Work

XevaDB enables AI-driven longitudinal pharmacogenomic modeling in patient-derived xenografts

Year: 2026
Type: preprint
Abstract: Patient-derived xenografts (PDXs) are critical models for translational oncology, but their computational use is limited by fragmented pharmacogenomic data, inconsistent response annotations, and poor access to raw longitudinal tumour-volume measurements. Here we present XevaDB (xevadb.ca), a standardized platform designed to support machine learning (ML) and artificial intelligence (AI) applications through the integration of longitudinal treatment and control-arm tumour growth curves with harmonized drug and gene annotations, response metrics, replicate-level experimental design, clinical annotation, and matched molecular profiles. XevaDB provides access to data from 689 PDX models across six tumour types, 69 drugs, 4,828 treatment-control experiments, and matched RNA-seq, copy-number, and mutation data for 337 models. Through four case studies, we show that preserving raw longitudinal response data enables analyses beyond conventional static endpoints: transcriptomic prediction of intrinsic tumour growth, image-based deep learning of drug-response trajectories, replicate-aware transfer learning from cell-line-trained models, and multi-omics prediction of paclitaxel response. These results establish XevaDB as a resource for sharing standardized longitudinal pharmacogenomic data and enabling reproducible AI/ML modeling, demonstrating how integrated PDX data can accelerate computational approaches in preclinical cancer research.
Language: English
Subfield: Cancer Research
Open Access status: green
OpenAlex ID: W7236800985
Cites: 45
Cited by:
Related to: 10