Public domain databases, such as ChEMBL and PubChem, are excellent sources from which to obtain compound data for modelling of target activities or other properties. However, despite the best attempts of the maintainers of these databases, significant pre-processing of the raw data downloaded from these databases is necessary before high quality models can be built. Issues include the mixing of qualified and unqualified values; high variability between different measurements of the same activity for a compound, leading to uncertainty about the most appropriate value to use; and duplicate, identical values of an activity for the same compound, resulting from multiple publications of the same measurement. In this presentation, we will describe an automated workflow to pre-process raw downloads from ChEMBL to generate a clean data set, for example filling in missing standardised activity values, checking comparability between different assays and discarding qualified data, where appropriate, while retaining as much of the initial information as possible. The resulting sets are ready for building of quantitative structure-activity relationship (QSAR) models. The Python code implementing this workflow is available for download. We will also describe how the clean data set can be used as input to an automated process for generation and validation of QSAR models. This builds multiple models using different statistical and machine learning methods, compares the resulting models to identify the best model and then validates the best model using an external independent test set. The combined workflow makes it easy to quickly develop, validate and deploy high quality models from the rich data available in the public domain. This workflow will be illustrated with example models for target activities and absorption, distribution, metabolism and elimination (ADME) properties.

Quality data to quality models

255th National Meeting (2018)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.