Abstract A030: Advancing Pharmacogenomic Modeling: Robust Dose-Response Curve Fitting and Drug Combination Analytics in the Next-Generation PharmacoGx Framework
Nikta Feizi, James Bannon, Kewei Ni, Jermiah Joseph, Petr Smirnov, Zhaleh Safikhani, Mathieu Lupien, Anna Goldenberg, Benjamin Haibe‐Kains · Clinical Cancer Research · 2025
Abstract Pharmacogenomic datasets continue to grow in volume and diversity, but their translation into clinical insights remains limited by fragmented data formats and inflexible toolkits. PharmacoGx has addressed this challenge since 2016 as a foundational package in computational pharmacogenomics, with over 54,000 downloads to date. It introduced the widely adopted PharmacoSet (PSet) structure, enabling reproducible integration of drug response and molecular profiles across diverse cancer models and supporting landmark studies in the field.We present a major update to the PharmacoGx framework, designed to meet the demands of modern high-throughput drug screening and pharmacogenomic modeling. This release introduces substantial improvements in robustness, scalability, and flexibility, especially for users working with complex or heterogeneous datasets. Notable updates include support for Huber loss-based dose-response curve fitting, enhancing resilience to outliers, and biphasic curve modeling, which better captures adaptive or nonlinear drug behaviors often seen in cancer systems.To address the rising interest in combination therapies, we introduce a built-in module for drug combination analysis, supporting synergy scoring, cross-platform benchmarking, and modeling of multi-agent regimens, an unmet need in current pharmacogenomic tools. These functions are compatible with both legacy PSet objects and a newly implemented TreatmentResponseExperiment (TRE) data structure.The TRE introduces a modular, long-format, data.table-backed architecture that replaces the static PSet sensitivity slot. It enables rapid filtering, subsetting, and group-wise operations (endoaggregation) within the object itself, minimizing memory overhead and supporting high-throughput analysis across millions of experimental points. The accompanying DataMapper utility lowers the barrier for researchers to import and analyze custom datasets, promoting reproducible and adaptable workflows. By extending the computational capabilities and usability of PharmacoGx, this update provides a powerful, scalable platform for drug response modeling in precision oncology. These enhancements facilitate robust AI-based analytics, including machine learning model training and predictive response profiling, enabling next-generation applications in translational cancer research. Citation Format: Nikta Feizi, James Bannon, Kewei Ni, Jermiah Joseph, Petr Smirnov, Zhaleh Safikhani, Mathieu Lupien, Anna Goldenberg, Benjamin Haibe-Kains. Advancing Pharmacogenomic Modeling: Robust Dose-Response Curve Fitting and Drug Combination Analytics in the Next-Generation PharmacoGx Framework [abstract]. In: Proceedings of the AACR Special Conference in Cancer Research: Artificial Intelligence and Machine Learning; 2025 Jul 10-12; Montreal, QC, Canada. Philadelphia (PA): AACR; Clin Cancer Res 2025;31(13_Suppl):Abstract nr A030.