For decades, cheminformatics has leaned on the same traditional tools to represent molecules—physicochemical properties and descriptors originally designed for drug discovery. While these methods have proven useful, they often fall short when applied to broader chemical spaces, especially environmental toxicology.
That’s where Sciome’s latest innovation, Saagar descriptors, comes in. Designed to push the boundaries of chemical modeling, Saagar is an extensible library of molecular substructures and moieties that goes beyond drug-like compounds to better capture the diversity of environmental chemicals. What sets Saagar apart is not only its strong performance but also its interpretability, giving scientists clearer insight into how specific structural features drive toxicity outcomes.
The Sciome team has also developed a semi-automated process to continuously expand Saagar’s coverage. By identifying underrepresented chemical substructures and creating new SMARTS patterns, the framework ensures that researchers can adapt descriptors to the unique chemistry of their datasets.
Benchmarking results have been impressive. In extensive tests covering over 400 assays and roughly 45,000 models across nine modeling methods, Saagar consistently matched or outperformed established descriptor sets. Notably, Saagar improved prediction accuracy for nitrosamine toxicity (boosting R² by up to 0.15) and mutagenicity (increasing ROC AUC by 0.1).
With Saagar, Sciome is reshaping cheminformatics by combining adaptability, transparency, and cutting-edge performance—delivering a new standard for chemical modeling in toxicology and beyond.
The American Chemical Society Fall 2025 Meeting (ACS Fall 2025), being held in Washington, D.C. August 17-21, brings together thousands of chemistry professionals to meet and share ideas and advance scientific and technical knowledge.
