Prediction of Acquired Taxane Resistance Using a Personalized Pathway-Based Machine Learning Method.

Purpose: This study was conducted to develop and validate an individualized prediction model for automated detection of acquired taxane resistance (ATR).

Materials and Methods: Penalized regression, combined with an individualized pathway score algorithm, was applied to construct a predictive model using publically available genomic cohorts of ATR and intrinsic taxane resistance (ITR). To develop a model with enhanced generalizability, we merged multiple ATR studies then updated the learning parameter via robust cross-study validation.

Results: For internal cross-study validation, the ATR model produced a perfect performance with an overall area under the receiver operating curve (AUROC) of 1.000 with an area under the precision-recall curve (AUPRC) of 1.000, a Brier score of 0.007, a sensitivity and a specificity of 100%. The model showed an excellent performance on two independent blind ATR cohorts (overall AUROC of 0.940, AUPRC of 0.940, a Brier score of 0.127). When we applied our algorithm to two large-scale pharmacogenomic resources for ITR, the Cancer Genome Project (CGP) and the Cancer Cell Line Encyclopedia (CCLE), an overall ITR cross-study AUROC was 0.70, which is a far better accuracy than an almost random level reported by previous studies. Furthermore, this model had a high transferability on blind ATR cohorts with an AUROC of 0.69, suggesting that general predictive features may be at work across both ITR and ATR.

Conclusion: We successfully constructed a multi-study-derived personalized prediction model for ATR with excellent accuracy, generalizability and transferability.