International Journal of Computational Bioinformatics and In Silico Modeling
ABSTRACT: There are number of common activation functions in use with artificial neural networks (ANNs). The most common choice of activation functions for artificial neural networks (ANNs) is used as transfer functions in research and engineering. The most common reasons for the use of this popularity were its boundedness in the unit interval, the functions, and its derivative’s fast computability, and a number of amenable mathematical properties in the realm of approximation theory. Objective: The purpose of this paper was to find out the most effective activation function in doing logic programming in the context of Hopfield network. Methods: A comparing hyperbolic tangent activation function, bipolar activation function, unipolar activation function and McCulloch-Pitts function were carried out based on Wan Abdullah’s method, evaluations from global minima ratio, Hamming distance and computational time. These functions are used in activation function in a logic program for experimental comparisons. Additionally, computer simulations have been tested using software NETLOGO 5.3.1(64bit) based on Wan Abdullah’s method doing the logic programs to demonstrate the ability of Hyperbolic tangent activation function, Unipolar activation function, Bipolar activation function and McCulloch-Pitts function. Hyperbolic tangent activation function resulted in the most successful one compared with bipolar activation function, unipolar activation function, and McCulloch-Pitts function. According to our experimental study, we can say that the hyperbolic tangent activation function can be used in the vast majority of ANN applications as a good choice to obtain high accuracy.
KeyWords: Logic Programming in Hopfield network, Types Activation Functions, Implementation of Activation Functions.
How to cite: Shehab Abdulhabib Saeed Alzaeemi et al. Analysis of Performance of Various Activation Functions for doing the logic programming in Hopfield Network. Int J Comput Bioinfo In Silico Model. 6(2) 2017: 911-921
ABSTRACT: Hemoglobins (Hbs) are proteins widely distributed in organisms from the three kingdoms of life. Genomic analysis revealed that genes coding for single domain Hbs (SDgbs), flavohemoglobins, globin-coupled sensors and truncated Hbs (tHbs) exist in rhizobial bacteria. Rhizobial Hb sequences have been characterized using bioinformatics methods, however nothing is known about the folding pathway and rate of rhizobial SDgbs and tHbs. Here, we report the prediction of folding pathway and rate for selected rhizobial SDgbs and tHbs using an Average Distance Map method. Results predicted that folding of most of the rhizobial SDgbs and tHbs analyzed in this work occurs throughout the formation of two compact modules, that helix composition for compact modules is rather variable and that protein folding mostly occurs at moderate rate either in the NC or CN direction.
KeyWords: ADM method, molecular dynamics, Rhizobium, unfolding, 2/2, 3/3.
How to cite: Masanari Matsuoka et al. Prediction of folding pathway and rate for selected rhizobial single domain and truncated hemoglobins using an average distance map method. Int J Comput Bioinfo In Silico Model. 6(2) 2017: 922-933
ABSTRACT: It is a major challenge in drug discovery pipeline to identify novel drug molecules against well-known bacterial targets. A potent gram negative bacterium, H. pylori is causative agent of gastrointestinal disorders in humans. Currently, there is no single therapy available to combat infections caused by the bacterium. Peptide deformylase (PDF) is a potent target which caters growth and is considered crucial for survival of bacterium. In assessment of its prominence, development of potent inhibitors against PDF can be considered as vital drug candidates against the bacterium. This paper illustrates a methodology for identification of potent drug molecules against PDF. The dataset comprised of a random sample of 34 experimentally proven inhibitors against protein target along with other drugs with no preliminary evidence against the target enzyme. Machine learning techniques are implemented to develop mathematical models for predicting the drug likeliness of a compound based on its molecular descriptors. Naïve bayes classifier gave better performance compared to other classifiers with an accuracy of 82.42% after ten-fold cross validation. The results suggested that virtual screening approach employing machine learning techniques appears like an encouraging approach for identification of antibacterial compounds. To aid experimentalists for developing novel inhibitors against H. pylori, an R package, PDFPred is deployed as a user friendly tool for prediction of drug likeliness of new compounds against PDF.
KeyWords: Peptide deformylase; naïve bayes classifier; R software; QSAR.
How to cite: Surekha Patil et al. PDFPred: A machine learning prediction tool for evaluation of novel drug candidates against peptide deformylase using naïve bayes classifier. Int J Comput Bioinfo In Silico Model. 6(2) 2017: 934-940