Intelligent Systems for Predictive Modelling in Cheminformatics: QSPR Models for Material Design using Machine Learning and Visual Analytics Tools

MARTÍNEZ, MARÍA JIMENA; PONZONI, IGNACIO; VAZQUEZ, GUSTAVO ESTEBAN; CRAVERO, FIORELLA; DIAZ, MÓNICA FÁTIMA

Advances in Intelligent Systems and Computing

Springer-Verlag

Año: 2016 vol. 477 p. 3 - 11

2194-5357

In this paper, the use of intelligence systems for feature extraction in predictive modelling applied to Cheminformatics is presented. In this respect, the application of these methods for predicting mechanical properties related to the design of the polymers constitutes, by itself, a central contribution of this work, given the complexity of in silico studies of macromolecules and the few experiences reported in this matter. In particular, the methodology evaluated in this paper uses a features learning method that combines a quantification process of 2D structural information of materials with the autoencoder method. Several inferred models for tensile strength at break, which is a mechanical property of materials, are discussed. These results are contrasted to QSPR models generated by traditional approaches using accuracy metrics and a visual analytic tool.