Abstract:Generative models for molecules based on sequential line notation (e.g. SMILES) or graph representation have attracted an increasing interest in the field of structure-based drug design, but they struggle to capture important 3D spatial interactions and often produce undesirable molecular structures. To address these challenges, we introduce Lingo3DMol, a pocket-based 3D molecule generation method that combines language models and geometric deep learning technology. A new molecular representation, fragment-based SMILES with local and global coordinates, was developed to assist the model in learning molecular topologies and atomic spatial positions. Additionally, we trained a separate noncovalent interaction predictor to provide essential binding pattern information for the generative model. Lingo3DMol can efficiently traverse drug-like chemical spaces, preventing the formation of unusual structures. The Directory of Useful Decoys-Enhanced (DUD-E) dataset was used for evaluation. Lingo3DMol outperformed state-of-the-art methods in terms of drug-likeness, synthetic accessibility, pocket binding mode, and molecule generation speed.

What problem does this paper attempt to address?

### Problems Addressed by the Paper The paper aims to address the issues encountered in generating 3D molecules in structured drug design. Specifically, existing methods based on sequence representations (such as SMILES) or graph representations fall short in capturing important 3D spatial interactions and often produce suboptimal molecular structures. To tackle these issues, the authors propose Lingo3DMol, a pocket-based 3D molecular generation method that combines language models and geometric deep learning techniques. ### Main Contributions 1. **Novel Molecular Representation (FSMILES)**: A new molecular representation method called FSMILES is introduced, which combines local and global coordinates, enabling the generated 3D molecules to have reasonable 3D conformations and 2D topological structures. 2. **3D Molecular Denoising Pretraining Method**: An independent non-covalent interaction (NCI)/anchor point prediction model is developed to help overcome the problem of limited data and identify potential NCI binding sites. 3. **Outperformance of Existing Methods**: Lingo3DMol outperforms current state-of-the-art methods on various metrics, including drug similarity, synthetic accessibility, and pocket binding patterns. ### Experimental Results and Discussion 1. **Molecular Geometry Evaluation**: By comparing with reference molecules, it is found that Lingo3DMol has the lowest Jensen-Shannon divergence (JSD) score in atomic distance distribution and performs better in ring size, reducing the probability of generating macrocycles larger than 7 atoms. 2. **Molecular Properties and Binding Mode Evaluation**: Using Glide scoring to evaluate the generated molecules, Lingo3DMol excels in min-in-place GlideSP scores and RMSD vs. low-energy conformers. Additionally, it performs well in generating molecules similar to known active compounds. 3. **Information Leakage Analysis**: An analysis of information leakage issues in benchmark models reveals that Lingo3DMol consistently outperforms Pocket2Mol under different levels of information leakage. ### Conclusion Lingo3DMol demonstrates excellent performance in generating drug-like 3D molecules and surpasses existing methods on multiple key metrics, showcasing its potential in drug discovery and design. Future research directions include further optimizing the capture of non-covalent interactions and improving the model's equivariance.

Generation of 3D Molecules in Pockets via Language Model

Generation of 3D molecules in pockets via a language model

3D Molecular Pocket-based Generation with Token-only Large Language Model

Advances in Deep Learning-Based 3D Molecular Generative Models

3DSMILES-GPT: 3D Molecular Pocket-based Generation with Token-only Large Language Model

Fragment and Geometry Aware Tokenization of Molecules for Structure-Based Drug Design Using Language Models

Molecule Generation For Target Protein Binding with Structural Motifs

RELATION: A Deep Generative Model for Structure-Based De Novo Drug Design

Pocket2Mol: Efficient Molecular Sampling Based on 3D Protein Pockets.

DNMG: Deep Molecular Generative Model by Fusion of 3D Information for De Novo Drug Design.

MOLUCINATE: A Generative Model for Molecules in 3D Space

In-Pocket 3D Graphs Enhance Ligand-Target Compatibility in Generative Small-Molecule Creation

Learning to design drug-like molecules in three-dimensional space using deep generative models

An Equivariant Generative Framework for Molecular Graph-Structure Co-Design

3D Molecular Generation via Virtual Dynamics

Domain-Agnostic Molecular Generation with Chemical Feedback

PocketFlow is a data-and-knowledge-driven structure-based molecular generative model

Learning Subpocket Prototypes for Generalizable Structure-based Drug Design

Generating 3D Molecular Structures Conditional on a Receptor Binding Site with Deep Generative Models

Learning on topological surface and geometric structure for 3D molecular generation

LS-MolGen: Ligand-and-Structure Dual-Driven Deep Reinforcement Learning for Target-Specific Molecular Generation Improves Binding Affinity and Novelty.