Machine learning applications for the topology prediction of transmembrane beta-barrel proteins

Grimaldi, Cédric Maxime (2022) Machine learning applications for the topology prediction of transmembrane beta-barrel proteins. Doctoral thesis, London Metropolitan University.

Abstract

The research topic for this PhD thesis focuses on the topology prediction of beta-barrel transmembrane proteins. Transmembrane proteins adopt various conformations that are about the functions that they provide. The two most predominant classes are alpha-helix bundles and beta-barrel transmembrane proteins. Alpha-helix proteins are present in larger numbers than beta-barrel transmembrane proteins in structure databases. Therefore, there is a need to find computational tools that can predict and detect the structure of beta-barrel transmembrane proteins. Transmembrane proteins are used for active transport across the membrane or signal transduction. Knowing the importance of their roles, it becomes essential to understand the structures of the proteins. Transmembrane proteins are also a significant focus for new drug discovery. Transmembrane beta-barrel proteins play critical roles in the translocation machinery, pore formation, membrane anchoring, and ion exchange. In bioinformatics, many years of research have been spent on the topology prediction of transmembrane alpha-helices. The efforts to TMB (transmembrane beta-barrel) proteins topology prediction have been overshadowed, and the prediction accuracy could be improved with further research. Various methodologies have been developed in the past to predict TMB proteins topology. Methods developed in the literature that are available include turn identification, hydrophobicity profiles, rule-based prediction, HMM (Hidden Markov model), ANN (Artificial Neural Networks), radial basis function networks, or combinations of methods. The use of cascading classifier has never been fully explored. This research presents and evaluates approaches such as ANN (Artificial Neural Networks), KNN (K-Nearest Neighbors, SVM (Support Vector Machines), and a novel approach to TMB topology prediction with the use of a cascading classifier. Computer simulations have been implemented in MATLAB, and the results have been evaluated. Data were collected from various datasets and pre-processed for each machine learning technique. A deep neural network was built with an input layer, hidden layers, and an output. Optimisation of the cascading classifier was mainly obtained by optimising each machine learning algorithm used and by starting using the parameters that gave the best results for each machine learning algorithm. The cascading classifier results show that the proposed methodology predicts transmembrane beta-barrel proteins topologies with high accuracy for randomly selected proteins. Using the cascading classifier approach, the best overall accuracy is 76.3%, with a precision of 0.831 and recall or probability of detection of 0.799 for TMB topology prediction. The accuracy of 76.3% is achieved using a two-layers cascading classifier. By constructing and using various machine-learning frameworks, systems were developed to analyse the TMB topologies with significant robustness. We have presented several experimental findings that may be useful for future research. Using the cascading classifier, we used a novel approach for the topology prediction of TMB proteins.

Documents
7931:41287
[img]
Preview
GrimaldiCédric-Maxime_PhD-thesis.pdf - Published Version

Download (7MB) | Preview
Details
Record
Statistics

Downloads

Downloads per month over past year



Downloads each year

View Item View Item