The Predictive Toxicology Evaluation Challenge

Abstract

Can an AI program contribute to scientific discovery? An area where this gauntlet has been thrown is that of understanding the mechanisms of chemical carcinogenesis. One approach is to obtain Structure-Activity Relationships (SARs) relating molecular structure to cancerous activity. Vital to this are the rodent carcinogenicity tests conducted within the US National Toxicology Program by the Nationa! Institute of Environmental Health Sciences (NIEHS). This has resulted in a large database of compounds classified as carcinogens or otherwise. The Predictive-Toxicology Evaluation project of the NIEHS provides the opportunity to compare carcinogenicity predictions on previously untested chemicals. This presents a formidable cha ll enge for programs concerned with knowledge discovery. Desirable features of this problem are: (1) involvement in genuine scientific discovery; (2) availability of a large database with expert-certified classifications; (3) strong competition from methods used by chemists; and ( 4) participation in true blind trials, with results available by next IJCAI. We describe the materials and methods constituting this cha llenge, and provide some initial benchmarks. These show the Inductive Logic Programming tool Progol to be competitive with current state-of-the-art. The cha llenge described here is aimed at encouraging AI programs to avail themselves the opportunity of contributing to an enterprise with immediate scientific value.