Mining toxicity structural alerts from SMILES: A new way to derive Structure Activity Relationships

Abstract

Encouraged by recent legislations all over the world, aimed to protect human health and environment, in silico techniques have proved their ability to assess the toxicity of chemicals. However, they act often like a black-box, without giving a clear contribution to the scientific insight; such over-optimized methods may be beyond understanding, behaving more like competitors of human experts' knowledge, rather than assistants. In this work, a new Structure-Activity Relationship (SAR) approach is proposed to mine molecular fragments that act like structural alerts for biological activity. The entire process is designed to fit with human reasoning, not only to make its predictions more reliable, but also to enable a clear control by the user, in order to match customized requirements. Such an approach has been implemented and tested on the mutagenicity endpoint, showing marked prediction skills and, more interestingly, discovering much of the knowledge already collected in literature as well as new evidences. The achieved tool is a powerful instrument for both SAR knowledge discovery and for activity prediction on untested compounds.