Treatment regimens, especially in cancer, often include more than one medicine in order to achieve durable outcomes. Identifying the optimal combination of treatments has historically been done through clinical trial and error. And for many conditions, such as pancreatic cancer, an optimal treatment protocol has remained elusive, and the best available treatment combinations provide only modest benefit. Recent developments have led to the application of both experimental screening approaches and in silico modeling methods to identify synergistic drug combinations and expand the therapeutic options for multiple diseases. Here we conduct a study to compare different predictive approaches for identifying new treatment combinations for pancreatic cancer using cell line growth as an initial proxy for clinical utility. NCATS performed screening involving 496 pairwise combinations of 32 antineoplastic drugs, tested against the PANC-1 human pancreatic carcinoma cell line in duplicates using a 10 × 10 matrix format. This dataset served as the basis for generating and training advanced AI/ML models focused on pancreatic cancer. Next, three independent groups (NCATS, UNC and MIT), though in a collaborative manner, utilized three different workflows with AL/ML approaches to discover new perspective drug combinations against pancreatic cancer among over 1.5 million drug combinations. As a result of this collaboration, 88 proposed combinations were tested in a cell-based assay; 53 of them were synergistic (hit rate ~60%). While all machine learning approaches demonstrate advances in the direction of predicting synergistic drug combinations, graph convolutional networks resulted in the best performance with a hit rate ~83%, and Random Forest delivered the highest precision of 65%. Interestingly, all utilized AL/ML methods among the three groups proposed different drug combinations with a small overlap of only two combos from 90. This study demonstrates the potential of a collaborative modeling approach for prioritizing drug combinations in large-scale screening campaigns, particularly when focusing on maximizing the efficacy of drugs known to exhibit synergy.
In the field of medicinal chemistry, the discovery of novel compounds with therapeutic potential is of utmost importance. However, the conventional approach to drug discovery, relying on high-throughput screening (HTS), encounters limitations such as reagent availability and labor costs. Although a hit-finding campaign starts with a virtual screen of millions of compounds, the hit-to-lead and lead optimization stages often require designing, synthesizing, and profiling thousands of analogs before selecting clinical candidates. Recent advances offer an innovative solution - the virtual generation of billions of synthetically feasible compounds with an impressive 80% success rate for synthesis. This breakthrough has the potential to significantly expand the chemical space available for purchase and experimental validation, bringing about a revolution in the field. Our study presents a comprehensive approach to compound discovery and optimization, harnessing quantitative high-throughput screening (qHTS), chemical databases, and reaction-based enumeration. To further enhance the synthesis process and gain deeper insights into compound formation, we utilize the Reaction Cookbook from Biosolveit, which comprises reaction SMARTs for around 300 chemical reactions. By employing this wide range of reactions, we aim to uncover the full spectrum of a chemotype's potential and customize its structure to optimize desired properties. As an illustration of this approach, our work on the ALDH3A1 project resulted in the synthesis of 50 compounds, with 21 of them exhibiting activity (negative curve class values; hit-rate ~42%). Among these active compounds, 6 displayed IC50 values lower than 30 µM and efficacy values less than -50%, with the most potent compound achieving an impressive potency of 35 nM. This study demonstrates the successful synergy between in-silico reaction-based analogs enumeration, molecular modeling and AI/ML-based techniques in identifying compounds with improved biological activity, offering promising prospects for the development of ALDH3A1-targeting agents as potential cancer therapeutics. Computational workflows developed in this study can be used for similar target-based drug discovery campaigns.
Traditional best practices for Quantitative Structure Activity Relationship (QSAR) modeling recommend dataset balancing and balanced accuracy (BA) as the key desired objective of model development. This study challenges the conventional norms by recommending the use of models with the highest positive predictive value (PPV) built for imbalanced training sets as preferred tools in virtual screening campaigns for drug discovery. As proof of concept, we developed QSAR models for five expansive datasets with different ratios of active and inactive molecules and assessed model performance using BA, PPV, and other metrics. We demonstrated that PPV-oriented models used in virtual screening have at least 30% higher first-batch hit rate as compared to all other models. These findings suggest that QSAR models trained on imbalanced datasets for the highest PPV are preferred tools for virtual screening campaigns in drug discovery.