Structure-activity relationships using locally linear embedding assisted by support vector and lazy learning regressor

Abstract

Motivation: Structure-activity relationships are characterized by large dimensions and conventional procedures become protracted while modeling these relationships. To enhance the modeling abilities in terms of reduced computational costs motivates the use of recently developed tools in machine learning. Method: Newly developed locally linear embedding is used in reducing the nonlinear dimensions in QSPR and QSAR. The reduced set is subsequently modeled with robust regressors, namely lazy learning and support vector regression. Result: Both the datasets show improved results with the reduced dimensions as compared to their original dimension counterparts. Conclusions: Locally linear embedding for nonlinear dimensionality reduction coupled with robust regressors such as lazy learning and support vector regression seems to be a promising option in analyzing the nonlinear datasets