Home // AIHealth 2025, The Second International Conference on AI-Health // View article
Authors:
Neel Banga
Keywords: Alzheimer's, Beta-Secretase 1, Machine Learning, Transformer model, Drug Discovery
Abstract:
Alzheimer's is a brain disorder that disproportionately affects older adults with its primary symptom being severe dementia. Worldwide, over 55 million people have Alzheimer's, with 6.7 million affected individuals living in the USA. Current methods to mitigate the effects of Alzheimer's are insufficient with most drugs (e.g., Memantine, Donepezil, Rivastigmine, etc.) being inconsistent while also causing heavy side effects. In order to address these issues, more drugs need to be tested for viability. To speed up the process, this research proposes AI-based models that can potentially detect which drugs will be able to effectively inhibit the crux of the Alzheimer's pathway, an enzyme named Beta Secretase 1. This study documented the investigation of four AI models—K-Nearest Neighbors (KNN), Random Forest, ChemBERTa, and PubChem10M—and their ability to predict drug efficacy for inhibiting BACE1, a vital target in the Alzheimer's Disease (AD) pathway. These models were trained on the ChEMBL4822 database. The KNN and RandomForest models were traditional descriptor-based models whereas the ChemBERTa and PubChem10M models were fine-tuned transformers. The KNN model showed a strong training performance of (R² = 0.6092); this score stayed consistent in the testing phase (R² = 0.6210). While having a lower score, the RandomForest model displayed similar consistency in the training (R² = 0.5651) and testing phase (R² = 0.5605). The ChemBERTa model showed significant improvement from the training phase (R² = 0.2641) to the testing one (R² = 0.6433), indicating high generalization potential. Similarly, the PubChem10M model exhibited large growth from the training (R² = 0.2641) to the testing phase (R² = 0.6194). These results highlight the unique strengths of each model and underscore the promising role of AI in AD drug discovery. Future work on the refinement and integration of these models could lead to more effective therapeutic agents for AD.
Pages: 29 to 34
Copyright: Copyright (c) IARIA, 2025
Publication date: March 9, 2025
Published in: conference
ISBN: 978-1-68558-247-0
Location: Lisbon, Portugal
Dates: from March 9, 2025 to March 13, 2025