Success Through Statistics in High Throughput Drug Discovery - Graeme Robb - AstraZeneca
[This talk was presented at the UK JMP Usergroup meeting on the 13th of July 2017]
In 1980s and 90s, drug discovery companies massively invested in High Throughput Screening (HTS) technologies, allowing millions of compounds to be rapidly screened. This has proven to be a naive approach. Bigger, it seems, is not better and the investment in HTS has not delivered a higher rate of drug approvals in the intervening years. What we now realise is that quantity is perhaps less important than quality. The 'predictive validity' of an assay suitable for HTS is necessarily compromised and may not directly relate to the disease phenotype of interest. High false-positive rates and noisy make it difficult to identify the true positives. Understanding our data and using statistical techniques help to improve our ability to recognise false positives. Statistical learning from past HTS campaigns is used to enhance the results and further spot systematic false positives. In the future, we will face further challenges of scale and predictive validity where statistics, modelling and machine learning can all be used to our advantage.
