University of Edinburgh
Preclinical discovery of biomarkers of response to NXP900 using large-scale pharmacogenomic screens
Abstract
dc:description.abstractPrecision oncology requires predictive biomarkers to stratify patients for targeted therapies. NXP900 is a novel, selective SRC/YES1 inhibitor currently in Phase I clinical trials, yet predictive biomarkers of response remain undefined. The purpose of this thesis was to identify predictive biomarkers of NXP900 response using integrated multi-omics analysis and to validate key findings through experimental approaches. This work aimed to characterise drug response patterns across cancer lineages, develop optimal predictive models, and elucidate mechanisms of response and resistance. Viability assays with NXP900 were conducted across four independent datasets comprising over 900 cancer cell lines. Multiple concentration-response metrics were evaluated to identify optimal approaches for quantifying drug response. Multi-omics data from the Dependency Map project were comprehensively characterised and integrated using machine learning approaches to identify biomarkers and signatures predictive of response to NXP900. Predictive performance of multiple models, feature selection, feature engineering, and dimensionality reduction methods were rigorously evaluated through cross-validation. Key findings were validated through targeted experimental investigation using CRISPR-Cas9 knockout studies. Analysis revealed lineage-specific patterns of NXP900 sensitivity and resistance, with notable therapeutic windows in specific cancer subtypes. Area-over-curve within the therapeutic range emerged as the optimal response metric for biomarker discovery. Analysis of response to NXP900 across cancers and subtypes revealed certain cancers consistently found to be sensitive to NXP900. Multi-omics characterisation identified confounding by cancer lineage and technical variation that posed significant challenges for biomarker identification. No approach was able to improve predictive performance over a simple model trained on only lineage and subtype information, suggesting the models were unable to learn consistent and generalisable molecular markers of response to NXP900. Experimental validation confirmed enhanced sensitivity of ARID1A-mutant ovarian clear cell carcinoma, with mechanistic investigation revealing potential drivers of response. This work establishes a comprehensive framework for multi-omics biomarker discovery, while highlighting critical challenges that threaten biomarker validity. The inability of multi-omics models to outperform lineage-based predictions suggests that further preclinical work is required, utilising alternative methods and datasets, to identify the mechanisms underlying response to NXP900. Clinical development could prioritise patient stratification by cancer subtype, as robust molecular determinants of response proved difficult to disentangle from lineage effects.
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- King, Ben G.
- Advisors dc:contributor.advisor
-
- Unciti-Broceta, Asier
- Carragher, Neil
- Hollis, Robb
- Brunton, Val
Subjects
dc:subject × 6Rights
- Language dc:language.iso
- en
Identifiers
dc:identifier.*- Identifier URI
- https://doi.org/10.7488/era/7425