Molecular Sub-structural Pattern Discovery using Transformer-CNN-OMBO Algorithm for Drug-target interaction (DTI) Prediction

Predicting drug-target interactions (DTIs) is critical in drug discovery and repositioning because it identifies the reaction between a chemical molecule and a target protein. Currently, researchers use supervised learning methods based on machine learning (ML) to handle the prediction issue. Predicting drug-target interactions is hindered by an imbalanced dataset. Substructures of DTI were not included in existing molecular representation learning algorithms, which resulted in a lack of precision and difficulty in articulating. Transformer-CNN-OMBO technique for DTI interaction prediction is proposed in this study. It is first necessary to transform input drug and protein datasets to explicit sub-structures using an Improved Frequent Subsequence Mining (IFSM) technique. Then the sub-structures are sent into a transformer binding phase, which generates a contextual binding for each structure. The neighbourhood interaction was extracted by adding an optimal Convolution Neural Network (CNN) layer to these interactions and predicting the resulting interactions. During the training procedure, the Opposition based Monarch Butterfly Optimization (OMBO) algorithm is used to optimise the hyper parameters of CNN layers. Using the BindngDB dataset, the suggested technique was tested with Transformer+CNN and MPNN+CNN models. The recommended Transformer-CNN-OMBO technique achieves improved accuracy, sensitivity, specificity metrics.