UCResponNet-X: cross-platform multi-dataset gene expression for predictive modeling of drug response in ulcerative colitis

Predictive modeling of biologic drug response using transcriptomic data is challenged by strong platform-specific effects between microarray and RNA-sequencing technologies. In this study, we propose UCResponNet-X, a computational framework designed to evaluate and improve cross-platform generalizability of machine-learning models for predicting infliximab response in ulcerative colitis. The framework integrates three independent microarray cohorts for training and validation and assesses model transferability on an external RNA-seq dataset. We systematically compare log2 transformation, quantile normalization, and z-score standardization in combination with batch-effect correction and biologically informed feature selection. Multiple classification algorithms are evaluated under a unified cross-validation protocol. Our results demonstrate that z-score and log2 normalization substantially outperform quantile normalization in preserving predictive signal across platforms, achieving mean cross-validation AUC values up to 0.824 and an external RNA-seq test AUC of 0.821. The findings highlight the normalization strategy as a decisive computational factor in cross-platform transcriptomic modeling and support the reuse of legacy microarray data for predictive biomedical engineering applications.