Scholarlink

<p>SARS-CoV-2 is still causing new deaths daily and new variants are appearing. Thus, the a need to identify potential drug-like inhibitors. QSAR and molecular docking were elaborated. The integration of soft computing techniques ensures precise and reliable predictions, accelerating drug discovery significantly. This study delves into soft computing applications in drug design through the use of the dragonfly algorithm, endorsed with ADMET properties and molecular docking studies. Thus, Bayesian Regularization-Backpropagation Neural Network BR-BPNN, Convolutional Neural Network CNN, Support Vector Regression SVR, and Dragonfly Algorithm DA were used because of their advantages over other algorithms. The mechanistic interpretation of the 10 selected descriptors revealed that the spectral diameter of the adjacency matrix and the unsubstituted benzenes had the most positive impact. The nonaromatic conjugated carbons and the number of neighboring atoms at the radius 7 from the chiral center had the most negative effect. The statistical validation of the three models showed an outperformance of the DA-SVR with determination coefficient R<sup>2</sup> = 0.92, coefficient of cross-validation Q<sup>2</sup> = 0.92, and root mean square error RMSE = 0.21. Henceforth, the present DA-SVR model is mechanistic and correlative and can be used for further computer drug design studies to develop 3CLpro inhibitors. This study also suggests that too complex (DA-CNN) or too simple (BR-BPNN) models are less likely to predict accurately the activity of the considered dataset. Molecules 19 (− 7.8 kcal/mole) and 15 (− 7.5 kcal/mol) were the most affine to the binding site among 24 selected drug-like molecules with 1 kcal/mol difference less than the reference molecule PF-07321332, with conventional Hydrogen bonds HBs, pi-donor HBs, halogen, and alkyl bonds. They share triazole and 1H-isoindole fragments with fluorine in the 1H-isondole in molecule 15. The results of this study pave the way for soft computing techniques enhancement of QSAR-based synthesis and analog generation with respect to molecules 19 and 15 and the structural features indicated and predict their IC<sub>50</sub> using the QSAR model to identify de novo potential inhibitors that can undergo wet-lab experiments.</p>

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

Ligand-based and structure-based approaches for the identification of potential SARS-CoV-2 3CLpro inhibitors

  • 期刊论文

  • 发表日期：

    2025年4月1日

查看全文
  • Achouak Madani,
  • Othmane Benkortbi,
  • Maamar Laidi,
  • Cherif Si-Moussa,
  • Naomie Salim

摘要

SARS-CoV-2 is still causing new deaths daily and new variants are appearing. Thus, the a need to identify potential drug-like inhibitors. QSAR and molecular docking were elaborated. The integration of soft computing techniques ensures precise and reliable predictions, accelerating drug discovery significantly. This study delves into soft computing applications in drug design through the use of the dragonfly algorithm, endorsed with ADMET properties and molecular docking studies. Thus, Bayesian Regularization-Backpropagation Neural Network BR-BPNN, Convolutional Neural Network CNN, Support Vector Regression SVR, and Dragonfly Algorithm DA were used because of their advantages over other algorithms. The mechanistic interpretation of the 10 selected descriptors revealed that the spectral diameter of the adjacency matrix and the unsubstituted benzenes had the most positive impact. The nonaromatic conjugated carbons and the number of neighboring atoms at the radius 7 from the chiral center had the most negative effect. The statistical validation of the three models showed an outperformance of the DA-SVR with determination coefficient R2 = 0.92, coefficient of cross-validation Q2 = 0.92, and root mean square error RMSE = 0.21. Henceforth, the present DA-SVR model is mechanistic and correlative and can be used for further computer drug design studies to develop 3CLpro inhibitors. This study also suggests that too complex (DA-CNN) or too simple (BR-BPNN) models are less likely to predict accurately the activity of the considered dataset. Molecules 19 (− 7.8 kcal/mole) and 15 (− 7.5 kcal/mol) were the most affine to the binding site among 24 selected drug-like molecules with 1 kcal/mol difference less than the reference molecule PF-07321332, with conventional Hydrogen bonds HBs, pi-donor HBs, halogen, and alkyl bonds. They share triazole and 1H-isoindole fragments with fluorine in the 1H-isondole in molecule 15. The results of this study pave the way for soft computing techniques enhancement of QSAR-based synthesis and analog generation with respect to molecules 19 and 15 and the structural features indicated and predict their IC50 using the QSAR model to identify de novo potential inhibitors that can undergo wet-lab experiments.