Topics covered

• Introduction to in silica toxicity
• Predictive models of toxicity and other properties of interest: risks, targets, classification, doses, similarities. Free tools to be addressed: Protox, Superpred, Emoltox, Swissadme, Predskin, Pred-Herg.
• Skin irritation: definition and biological mechanisms. Relevance to cosmetics, drugs and chemicals. Relevant public database (eg oecd qsar toolbox, toxcast, echa).
• Molecular representation and descriptors: introduction of RDKIT and Dragon. Molecule formats (Smiles, SDF, Mol2, etc.). RDKIT: Introduction and installation on Google Colab. How to visualize molecules and generate basic descriptors.
• Dataset construction and predictive modeling: creation of a simple model to predict skin irritation.
• How to format data for Machine Learning (CSV with columns: Smiles + toxicity). Data pre-processing: Smiles transformation into descriptors.
• Creation of the simple AI model: use of random forest or logistics regression. Training and evaluation of the model.
• Testing the model and predicting new molecules: making predictions with new molecules and discussing model limitations.
• Introduction to Integrated Chemical Environment (ICE) and its tools: Search, Curve Surfer, PBPK, Ivive and Chemical Quest. Presentation of the High-ThroughPut Screening (HTS) database, composed of in vitro and in vivo information from Tox21 and Toxcast programs for various chemicals. Explanation of the data curatorship process, including chemical quality control and curved adjustments. Association of mechanistic target tests and their relevance to regulatory toxicology. Practical exploitation of ICE and its tools, with demonstration of the use of Search to search for compounds and data, curve surfer for dose-response curves and PBPK and IVIVE for toxicological modeling.
• Introduction to Modeling and Visualization Pipeline (Moviz), which aims to democratize the methods of chemoinformatics for non -expert and simplify their application to the community.
• Moviz’s first tool is a knime workflow to facilitate chemical grouping using supervised and non -supervised machine learning methods. This tool aims to assist researchers in assessing the chemical diversity of data sets, obtaining information on the mechanisms of action, extraction of structure-activity and prioritization of chemicals for experimental tests. Case studies will be presented with practical applications to illustrate the use of these tools.