In European Biopharmaceutical Review, Optibrium’s CEO Dr Matthew Segall discusses how we can elevate drug discovery with deep learning imputation. He shares insights on the method’s key benefits, including how they help to unlock a world of pharmacokinetics knowledge.


Making effective decisions based on the data available in drug discovery is challenging. Researchers can’t measure all the activities and properties of every single compound of interest for a drug discovery project – that would be too time-consuming and expensive. In a typical pharma company, only about 1% of the possible data will be available. Even within a project, commonly, only 10-20% of the potential data will have been measured in practice. So, decisions must be made based on incomplete or ‘sparse’ data. 

Furthermore, the experimental data that have been measured are noisy. Experiments performed by drug discovery scientists are variable and data will have experimental errors and artefacts. False negatives can lead to missed opportunities, and false positives can mean that time and effort is wasted pursuing hypotheses that later turn out to be based on faulty data. These errors exist, but it’s very difficult to spot them. 

Imputation using deep learning is a recent approach applied to drug discovery that addresses these challenges, and is the process of ‘filling in’ missing data based on limited available measurements ​[1]

About the author

Matt Segall, PhD

CEO, Optibrium


The image shows Optibrium CEO Matthew Segall

[1] ​T. Whitehead, B. Irwin, P. S. M. Hunt and G. Conduit, “Imputation of Assay Bioactivity Data Using Deep Learning,” J. Chem. Inf. Model., vol. 59, no. 3, pp. 1197-1204, 2019. 

Related content from across the site

Meet with us atACS Fall 2024

Join Daniel Barr to hear more about how deep learning imputation prioritises the most relevant data, accounts for uncertainty, and guides experiment selection to bring additional value to small molecule discovery.