Determination of anti-coronavirus peptides by protein coding methods

Pathogens settle on living beings and leave their own proteins in the cell nucleus of the host, disrupting the protein structure. The deteriorated protein cannot perform its functions and the biological functioning of the living being is disrupted. Epidemics occur when the pathogen settles and multiplies on the living being. Today, the Covid-19 disease, which has turned into a major epidemic, causes mass casualties around the world. The discovery of the vaccine against coronavirus could not give definite results in the prevention of the Covid-19 disease. Hence, drug development studies go on all over the world. Bioinformatics and machine learning-based studies make a great contribution to drug development for the Covid-19. Anti-coronavirus (anti-CoV) peptides are a kind of therapeutic agent that inhibits the coronavirus and eliminates the harms of the coronavirus. Prediction of anti-CoV peptide sequences with machine learning methods is very important for the treatment of coronavirus disease and drug development steps. However, distinguishing anti-CoV peptides from other peptides requires much more cost and time consuming in vivo and in vitro. For this reason, it is more advantageous to model and predict in silico environment with machine learning-based studies. In this thesis, it is aimed to digitize protein data with protein coding methods so that anti-CoV peptide sequences, which are therapeutic agents against Covid-19, can be predicted with high accuracy. In this regard, two new protein coding methods named 2gBLO and 2gBLOTVD have been developed. Anti-CoV peptide dataset positive class, antimicrobial (AMP), non-antimicrobial (non-AMP) and non-antiviral (non-AVP) three negative class peptide datasets were used. These protein data were combined with each negative dataset separately to be positive anti-CoV, and datasets with three negative and positive classes were obtained. Data sets were digitized with the developed protein coding methods. The 2gBLO method is obtained by updating the amino acid pair in each coded sequence using the digram method by multiplying the displacement value in the BLOSUM62 matrix. In the 2gBLOTVD method, the number of common physicochemical properties in Taylor's Venn diagram is multiplied by the value obtained from the 2gBLO method. Due to the small sample size in the positive class anti-CoV peptide dataset, unstable datasets were subjected to class balancing in order to increase their performance values. In addition, in order to increase the effect of classification results, discretization method is applied to performance metrics, Naive Bayes, Bayesian Networks, Random Forest, Linear Support Vector Machines (Linear SVM), Radial Support Vector Machines (Radial SVM), k-Nearest Neighborhood (k-NN), Multi-Layer Perceptron (MLA) and Logistic Regression classifier algorithms. According to the experimental results obtained; protein coding methods developed on classifier algorithms give higher performance than hitherto protein coding methods that are prominent. Keywords: Covid-19, Therapeutic agent, Feature encoding, Classification, Machine learning