U-SAR solution is a machine learning insight generation and data visualization tool used to rationalize and streamline multiparametric molecular optimization.
Traditional SAR models trained on large chemical spaces systematically fail to predict molecular properties when it comes to compound optimization in limited chemical spaces. U-SAR turns the usual tedious and time consuming iterative multiparametric optimization process into an increasingly productive and converging experience through medicinal chemists’ cognitive augmentation.
The product was designed with chemists and for chemists and proved to significantly increase the efficiency of multiparametric compound optimization where traditional SAR models usually reach their limits.
The U-SAR methodology starts with the identification of promising combinations of molecular features for each criterion to be optimized. These criteria include for instance solubility, potency, and metabolic stability of the compound profiles. This initial step is followed by the augmented molecular optimization step aimed at maximizing chances to synthesize compounds satisfying simultaneously multiple target properties. Our tool also allows getting a strategic overview of the project effort and resources thanks to an interactive dashboard.
Quinten Health initially proved the efficiency of the approach by comparing it with the usual empirical approach to multiparametric lead optimization. A 2-years long chemical lead optimization program in oncology led a pharmaceutical company to retaining 5 molecular candidates out of 1,600 iterative chemical designs. Quinten Health worked on the retrospective data to showcase its ability to generate insights relevant to streamline the drug development by leveraging machine learning.
Our data scientists first explored the data concerning the 400 compounds generated during the 6 first months of the program. We identified 4 compound profiles with properties above average. These profiles were then mapped to the 1,200 more compounds which were further generated during the program.
The results of the exploration demonstrated that the chemists could have identified the 5 promising molecules in less than a year instead of two years if they had given priority to the compounds matching with at least one of the 4 profiles identified by Quinten Health based on the data accumulated in the first 6 months.
Since then, U-SAR was used in several chemical lead optimization programs, saving considerable amounts of time and resources for our partners.