Abstract
Rationale: Increased levels of advanced glycation end products (AGEs) have been observed in the brain tissues of patients with Alzheimer's disease (AD). Methylglyoxal (MGO) is a potent precursor of AGEs. To date, there have been no reports of utilizing deep learning (DL) technologies to target MGO scavengers for the development of AD therapeutics. Therefore, DL-driven approaches may play a crucial role in identifying potential MGO scavengers and candidates for Alzheimer's treatment. Methods: We developed "DeepMGO," a novel DL-based MGO scavenging activity prediction model, trained on 2,262 MGO scavenging activity assays from 660 compounds. Using this approach, we identified and validated TP-41 as a potential MGO scavenger in a mouse model of memory impairment. Results: DeepMGO demonstrated robust predictive performance and identified novel compounds with high MGO scavenging activity. TP-41 ameliorated depression symptoms and memory deficits in mouse models. Conclusions: Using DeepMGO, we identified TP-41 as a potential therapeutic agent for AD.
Keywords:
Alzheimer's disease; deep learning; drug discovery; memory impairment; methylglyoxal.