ersilia-os/ai2050-h3d-symposium-workshop

Materials for a two-hour introductory workshop on AI/ML for drug discovery, delivered at the 5th H3D Symposium in Zambia. # Introductory workshop to AI/ML for antimicrobial drug discovery This repository contains the materials for an introduction to AI tools for drug discovery, delivered at the 5th H3D Symposium in Livingstone, Zambia. The demo contained within this repository is only brought alive online during the workshop delivery. If you wish to use it for demo purposes, please email us at hello[at]ersilia[dot]io ## Hands on Activity You can find detailed information about the proposed demo in this page. Find below a quick summary! The participant is presented with the following problem statement: _We are a laboratory specialised in antimicrobial research. We have received a library of compounds from a collaborator and we need to explore it and identify the best candidates._ Some of the considerations to take into account: * We can synthesise compounds but our throughput is 50 compounds per month * We have a tight timeline (2 months) to provide the results * Our collaborators are keen on exploring analogues of the molecules in their library as well * The selected compounds need to meet the following criteria: * Synthetically accessible in the laboratory * Good ADME profile * High activity against at least one pathogen in the WHO priority list ### Step 0: AMA A simple plug to GPT3.5 to ask questions around pathogens. ### Step 1: Train an ML Model Train a simple classifier model based on the dataset for _A.baumannii_ activity reported in Liu et al, 2023. The participants can play with different cut-offs and the demo performs an automated 5-fold cross-validation. ### Step 2: Prioritize candidates Participants will use a given dataset (prepared from ChEMBL) and run predictions using the just trained _A.baumannii_ model as well as two models from the Ersilia Model Hub: Synthetic Accessibility Score (Ertl et al, 2009) and hERG cardiotoxicity (Jiménez-Luna et al, 2021). The goal is to select the best molecule according to the predicted values (high activity against _A.baumannii_, g …