18, although the model can accurately position the nth atom to create a benzene ring when the preceding n − 1 carbon atoms are already in the same plane 16, accurate placement of the initial atoms is often problematic because of insufficient context information, resulting in unrealistic fragments. It can easily get stuck in local optima during the initial stages of molecule generation and may accumulate errors introduced at each step of the sampling process. An autoregressive sampling method has its inherent limitations. Although these methods can generate molecules with 3D conformations, they share some common drawbacks: (1) the generated molecules often contain problematic, non-drug-like or not synthetically available substructures such as very large rings (rings containing seven or more atoms) and honeycomb-like arrays of parallel, juxtaposed rings (2) problematic topology: the generated molecules often contain an excessive number of rings or none at all. These GNN models use an autoregressive generation process that linearizes a molecule graph into a sequence of sampling decisions. Some studies 14, 15, 16, 17 proposed to represent pocket and molecule as 3D graphs and used graph neural networks (GNNs) for encoding and decoding. For example, methods using 3D convolutional neural networks 13 are used to capture 3D inductive bias, but they still struggle to convert atomic density grids into discrete molecules. The increase of 3D protein–ligand complex structures data 10 and advances in geometric deep learning have paved the way for artificial intelligence algorithms to directly design molecules with 3D binding poses 11, 12. However, both representations disregard three-dimensional (3D) spatial interactions, rendering them suboptimal for target-aware molecule generation. Earlier molecular generative models relied on either molecular string representations 2, 3, 4, 5 or graph representations 6, 7, 8, 9. De novo molecule generation using artificial intelligence has recently gained attention as a tool for drug discovery. Structure-based drug design, which involves designing molecules that can specifically bind to a desired target protein, is a fundamental and challenging drug discovery task 1. Lingo3DMol outperformed state-of-the-art methods in terms of drug likeness, synthetic accessibility, pocket binding mode and molecule generation speed. The Directory of Useful Decoys-Enhanced dataset was used for evaluation. Lingo3DMol can efficiently traverse drug-like chemical spaces, preventing the formation of unusual structures. Additionally, we trained a separate non-covalent interaction predictor to provide essential binding pattern information for the generative model. A new molecular representation, the fragment-based simplified molecular-input line-entry system with local and global coordinates, was developed to assist the model in learning molecular topologies and atomic spatial positions. To address these challenges, we introduce Lingo3DMol, a pocket-based 3D molecule generation method that combines language models and geometric deep learning technology. Generative models for molecules based on sequential line notation (for example, the simplified molecular-input line-entry system) or graph representation have attracted an increasing interest in the field of structure-based drug design, but they struggle to capture important three-dimensional (3D) spatial interactions and often produce undesirable molecular structures.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |