphasis on miRNAs with high out-degree centrality.
• Classification and Prognosis: Evaluates predictive performance using Support Vector Machine (SVM), Partitioning Around Medoids (PAM), and Random Forest (RF) with 10-fold cross-validation and reports Area Under Curve (AUC) metrics.
• Survival Prognosis Study: Conducts Cox regression survival analysis to assess prognostic relevance and reports statistically significant p-values.

Scientific Applications:

• Prognostic biomarker discovery in cervical cancer: Identification and validation of gene signatures associated with prognosis in cervical cancer cohorts.
• Differential expression and miRNA–target analysis: Detection of dysregulated genes and miRNAs and assessment of miRNA targeting and network centrality (out-degree) within expression data.
• Classifier benchmarking and survival validation: Evaluation of classifier performance (SVM, PAM, RF) via 10-fold cross-validation and AUC, combined with Cox regression for survival prognosis studies.
• Application to RNA-seq profile analyses: Detection of precise gene signatures from RNA-seq expression profiles beyond cervical cancer datasets.

Methodology:

Uses Limma Empirical Bayes testing to identify dysregulated genes/miRNAs; applies Quasi-Clique detection to find dense co-expressed modules evaluated by average correlation coefficients (selecting the highest correlated module as the signature); performs miRNA targeting and out-degree centrality analysis; trains SVM, PAM, and Random Forest classifiers with 10-fold cross-validation and AUC assessment; and conducts Cox regression survival analysis.