TAGMol: Target-Aware Gradient-guided Molecule Generation

Vineeth Dorna,D. Subhalingam,Keshav Kolluru,Shreshth Tuli,Mrityunjay Singh,Saurabh Singal,N. M. Anoop Krishnan,Sayan Ranu
2024-06-03
Abstract:3D generative models have shown significant promise in structure-based drug design (SBDD), particularly in discovering ligands tailored to specific target binding sites. Existing algorithms often focus primarily on ligand-target binding, characterized by binding affinity. Moreover, models trained solely on target-ligand distribution may fall short in addressing the broader objectives of drug discovery, such as the development of novel ligands with desired properties like drug-likeness, and synthesizability, underscoring the multifaceted nature of the drug design process. To overcome these challenges, we decouple the problem into molecular generation and property prediction. The latter synergistically guides the diffusion sampling process, facilitating guided diffusion and resulting in the creation of meaningful molecules with the desired properties. We call this guided molecular generation process as TAGMol. Through experiments on benchmark datasets, TAGMol demonstrates superior performance compared to state-of-the-art baselines, achieving a 22% improvement in average Vina Score and yielding favorable outcomes in essential auxiliary properties. This establishes TAGMol as a comprehensive framework for drug generation.
Biomolecules,Artificial Intelligence,Machine Learning
What problem does this paper attempt to address?
The paper focuses on the problem of molecular generation in Structure-Based Drug Design (SBDD), specifically how to create effective ligands for specific targets. Existing algorithms typically only consider the binding affinity between the ligand and the target, while ignoring other important properties in the drug discovery process, such as drug similarity and synthesizability. To address this issue, the paper proposes a method called Target Aware Gradient-guided Molecule Generation (TAGMOL) that decomposes the problem into molecule generation and property prediction. In TAGMOL, a time-dependent guidance model is first trained to predict properties of inputs with similar noise levels. Then, the gradient-guided diffusion sampling process is utilized using the guidance model to ensure that the generated molecules have the desired properties. This approach overcomes the limitations of traditional autoregressive models and leverages the advantages of diffusion models to effectively generate molecules with high binding affinity. Experimental results in the paper demonstrate that TAGMOL outperforms existing state-of-the-art methods on benchmark datasets, with an average improvement of 22% in Vina scores, and achieves good results in key auxiliary properties. Additionally, TAGMOL considers multi-objective guidance to adapt to real-world scenarios with multiple attribute constraints by training separate guidance models for each attribute to effectively guide the diffusion process. In conclusion, the paper aims to address the limitation of drug generation models focusing solely on optimizing binding activity. It proposes a comprehensive framework, which simultaneously optimizes target awareness and molecular properties in 3D space, thereby improving the efficiency and effectiveness of drug discovery.