Abstract
The issue of cancer remains a significant health challenge in the world because it causes serious mortalities. Despite the different traditional therapies and wet-laboratory technologies, which can be employed to treat cancerous cells, these can be faced with several setbacks, including extreme costs and severe side effects. Peptides have become a growing focus of interest over the past few years due to the high selectivity, reliable targeting ability, and low toxicity of peptides. After developing the existing computational tools, we offered a reasonable and functional framework, BT-TransACP, to the appropriate forecast of anticancer peptides. In this framework handles mathematically coded peptide sequences through the aid of an consideration-based ProtBERT-BFD encoder to encode semantics data with the addition of CTDT-derived structure features. A suboptimal set of properties of the integrated representation is then chosen based on binary tree growth (BTG) algorithm, which is a k-nearest neighbor algorithm. Our suggested framework is the BT-TransACP, a Binary Tree-Growth Transformer framework with a hybrid deep learning model to predict anticancer peptides. The fined feature vector is applied and a hybrid deep learning architecture that consists of CNN and LSTM layers are trained. The training accuracy of the model was 95.33% and the AUC was 0.97 with BT-TransACP model. The model was also tested on three separate datasets with the accuracies of 94.92, 92.26 and 91.16, respectively, to test its generalization capability. These findings indicate the high efficacy and effectiveness of BT-TransACP framework, which justifies the importance of the framework in research by scholars and in the development of pharmaceutical drugs.