Skip to main content
Contact Info
Swapnil Chavan

OpenTox 2022 Virtual Conference 

AI-TOX: An end-to-end Solution for Automating Predictive Toxicology by AI

The performance of any machine learning model is governed by choice of input features. These features usually either hand crafted using prior human knowledge or derived using mathematical procedures. During model building process, highly correlated features and features that are constant are usually discarded, followed by a feature selection step. Such a feature engineering has well supported machine learning algorithms, to date. However, robustness of these models heavily dependent on devising the correct features predictive enough for the given task. Failing in devising right features, an algorithm cannot be trained to solve the given task better than a guesswork.

The rise of artificial intelligence (AI) based algorithms has gained a lot of interest in the pharmaceutical development field. AI algorithm can automatically learn to extract precisely right features using observations alone, without any human intervention. These algorithms can simulate ionic bonds, can handle all periodic table elements, can dynamically use augmented input SMILES, and allows reversible operation for encoding any SMILES and reconverting them back. On top, AI algorithms can allow limitless size of SMILES encodings which otherwise is limited to ~6000 two-dimensional descriptors or ~40000 Gobbi Pharm 2D fingerprints with traditional methods. With AI, one can trace back important structural features in the given SMILES from a predicted outcome using various explainable-AI techniques, which is not possible with traditional methods. The advanced deep learning techniques allowing newer ways to address model’s uncertainty which is not available with traditional methods. In our AI-TOX program, we have developed end-to-end solutions for predicting various toxicities. We believe this is a comprehensive step towards automating predictive toxicology by virtue of predictivity, certainty and explanability.

CV: Swapnil Chavan (RISE AB)

Swapnil Chavan got his masters in Pharmacoinformatics from NIPER, India, and received his PhD in Computational Chemistry from Linnaeus University, Sweden. Then he went to National Center for Computational Toxicology at US – Environmental Protection Agency for his postdoctoral research where he served as a Unilever postdoc. Swapnil currently works as a senior data scientist in computational toxicology at RISE AB. He leads innovative research towards developing novel chemical risk assessment methods using cheminformatics, bioinformatics, image-informatics and machine learning techniques.