Structure-Based Drug Design via 3D Molecular Generative Pre-training and Sampling (2402.14315v2)

Abstract: Structure-based drug design aims at generating high affinity ligands with prior knowledge of 3D target structures. Existing methods either use conditional generative model to learn the distribution of 3D ligands given target binding sites, or iteratively modify molecules to optimize a structure-based activity estimator. The former is highly constrained by data quantity and quality, which leaves optimization-based approaches more promising in practical scenario. However, existing optimization-based approaches choose to edit molecules in 2D space, and use molecular docking to estimate the activity using docking predicted 3D target-ligand complexes. The misalignment between the action space and the objective hinders the performance of these models, especially for those employ deep learning for acceleration. In this work, we propose MolEdit3D to combine 3D molecular generation with optimization frameworks. We develop a novel 3D graph editing model to generate molecules using fragments, and pre-train this model on abundant 3D ligands for learning target-independent properties. Then we employ a target-guided self-learning strategy to improve target-related properties using self-sampled molecules. MolEdit3D achieves state-of-the-art performance on majority of the evaluation metrics, and demonstrate strong capability of capturing both target-dependent and -independent properties.

Enhancing Structure-Based Drug Design with 3D Molecular Generative Pre-Training and Sampling

Introduction to MolEdit3D

Structure-based drug discovery (SBDD) presents a critical area within pharmaceutical research, where the goal is to design ligands (potential drug molecules) with high affinity towards target biomolecules, leveraging the knowledge of the 3D structure of the target. The paper introduces MolEdit3D, a novel approach that combines 3D molecular generation with optimization frameworks to address the challenges faced by current SBDD methods. MolEdit3D leverages a pre-trained 3D graph editing model, utilizing fragments as building blocks for molecule construction. The model is then fine-tuned via a target-guided self-learning strategy to enhance target-related properties. The paper details substantial improvements over previous methods in generating molecules with higher binding affinities and maintaining drug-like properties.

Overview of Existing SBDD Methods

Existing SBDD methods are broadly categorized into two types: conditional generation methods and optimization-based approaches. Conditional generation methods are constrained by the limited availability and quality of 3D target-ligand complex data, affecting their ability to learn both target-dependent and -independent properties. Optimization-based approaches, such as those using molecular docking, tend to operate on 2D molecular representations, which introduces a misalignment with the inherently 3D nature of molecular interactions. The paper critiques these existing methodologies for their limitations in efficiently exploring the molecular space and accurately approximating real-world biochemical interactions.

Key Contributions of MolEdit3D

MolEdit3D's development and implementation introduce several key innovations in the field of SBDD:

• The introduction of a 3D molecular graph editing model allows for direct manipulation and generation of molecules in three-dimensional space, improving the relevance and accuracy of generated ligand structures.
• Pre-training the model with a significant dataset of 3D molecules enables it to recognize and reproduce general drug-like properties, making the generated molecules more viable as potential drugs.
• The application of a target-guided self-learning strategy, which leverages self-generated samples to fine-tune the model, enhances its capacity to produce molecules with higher affinity for specific targets.
• MolEdit3D demonstrates leading performance across various metrics, notably in producing molecules with high binding affinity, while also ensuring the molecules exhibit desirable drug-like and synthesizable properties.

Theoretical and Practical Implications

MolEdit3D's approach to incorporating 3D generation and optimization within SBDD processes holds significant implications for the field. Practically, it offers a more efficient pathway to identify potential drug candidates by directly generating and optimizing molecules within the spatial confines of target binding sites. This direct approach could significantly shorten drug discovery timelines and increase the specificity and efficacy of resulting compounds. Theoretically, MolEdit3D's success supports the hypothesis that closer alignment of the model's action space with the natural three-dimensional interaction space of molecules can lead to more accurate and effective drug design methodologies.

Future Prospects in AI and Drug Discovery

Looking forward, the integration of 3D molecular generation and machine learning optimization presents a fertile ground for innovation in drug discovery. Potential areas for further research and development include the exploration of more complex molecular interaction models, incorporation of dynamic molecular simulations, and the expansion of self-learning strategies to incorporate more nuanced biochemical and pharmacological properties. As computational power and machine learning algorithms continue to evolve, methods like MolEdit3D stand at the forefront of transforming drug discovery, rendering it more precise, efficient, and rooted in the complex realities of molecular biology.

MolEdit3D's development represents an important step forward in the application of generative AI and machine learning to structure-based drug design. It not only addresses current limitations within SBDD methodologies but also opens up new avenues for research and application in the quest for more effective medicinal compounds.