Introduction: Immune Checkpoint Inhibitors (ICIs) have revolutionized cancer therapy, offering clinical benefits across multiple cancer types. However, patient responses remain highly variable, underscoring the urgent need for robust biomarkers to characterize immunotherapy outcomes. Methods: To capture variations in immunotherapy responses, we developed a pathway-masked approach to construct a Phenotype-Specific Network (PSN) for biomarker discovery. The method consists of two key steps: pathway ranking and selection. First, we applied an improved strategy, Concordance Enrichment Analysis (CEA), to rank Reactome pathways, benchmarking its performance via network modularity and gene set coverage. Next, gene set distance and pathway semantic similarity were integrated to define thresholds for selecting phenotype-specific pathways. Genes from these pathways were used as a mask to extract a PSN, represented as a subnetwork of Protein- Protein Interactions (PPIs). Topological analysis and network propagation were then performed to identify critical molecular determinants, which, along with their neighbors, were developed into informative gene signatures. Results: The proposed CEA outperforms established pathway ranking methods. Two pathways, selected based on gene-set distance and pathway semantic similarity, were used to construct a PSN comprising 148 genes. Within this network, ZAP70 emerged as a gene with high betweenness centrality, associated with multiple T-cell receptor modules. Heat diffusion analysis centered on PD1 and CTLA4 consistently highlighted ICOSLG, a key ligand in the ICOS–ICOSLG axis, which is involved in T-cell costimulation. ZAP70 and ICOS, together with their neighboring genes, were used to construct gene signatures. These signatures showed moderate predictive performance for immunotherapy response, with the ICOS-related signature notably outperforming others in predicting progression- free survival. Discussion: This study highlights a pathway-masked PSN strategy that integrates transcriptomic data with biological networks to capture phenotype-specific insights into the immunotherapy response. By combining pathway semantics and network topology, this approach provides a conceptually novel framework for biomarker discovery. Taken together, this strategy contributes to a systems-level perspective with strong translational relevance, offering new opportunities to inform clinically meaningful applications. Conclusion: This pathway-masked framework enables the identification of the phenotype-specific network and key molecular regulators associated with immunotherapy response variations. Based on these findings, we have derived gene signatures that hold potential for improving clinical decisionmaking in immunotherapy.
Ding et al. (Thu,) studied this question.