Enhancing Drug-Target Interaction Predictions Using a Divisive Computational Framework

Computational prediction of drug-target interactions (DTIs) is crucial for drug discovery. However, the sparse distribution of DTIs and the imbalance in the number of interactions among targets pose challenges. This study proposes a divisive computational framework. Firstly, it includes a novel preprocessing algorithm that adjusts the interaction matrix based on the number of interactions of a target and its neighbors, enhancing DTI predictions for targets with fewer interactions. Additionally, a new divisive computational testing method is introduced, which evaluates targets with similar numbers of interactions separately, ensuring that the results are not disproportionately influenced by targets with a large number of interactions. Furthermore, a weighted global testing method is proposed to provide a more comprehensive assessment of the enhanced prediction capabilities, which reduces the negative impact of low-interaction targets on the overall evaluation and offers a more balanced perspective on the algorithm's effectiveness. Experimental results demonstrate the efficacy of the proposed framework, where the means of AUCs in the divisive computational framework are respectively 9.45%, 10.64%, 4.21%, 7.04%, 3.67%, and 6.50% higher than those in the traditional framework on six DTI datasets.