Multi-perspective neural network for dual drug repurposing in Alzheimer’s disease

In the field of drug discovery, the large-scale prediction of drug-target affinity (DTA) is essential. Despite recent advancements in deep learning enhancing DTA prediction, many single-model approaches do not adequately capture the multifaceted features of drugs and proteins, encompassing sequence, 3D structure, and interaction characteristics. Addressing this, we introduce MNNEL-DTA, a novel multi-perspective neural network ensemble learning approach for drug-target affinity prediction, which integrates multiple deep learning models, each emphasizing different feature perspectives. MNNEL-DTA not only extracts these diverse features but also quantifies the contribution of each, achieving test MSEs of 0.178 and 0.113 on the benchmark datasets Davis and KIBA, respectively. These results signify a relative improvement of 9.2% and 8.9% over the state-of-the-art (SOTA) models. Moreover, we applied MNNEL-DTA to drug repurposing for Alzheimer's disease (AD) and identified Glutethimide as an inhibitor for the target medin and Memantine as a joint agonist for S1R and SYK. Importantly, Memantine is already FDA-approved for AD. Meanwhile, Glutethimide, historically prescribed for insomnia, has demonstrated potential in preliminary studies for AD treatment, particularly targeting the medin protein. The success in identifying the dual drugs as potential therapeutic agents underscores the efficacy of our model in drug repurposing for AD.