Christina Maria Founti
21st October 2016 at 10am, Information School, RC-231
Title: Incorporating prediction error estimates in the evaluation of QSAR models
Abstract: Quantitative Structure-Activity Relationship (QSAR) modelling is a widely accepted, non-testing method for generating data in the chemical industries. Limitations of the method are well understood and often require skillful assessment of the accuracy and reliability of models. However, standard measures of model quality are only based on accuracy. This project investigates methods for QSAR model evaluation that account for reliability by incorporating measurement error in collected data and the estimated error of predictions. The literature review focuses on QSAR methods implementing physicochemical and topological features of molecules, supervised machine learning algorithms and main approaches for obtaining prediction error estimates. Benchmark results of a preliminary modelling experiment are reported and assessed using a statistical divergence method reported in a recent publication.
This event is free and open to all. There is no requirement to pre-register.
21st October 2016 at 10am, Information School, RC-231
Title: Incorporating prediction error estimates in the evaluation of QSAR models
Abstract: Quantitative Structure-Activity Relationship (QSAR) modelling is a widely accepted, non-testing method for generating data in the chemical industries. Limitations of the method are well understood and often require skillful assessment of the accuracy and reliability of models. However, standard measures of model quality are only based on accuracy. This project investigates methods for QSAR model evaluation that account for reliability by incorporating measurement error in collected data and the estimated error of predictions. The literature review focuses on QSAR methods implementing physicochemical and topological features of molecules, supervised machine learning algorithms and main approaches for obtaining prediction error estimates. Benchmark results of a preliminary modelling experiment are reported and assessed using a statistical divergence method reported in a recent publication.
This event is free and open to all. There is no requirement to pre-register.
Comments