*(MAE : Mean absolute error)
[Ranked No.1 with AI Technology] Calici’s proprietary model ‘CaliciBoost’ achieved the best performance (MAE 0.256)!
Calici, a bio-AI startup, announced that it has secured first place in the Caco‑2 permeability prediction task of the Therapeutics Data Commons (TDC)—a global benchmark competition for AI-driven drug discovery. This achievement highlights a significant advancement in accurately predicting the absorption of drug candidates, gaining international recognition.
The Therapeutics Data Commons (TDC) is a widely recognized open benchmarking platform where bio and AI research institutions and companies from around the world evaluate the performance of predictive models for drug development, particularly in ADMET (Absorption, Distribution, Metabolism, Excretion, and Toxicity) properties.
The Caco‑2 task, known for its difficulty, involved predicting the absorption rate of orally administered drugs using a dataset of 906 compounds, with performance measured by MAE (Mean Absolute Error).
Calici’s proprietary AutoML-based model, ‘CaliciBoost,’ achieved a leading MAE of 0.256, outperforming all other participants. The model distinguishes itself through its hybrid analysis of 2D/3D molecular features and automated algorithm optimization, which together significantly improve predictive accuracy.
This accomplishment demonstrates Calici’s technical excellence on a globally recognized benchmark and marks a major milestone as the company expands its role in AI-powered ADMET prediction and drug candidate discovery through international collaboration.