Prioritizing treatments for individual cancer patients remains challenging, and performing co-clinical studies using patient-derived models in real-time is often unfeasible. To circumvent these challenges, we introduce OncoLoop, a precision medicine framework that predicts drug sensitivity in human tumors and their pre-existing high-fidelity (cognate) model(s) by leveraging drug perturbation profiles. As proof-of-concept, we applied OncoLoop to prostate cancer (PCa) using genetically-engineered mouse models (GEMMs) that recapitulate a broad spectrum of disease states, including castration-resistant, metastatic, and neuroendocrine prostate cancer. Interrogation of human PCa cohorts by Master Regulator (MR) conservation analysis revealed that most advanced PCa patients were represented by at least one cognate GEMM-derived tumor (GEMM-DT). Drugs predicted to invert MR activity in patients and their cognate GEMM-DTs were successfully validated in allograft, syngeneic, and patient-derived xenograft (PDX) models of tumors and metastasis. Furthermore, Oncoloop-predicted drugs enhanced the efficacy of clinically-relevant drugs, namely the PD1 inhibitor, nivolumab, and the AR-inhibitor, enzalutamide.

No abstract

The classification of high dimensional gene expression data is key to the development of effective diagnostic and prognostic tools. Feature selection involves finding the best subset with the highest power in predicting class labels. Here, we conducted a comparative study focused on different combinations of feature selectors (Chi-Squared, mRMR, Relief-F, and Genetic Algorithms) and classification learning algorithms (Random Forests, PLS-DA, SVM, Regularized Logistic/Multinomial Regression, and kNN) to identify those with the best predictive capacity. The performance of each combination is evaluated through an empirical study on three benchmark cancer-related microarray datasets. Our results first suggest that the quality of the data relevant to the target classes is key for the successful classification of cancer phenotypes. We also proved that, for a given classification learning algorithm and dataset, all filters have a similar performance. Interestingly, filters achieve comparable or even better results with respect to the GA-based wrappers, while also being easier and faster to implement. Taken together, our findings suggest that simple, well-established feature selectors in combination with optimized classifiers guarantee good performances, with no need for complicated and computationally demanding methodologies.