GALAS Modeling Methodology Applications In The Prediction Of Drug Metabolism Related Properties
Poster Feb 21, 2017
Remigijus Didziapetris, Justas Dapkunas, Andrius Sazonovas and Pranas Japertas
Every model, no matter what data, descriptors, or modeling techniques used to build it, has a certain applicability domain, beyond which the quality of predictions becomes highly questionable. This reality is one of the fundamental issues concerning the effective use of third-party predictive algorithms in industry. The simple reason for this is that literature based training sets rarely cover the specific part of the chemical space that ‘in-house’ projects are focused on. Discrepancies between ‘in-house’ experimental protocols and methods used to measure properties for compounds in publicly available sources further affect the quality of resulting in silico predictions. Therefore the need has long existed for a method that would allow any company to effectively assess the Applicability Domain of any third-party model and to tailor it to its specific needs using proprietary ‘in-house’ data.
Addressing the aforementioned issue, a GALAS (Global, Adjusted Locally According to Similarity) model concept has been developed providing a novel solution to this problem.