Machine Learning modules – Toxophores / Pharmacophores – 1st Feb 2019
We aim for each annual major release to include a new tool. V1.6 is one of our biggest upgrades in functionality. A new suite of Machine Learning (ML) models for predicting ‘potency’ against primary targets has been added. MCPairs Online is now running this version and we have added over 1000 ML models, using data from ChEMBL and BindingDB. A single or multiple compound predictions can be completed and each compound is run against Regression Forest and k-Nearest Neighbour models. Critically the output show the parts of molecule that lead to the prediction. This makes the ML models are Explainable Artificial Intelligence (AI).
Managing the meta-data:
The Enterprise version has a meta-data management panel. Loading datasets to MCPairs for matched pair analysis automatically cleans the data and makes it AI ready. Once loaded the dataset can be registered as a new Pharmacophore and the ML models will build and update automatically.