CheckDyn: a multi-cohort computational framework for profiling treatment-induced immune checkpoint dynamics and predicting adaptive resistance to immune checkpoint blockade.

Adaptive resistance limits durable benefit from immune checkpoint blockade (ICB) in the majority of cancer patients, yet the transcriptomic dynamics of the broader checkpoint landscape during treatment remain poorly characterized across tumor types. Here we present CheckDyn, a multi-cohort computational framework that profiles paired pre- and post-treatment transcriptomes to quantify treatment-induced changes across 38 immune checkpoint and exhaustion-associated genes and to predict adaptive resistance. We integrated publicly available RNA-seq and scRNA-seq data from 64 paired tumor samples spanning melanoma, basal cell carcinoma, and non-small-cell lung cancer (GSE91061, GSE120575, GSE123813, GSE176021), applying pseudo-bulk aggregation, Z-score batch correction, and Stouffer meta-analysis for cross-cohort harmonization. Paired Wilcoxon signed-rank testing and linear mixed-effects meta-analysis identified LAG3 (log2FC = 0.596, padj = 0.015), PDCD1 (log2FC = 0.810, padj = 0.003), TOX2 (log2FC = 0.605, padj = 0.003), CD274 (log2FC = 0.402, padj = 0.015), and IDO1 (log2FC = 0.381, padj = 0.026) as consistently upregulated post-treatment across cohorts. Co-expression network analysis revealed extensive rewiring, with ENTPD1 (ΔDegree = +0.297) emerging as the largest hub-degree shift, suggesting a shift toward metabolic immune suppression. Temporal trajectory modeling showed that all 38 checkpoint genes followed linear upregulation trajectories, with PDCD1 and LAG3 carrying the steepest slopes. An ensemble classifier combining logistic regression and random forest on pre-to-post expression deltas achieved an area under the receiver operating characteristic curve (AUC) of 0.812 (95% CI: 0.694-0.930; 5-fold cross-validated AUC = 0.806) for adaptive resistance prediction across n = 59 patients with available response annotations. These findings establish a consistent transcriptional signature of compensatory checkpoint upregulation during ICB therapy and provide a data-driven framework for early identification of adaptive resistance that warrants external prospective validation.