A comprehensive and rigid computational framework to construct a patient-level personalized prognostic signature .
This repository provides the code for the paper "A mechanistically-derived patient-level immune prognostic signature in high-grade serous ovarian cancer".
To acquire comprehensive, precise and explicable prognostic genes, univariate Cox regression analysis was performed to assess prognostic value of each gene in each dataset. Then, meta-analysis was conducted to integrate gene’s HR value from multiple dataset and evaluate their overall impact on prognosis. We assessed the heterogeneity using the Q-test and chose to conduct a random-effects model when P < 0.05, otherwise a fixed-effects model was implemented. Next, multiple testing correction (Benjamini-Hochberg, BH) was performed, and candidate survival-related markers were identified using an adjusted P value threshold of 0.01. Finally, to acquire comprehensive, precise and explicable prognostic biomarkers, functional enrichment analysis for candidate prognostic genes was performed using R package ‘clusterProfiler’. Candidate prognostic genes involved in significantly enriched biological processes were retained as biologically plausible prognostic genes.
To construct a patient-level prognostic signature in a personalized manner without the need to normalize data from various sources, a modified single sample gene set enrichment analysis (ssGSEA) was used to constructed an patient-level personalized prognostic signature (PLPPS) based on gene expression levels of individual samples. The difference of overall survival between subgroups was assessed using Kaplan-Meier estimates and the statistical significance was assessed by log-rank test. Uni- and multi-variate Cox regression analysis was performed to measure the correlation between various factors and overall survival. Hazard ratios (HR) and 95% confidence intervals (CI) were calculated to assess the factor’s impact on prognosis. Comparative analysis of performance from diverse signatures was conducted according to HR and Harrell's concordance index (C-index) statistic.