Development of new feature encoding methods in prediction of anticancer peptides

Cancer is one of the diseases that are likely to cause death. The methods used in known cancer therapies destroy tumor cells while tumor uninfected cells are also affected during treatment. In recent years, promising peptide-based strategies have been used in various tumor therapies. In this respect, anti-cancer peptides are in the process of development. With the emergence of anti-cancer peptides, only tumor cells can be destroyed without damaging healthy cells. Thus, as anti-cancer peptides are cationic in nature, they can interfere with the anionic cell membrane components of cancer cells, and in particular eliminate cancer cells. In addition, anti-cancer peptides do not physically damage the body and are more effective and reliable than artificial drugs. Detection of anti-cancer peptides is an important step in the treatment of diseases and drug development. However, distinguishing anti-cancer peptides from other peptides is costly and difficult. It is appropriate to estimate peptide sequences by statistical methods. For this reason, it is more advantageous to estimate the disease in computer environment with machine learning based studies. When studies on anti-cancer peptides are examined, it can be seen that sequence based methods give more effective results. In our proposed new method, 2-grams feature extraction method and Taylor Venn Diagram were used to extract the properties of the peptides on the dataset published by Universal Protein Resource and the values of the extracted properties were updated using Blosum 30 matrix. In order to increase the effect of the results of the study, Fisher's multiple class linear discriminant analysis method was used to reduce the size of the features and classified them using Dual Perturb and Combine Tree, Multilayer Perceptron, Random Forest, Naive Bayes, Bayes Network Support Vector Machines, K-Nearest Neighbor, Adaboost, Bagging, KStar and Logistic classifier algorithms. The experimental method has been compared and analyzed with the existing methods. As a result of the analysis, the method we developed for the detection of anti-cancer peptides showed the highest performance in the literature compared to the studies performed on the same dataset.