In silico/Mechanistic and Discovery Toxicology speciality sections’ joint workshop: hands-on modelling of safety data for non-experts
Free, online, open to all!
29 November 2022 – 14:00 – 17:00 UK time, online via Zoom
While machine learning (ML) / artificial intelligence (AI) models are gaining popularity across scientific fields, their inner workings are not transparent to everyone, and hence they are often approached with hesitation. In this hands-on workshop, we will make the construction, and use, of machine learning models clear to a non-expert audience with a general life science/safety sciences background.
Firstly, a lecture will introduce the basics of predictive ML models: Data, descriptors, ML-algorithms, and validation of models. Secondly, two hands-on sessions (on target prediction and mitochondrial toxicity classification) will illustrate step-by-step how to prepare input data, how to train models, and how to apply them on new datasets.
We will conclude by discussing possible use cases and pitfalls of the resulting models. The models will be trained in Google Colab which is run through most common browser interfaces, so no local software installation is required, and steps of model construction will be explained throughout.
Programming knowledge is not necessary to be able to follow the course, though this will be useful to understand details of the code and to be able to modify it. We hope that this course will bring the fields of ML/AI and of safety/toxicology closer together, to the mutual benefit of both fields.
14:00 to 14:10 Session intro Nic Coltman (Apconix) and Andreas Bender (University of Cambridge)
14:10 to 15:00 Fundamentals of modelling Andreas Bender
15:00 to 15:50 Exercise 1: Regression approach to target prediction Layla Hosseini-Gerami (Absolute AI)
Break 15 min
16:05 to 16:55 Exercise 2: Classification models for mitochondrial toxicity Srijit Seal (University of Cambridge)