Scholarlink

Alzheimer’s disease (AD), the leading cause of dementia, is marked by progressive neurodegeneration driven by \(\beta \) -amyloid (A \(\beta \) ) plaque accumulation and neurofibrillary tangle formation. \(\beta \) -Secretase 1 (BACE1) plays a critical role in the amyloidogenic pathway by cleaving amyloid precursor protein to produce neurotoxic A \(\beta \) peptides, making it a promising therapeutic target. Emerging evidence demonstrates the transformative potential of large language models (LLMs) in accelerating drug discovery through enhanced bioactivity prediction and virtual screening capabilities. In this study, we introduce EmSVR-BACE, an LLM-SVR hybrid model that accurately predicts BACE1 inhibition ( \(R^2\) = 0.87 \(_{\pm 0.007}\) ; MAE = 0.30 \(_{\pm 0.010}\) ), outperforming existing approaches. To elucidate model interpretability, atom-level attention weights were correlated with docking results, revealing that the chlorine atom enhances binding affinity through hydrophobic and halogen interactions with Tyr71, a key residue for BACE1 selectivity and potency. Furthermore, EmSVR-BACE was applied to screen experimentally validated blood–brain barrier penetrant compounds, emphasizing the need for designing multifunctional molecules with strong BBB permeability, low IC \(_{50}\) values, and optimal ADMET profiles for effective AD therapy.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

EmSVR-BACE: LLM-Based Molecular Embedding for Predicting BACE1 Inhibitors in Alzheimer’s Therapy

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Dhamodharan Ganesan,
  • Pratiti Bhadra

摘要

Alzheimer’s disease (AD), the leading cause of dementia, is marked by progressive neurodegeneration driven by \(\beta \) -amyloid (A \(\beta \) ) plaque accumulation and neurofibrillary tangle formation. \(\beta \) -Secretase 1 (BACE1) plays a critical role in the amyloidogenic pathway by cleaving amyloid precursor protein to produce neurotoxic A \(\beta \) peptides, making it a promising therapeutic target. Emerging evidence demonstrates the transformative potential of large language models (LLMs) in accelerating drug discovery through enhanced bioactivity prediction and virtual screening capabilities. In this study, we introduce EmSVR-BACE, an LLM-SVR hybrid model that accurately predicts BACE1 inhibition ( \(R^2\) = 0.87 \(_{\pm 0.007}\) ; MAE = 0.30 \(_{\pm 0.010}\) ), outperforming existing approaches. To elucidate model interpretability, atom-level attention weights were correlated with docking results, revealing that the chlorine atom enhances binding affinity through hydrophobic and halogen interactions with Tyr71, a key residue for BACE1 selectivity and potency. Furthermore, EmSVR-BACE was applied to screen experimentally validated blood–brain barrier penetrant compounds, emphasizing the need for designing multifunctional molecules with strong BBB permeability, low IC \(_{50}\) values, and optimal ADMET profiles for effective AD therapy.