ABSTRACT With a rapid diffusion, Leishmaniasis now appears as a severe tropical disease with millions of people affected. Currently used drugs are not devoid of detrimental side-effects and there is a crucial need for new alternative, active anti-parasitic chemicals. Recently, Nobrega et al. synthetized 32 analogues of piplartine, determined their activity against Leishmania amazonensis promastigote forms and presented a comparative molecular field analysis (CoMFA) treatment. Here we revisited these results and proposed topology-based 2D correlation models with special attention to robustness and predictive ability. From PaDEL and QSARINS softwares, a set of 3 descriptors was selected, and imported into multilinear regression and various machine learning approaches: partial least squares (PLS), projection pursuit regression (PPR), support vector machine (SVM with linear or Gaussian kernel) and three-layer perceptron (TLP, neural network with back-propagation algorithm). Although a reduced set of structural descriptors, these different models appeared attractive with satisfactory and consistent performances. The best results, obtained from linear SVM and three-layer perceptron, suggested that these models might be applied for screening new possible drugs.
View Full Article
|