Search

Most of the proteins locate more than one organelle in a cell. Unmixing the localization patterns of proteins is critical for understanding the protein functions and other vital cellular processes. Herein, non-linear machine learning technique is proposed for the first time upon protein pattern unmixing. Variable-weighted support vector machine (VW-SVM) is a demonstrated robust modeling technique with flexible and rational variable selection. As optimized by a global stochastic optimization technique, particle swarm optimization (PSO) algorithm, it makes VW-SVM to be an adaptive parameter-free method for automated unmixing of protein subcellular patterns. Results obtained by pattern unmixing of a set of fluorescence microscope images of cells indicate VW-SVM as optimized by PSO is able to extract useful pattern features by optimally rescaling each variable for non-linear SVM modeling, consequently leading to improved performances in multiplex protein pattern unmixing compared with conventional SVM and other exiting pattern unmixing methods., (Copyright © 2015 Elsevier B.V. All rights reserved.)

Abstract: One problem with discriminant analysis of microarray data is representation of each sample by a large number of genes that are possibly irrelevant, insignificant or redundant. Methods of variable selection are, therefore, of great significance in microarray data analysis. To circumvent the problem, a new gene mining approach is proposed based on the similarity between probability density functions on each gene for the class of interest with respect to the others. This method allows the ascertainment of significant genes that are informative for discriminating each individual class rather than maximizing the separability of all classes. Then one can select genes containing important information about the particular subtypes of diseases. Based on the mined significant genes for individual classes, a support vector machine with local kernel transform is constructed for the classification of different diseases. The combination of the gene mining approach with support vector machine is demonstrated for cancer classification using two public data sets. The results reveal that significant genes are identified for each cancer, and the classification model shows satisfactory performance in training and prediction for both data sets. [Copyright &y& Elsevier]

Abstract: Representing a compound by a numerous structural descriptors becomes common in quantitative structure–activity relationship (QSAR) studies. As every descriptor carries molecular structure information more or less, it seems more advisable to investigate all the possible descriptor vectors rather than traditional variable selection when building a QSAR model. Based on particle swarm optimization (PSO) algorithm, a more flexible descriptor selection and model construction method variable-weighted support vector machine (VW-SVM) is proposed. The new strategy adopted in this paper is to weight all structural descriptors with continuous non-negative values rather than removing or reserving any ones arbitrarily. The manner of invoking PSO to seek non-negative weights of variables can be regarded as a process of searching optimized rescaling for every molecular structural descriptor. Moreover, PSO is employed to search the optimal parameters of VW-SVM model besides variable weights, enables the construction of a rational and adaptive parameter-free QSAR model according to the performance of the total model. Results obtained by investigating glycogen synthase kinase-3α inhibitors and carbonic anhydrase II inhibitors indicate VW-SVM can hold more useful structure information of compounds than other methods as optimally weighting all the descriptors, consequently leading to precisely QSAR models coupled with developed performance both in training and in prediction. [Copyright &y& Elsevier]