PAIRWISE: Deep Learning-based Prediction of Effective Personalized Drug Combinations in Cancer

Area of Science:

• Computational biology
• Oncology
• Pharmacology

Background:

• Combination therapies are crucial for enhancing cancer treatment efficacy and preventing recurrence.
• Identifying optimal drug combinations is challenging due to numerous possibilities and tumor heterogeneity.
• Preclinical screening can prioritize synergistic drug combinations, but personalized approaches are needed.

Purpose of the Study:

• To develop a computational model, PAIRWISE, for predicting synergistic drug combinations in cancer.
• To address the challenge of identifying personalized drug combinations tailored to specific cancer subtypes and patients.
• To accelerate the development of precision oncology through effective nomination of drug combinations.

Main Methods:

• Developed PAIRWISE, a model explicitly designed to predict synergistic effects of drug combinations.
• Applied PAIRWISE to held-out cancer cell lines and an independent dataset of Diffuse Large B Cell Lymphoma (DLBCL) treated with Bruton Tyrosine Kinase (BTK) inhibitors.
• Validated predictions using high-throughput screening (HTS) of approved or investigational agents for DLBCL.

Main Results:

• PAIRWISE demonstrated superior performance compared to competing models, achieving an AUROC of 0.847 on cancer cell lines.
• The model accurately predicted synergistic combinations in DLBCL with an AUROC of 0.720.
• PAIRWISE showed strong concordance with in vitro screening results, validating its predictive capability.

Conclusions:

• PAIRWISE effectively models synergistic drug effects and nominates personalized drug combinations for cancer treatment.
• The model shows significant potential for accelerating precision oncology development.
• This approach can guide the selection of effective combination therapies for individual patients.