Guided Multi-objective Generative AI to Enhance Structure-based Drug Design (2405.11785v3)

Abstract: Generative AI has the potential to revolutionize drug discovery. Yet, despite recent advances in deep learning, existing models cannot generate molecules that satisfy all desired physicochemical properties. Herein, we describe IDOLpro, a generative chemistry AI combining diffusion with multi-objective optimization for structure-based drug design. Differentiable scoring functions guide the latent variables of the diffusion model to explore uncharted chemical space and generate novel ligands in silico, optimizing a plurality of target physicochemical properties. We demonstrate our platform's effectiveness by generating ligands with optimized binding affinity and synthetic accessibility on two benchmark sets. IDOLpro produces ligands with binding affinities over 10%-20% better than the next best state-of-the-art method on each test set, producing more drug-like molecules with generally better synthetic accessibility scores than other methods. We do a head-to-head comparison of IDOLpro against a classic virtual screen of a large database of drug-like molecules. We show that IDOLpro can generate molecules for a range of important disease-related targets with better binding affinity and synthetic accessibility than any molecule found in the virtual screen while being over 100x faster and less expensive to run. On a test set of experimental complexes, IDOLpro is the first to produce molecules with better binding affinities than experimentally observed ligands. IDOLpro can accommodate other scoring functions (e.g. ADME-Tox) to accelerate hit-finding, hit-to-lead, and lead optimization for drug discovery.

• The paper introduces IDOLpro, a novel generative AI that combines deep diffusion models with multi-objective optimization for ligand design.
• It achieves over 10% improvement in binding affinity while generating molecules with enhanced synthetic accessibility compared to prior models.
• The research employs iterative refinement and differentiable scoring to accelerate drug discovery and improve practical drug candidate design.

Guided Multi-objective Generative AI for Drug Design

Hello, fellow data enthusiasts! Today, we will dive into a noteworthy AI application in the field of drug discovery. This research presents a novel approach using generative AI to enhance the process of structure-based drug design (SBDD). Let's break it down into digestible bits!

The Premise

Structure-based drug design (SBDD) aims to create molecules (ligands) that fit snugly into a 3D protein pocket to form effective drugs. However, designing these molecules with the desired properties (like high binding affinity and synthesizability) isn't straightforward.

Inverse design, commonly used across various sciences, involves defining a desired set of properties and then figuring out how to create a molecule that meets these requirements. This method relies on two critical steps:

1. Sampling the chemical space.
2. Scoring the molecules based on their properties.

Traditionally, researchers sample from huge databases of molecules, but these databases cover just a tiny fraction of possible chemical space. Generating novel molecules that don't exist in current databases is the way forward, and AI models greatly aid in this quest.

IDOLpro: The New Kid on the Block

The research introduces IDOLpro (Inverse Design of Optimal Ligands for Protein pockets), developed to streamline the drug design process using a combination of deep diffusion models and multi-objective optimization. Here's how IDOLpro stands out:

• Generative Chemistry: It combines generative AI techniques with multi-objective optimization to produce ligands catered to specific proteins.
• Differentiable Scoring: The model uses various scoring functions that assess desirable properties like binding affinity and synthesizability, which are differentiable and integrated into the model.
• Iterative Improvement: Unlike some other models, IDOLpro refines its predictions iteratively to improve molecule properties.

Performance Highlights

To validate IDOLpro's capabilities, the authors tested it on benchmark datasets including CrossDocked and Binding MOAD. These datasets contain protein-ligand pairs used to evaluate models in similar research.

Here's what IDOLpro managed to achieve:

• Binding Affinity: Generated ligands with binding affinities over 10% better than state-of-the-art models.
• Synthetic Accessibility: Produced more synthetically accessible molecules, even while improving their binding affinities.

Table Comparison: When comparing average Vina scores (a metric for binding affinity), IDOLpro outperformed other models substantially. For example, in the Binding MOAD dataset, IDOLpro achieved an average Vina score of -8.48 kcal/mol compared to -7.31 kcal/mol by the next best model.

Practical Implications and Future Directions

Implications:

1. Accelerated Drug Discovery: By generating and screening optimized molecules computationally, drug discovery can move faster and more cost-effectively.
2. Enhanced Molecular Properties: Multi-objective optimization ensures both binding affinity and synthesizability are taken into account, leading to more practical drug candidates.

Future Directions:

1. Additional Properties: Integrate more metrics like toxicity and solubility to ensure even better-quality ligand generation.
2. Broader Applications: While IDOLpro focuses on drugs, similar approaches could benefit materials science and other fields where molecular design is crucial.

In essence, IDOLpro represents a significant step towards making drug discovery not just more efficient but also more effective by utilizing the power of generative AI. As the model continues to evolve, we can anticipate further advancements that will make the dream of rapid, cost-effective drug discovery a reality.

What an exciting time for AI and its myriad applications in the real world!