PeerJ · 2026
AI-driven de novo design of BRAF inhibitors with enhanced binding affinity and optimized drug-likeness.
СтатьяНазвание и аннотация: Перевод пока не подготовлен
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Del-Prado-Sánchez Cristina, Llorca-Cardeñosa Ana, Yeste-Mayoral Begoña, Grixolli-Mazzon Stefano, Barros-Centeno María Fernanda, Díaz-Cascajosa Jesús, Antón Jordi, Calzada-Hernández Joan, Mosquera Juan Manuel, Carriqui Sonia, Zacarías Andrea, Morales-Ballús Marta
Аннотация
BACKGROUND: Traditional drug discovery methods, such as high-throughput screening (HTS), are often inefficient and costly, especially in complex areas like oncology. The BRAF V600E mutation is a validated therapeutic target in cancers such as melanoma, thyroid carcinoma, and colorectal cancer. However, existing BRAF inhibitors face challenges like acquired resistance and off-target toxicity. Artificial intelligence (AI) has emerged as a transformative tool for designing novel inhibitors more efficiently.
METHODS: This study employed REINVENT 4, an advanced machine learning (ML) framework using recurrent neural networks and transformer architectures, for targeted generation and property optimization of BRAF V600E inhibitors, integrating reinforcement learning (RL) for drug-likeness optimization and transfer learning (TL) for mutation-specific design. Molecular docking and dynamics simulations were used to evaluate binding affinity and stability.
RESULTS: The AI-driven approach generated 41,721 novel BRAF V600E inhibitor candidates with enhanced drug-likeness (mean Quantitative Estimate of Drug-likeness (QED) score: 0.61 ± 0.17 vs. the training set 0.40 ± 0.13) and predicted inhibitory activity (83.8% with predicted pIC50 > 6). The generated compounds showed a 32% reduction in mean molecular weight (326.8 ± 45.6 g/mol vs. 480.8 ± 84.2 g/mol in the training set) while maintaining inhibitory potency. Pharmacokinetic analysis revealed that 99.7% of generated compounds satisfied Lipinski's Rule of Five criteria, suggesting favorable absorption and distribution profiles. Molecular docking analysis of selected compounds revealed strong binding affinities, with an average free energy of -8.03 ± 1.12 kcal/mol and top-performing compounds reaching -11.5 kcal/mol. Molecular dynamics simulations conducted over 200 ns confirmed complex stability, demonstrating protein backbone RMSD values of 0.35-0.55 nm and ligand RMSD values of 0.086-0.161 nm. Structural novelty assessment using Tanimoto similarity coefficients showed values below 0.45 when compared with FDA-approved BRAF inhibitors (including Sorafenib and Vemurafenib).
DISCUSSION: This work highlights a reproducible, integrated AI-driven workflow demonstration for targeted inhibitor generation. The generated inhibitors exhibit favorable drug-like properties and inhibitory activity, offering a scalable solution for designing safer cancer therapies. Experimental validation is needed to address potential discrepancies between computational predictions and biological behavior.
METHODS: This study employed REINVENT 4, an advanced machine learning (ML) framework using recurrent neural networks and transformer architectures, for targeted generation and property optimization of BRAF V600E inhibitors, integrating reinforcement learning (RL) for drug-likeness optimization and transfer learning (TL) for mutation-specific design. Molecular docking and dynamics simulations were used to evaluate binding affinity and stability.
RESULTS: The AI-driven approach generated 41,721 novel BRAF V600E inhibitor candidates with enhanced drug-likeness (mean Quantitative Estimate of Drug-likeness (QED) score: 0.61 ± 0.17 vs. the training set 0.40 ± 0.13) and predicted inhibitory activity (83.8% with predicted pIC50 > 6). The generated compounds showed a 32% reduction in mean molecular weight (326.8 ± 45.6 g/mol vs. 480.8 ± 84.2 g/mol in the training set) while maintaining inhibitory potency. Pharmacokinetic analysis revealed that 99.7% of generated compounds satisfied Lipinski's Rule of Five criteria, suggesting favorable absorption and distribution profiles. Molecular docking analysis of selected compounds revealed strong binding affinities, with an average free energy of -8.03 ± 1.12 kcal/mol and top-performing compounds reaching -11.5 kcal/mol. Molecular dynamics simulations conducted over 200 ns confirmed complex stability, demonstrating protein backbone RMSD values of 0.35-0.55 nm and ligand RMSD values of 0.086-0.161 nm. Structural novelty assessment using Tanimoto similarity coefficients showed values below 0.45 when compared with FDA-approved BRAF inhibitors (including Sorafenib and Vemurafenib).
DISCUSSION: This work highlights a reproducible, integrated AI-driven workflow demonstration for targeted inhibitor generation. The generated inhibitors exhibit favorable drug-like properties and inhibitory activity, offering a scalable solution for designing safer cancer therapies. Experimental validation is needed to address potential discrepancies between computational predictions and biological behavior.