An integral framework for QSAR modelling using computational intelligence and visual analytics

CRAVERO, FIORELLA; MARTINEZ, MARÍA JIMENA; DIAZ, MÓNICA F.; VAZQUEZ, GUSTAVO E.; PONZONI, IGNACIO

Bahía Blanca

Conferencia; VI Argentinean Conference on Computational Biology and Bioinformatics; 2015

Asociación Argentina de Bioinformática y Biología Computacional

This paper presents an integrated methodology for the development of predictive models in this field. It includes the use of QSPR technique (Quantitative Structure-Activity Relationship), the use of an automatic feature selection algorithm (Delphos), the use of a feature learning technique that encodes 2D molecular information and a methodology for visual analytics and expert assistance (VIDEAN). QSPR is a technique that allows to build predictive models relating molecular descriptors of the object of interest with some experimental property. To build-up these models, it is desirable to discover the best parameters describing the property of interest. In one hand, this task is performed by Delphos, an in-house developed software based on multi-objective genetic algorithms. On the other hand, a feature learning technique allows to extract the main 2D features of a molecular entity and to reduce the dimensionality of this information by using an autoencoder method. Finally, VIDEAN provides a web based tool for the analysis and development of prediction models assisted by visual analytics techniques. Different case studies that exemplify the interaction of these tools for the development of a particular prediction model are presented.