ACCELERATE

BACKGROUND: Despite advances in understanding pancreatic cancer (PC), clinical management has not significantly improved and many patients do not benefit from standard therapies and experience significant toxicity. Combination therapies offer better survival outcomes than monotherapy, but the large number of possible combinations and high costs make experimental screening infeasible. To address these challenges, computational methods have been developed to accelerate the identification of effective drug combinations.
HYPOTHESIS: We hypothesise that deep learning approaches fine-tuned with patient’s specific data can effectively prioritize and identify optimal drug combinations tailored to individual patients. Multiple precision medicine trials have highlighted significant challenges for patients who lack treatments designed to their specific mutations. Functional precision medicine (FPM) is an approach that involves directly exposing patient-derived tumour cells to drugs, generating data that can reveal key vulnerabilities of the tumour that are not necessarily driven by genomic aberrations. We hypothesise that integrating the FPM approach, which combines molecular tumour profiling, clinical data and high-throughput drug sensitivity testing on patient-derived organoids (PDOs), will uncover new treatment options for tumours with limited therapeutic choices, such as PC.
AIMS: First: build an FPM platform for PC that integrates the input of in vitro drug testing, omics and clinical data for individual patients with the main goal of matching patients with effective and optimal therapies and returning treatment recommendations in real time to the clinicians in a clinically exploitable timeframe. Second: identify new therapy response/resistance biomarkers. Third: identify resistance pathways and mechanisms.
METHODS: Using a panel of FDA-approved drugs, single drug response on each PDO will be assessed. Then, a deep learning model, pre-trained on large drug pair synergy measurements on cell lines, will be fine-tuned on the patient-specific molecular data to make personalized drug combination synergy predictions for each patient. The prioritized patient's specific drug combinations will be evaluated on the patient’s matched PDOs; best-scored combinations will be validated also on patient-derived xenograft (PDX). Lastly, results from FPM and molecular characterization will be integrated with patient specific clinical data to build a therapy decision supporting report.
EXPECTED RESULTS AND POTENTIAL IMPACT: We expect to build a platform for PC combinatorial drug testing, integrating patient specific clinical, genomic and transcriptomic data with FPM results. Potential clinical impact: i) to return treatment recommendations in real time to the physician in a clinically exploitable timeframe ii) to identify more effective treatment options for PC patients iii) to provide a proof of concept platform for clinical study design.