(1274-C) DiscoveryAI SAFIRE: a collection of machine learning models to predict ADMET liabilities for small molecule drug discovery
Tuesday, February 6, 2024
12:00 PM – 1:00 PM EST
Location: Exhibit Halls AB
Abstract: As the small molecule drug discovery process is time-consuming and expensive, computational prediction of ADMET properties of drug candidates can lead to the reduction of time and resources exerted by eliminating unsuccessful compounds. Successful predictions are reliant on quality datasets, such as Eurofins Discovery’s proprietary BioPrint database, to effectively train machine learning algorithms. We have developed our SAFIRE models with the BioPrint database and public data to predict ADMET properties from chemical structure alone. SAFIRE models perform comparably with other industry efforts, with target performance of Matthews Correlation Coefficient (MCC) greater than 0.4 and accuracy greater than 75%. Our efforts reiterate that the BioPrint database remains a valuable resource for machine learning efforts and demonstrate the utility of our tools to advise the decision-making process during drug discovery projects. SAFIRE models are expected to launch this month.