Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Zero-Shot 3D Drug Design by Sketching and Generating

Siyu Long, Yi Zhou, Xinyu Dai, Hao Zhou

TL;DR

This study proposes the zero-shot drug design method DESERT, which splits the design process into two stages: sketching and generating, and bridges them with the molecular shape to utilize the large-scale molecular database to reduce the need for experimental data and docking simulation.

Abstract

Drug design is a crucial step in the drug discovery cycle. Recently, various deep learning-based methods design drugs by generating novel molecules from scratch, avoiding traversing large-scale drug libraries. However, they depend on scarce experimental data or time-consuming docking simulation, leading to overfitting issues with limited training data and slow generation speed. In this study, we propose the zero-shot drug design method DESERT (Drug dEsign by SkEtching and geneRaTing). Specifically, DESERT splits the design process into two stages: sketching and generating, and bridges them with the molecular shape. The two-stage fashion enables our method to utilize the large-scale molecular database to reduce the need for experimental data and docking simulation. Experiments show that DESERT achieves a new state-of-the-art at a fast speed.

Zero-Shot 3D Drug Design by Sketching and Generating

TL;DR

This study proposes the zero-shot drug design method DESERT, which splits the design process into two stages: sketching and generating, and bridges them with the molecular shape to utilize the large-scale molecular database to reduce the need for experimental data and docking simulation.

Abstract

Drug design is a crucial step in the drug discovery cycle. Recently, various deep learning-based methods design drugs by generating novel molecules from scratch, avoiding traversing large-scale drug libraries. However, they depend on scarce experimental data or time-consuming docking simulation, leading to overfitting issues with limited training data and slow generation speed. In this study, we propose the zero-shot drug design method DESERT (Drug dEsign by SkEtching and geneRaTing). Specifically, DESERT splits the design process into two stages: sketching and generating, and bridges them with the molecular shape. The two-stage fashion enables our method to utilize the large-scale molecular database to reduce the need for experimental data and docking simulation. Experiments show that DESERT achieves a new state-of-the-art at a fast speed.
Paper Structure (16 sections, 4 equations, 11 figures, 3 tables)

This paper contains 16 sections, 4 equations, 11 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Proposed Method: Desert
  3. Zero-Shot Pipeline
  4. Sketching Molecular Shapes
  5. Generating 3D Molecules by Shape2Mol
  6. Encoder: Voxelized Shape
  7. Decoder: Linearized Molecule
  8. Training & Decoding
  9. Results and Discussions
  10. Experiments
  11. Main Results
  12. Comparison with Related Shape-based Models
  13. Ablation Study of Generating
  14. Ablation Study of Sketching
  15. Apply Desert to More Protein Targets
  16. ...and 1 more sections

Figures (11)

  • Figure 1: When a drug is binding to a pocket, its shape does not change too much (with an RMSD less than $1.391$Å) and is complementary to the pocket.
  • Figure 2: Desert splits the whole drug design process into two stages: sketching the shape and generating the molecules.
  • Figure 3: Overview of Desert. In Figure (a), we use massive unbound molecules$^2$ to train the Shape2Mol, which includes two sketching variants. In Figure (b) and (c), we use the Shape2Mol to generate 3D molecules to fill in the given shapes. In Figure (b) where existing drugs are available, we treat their shapes as desired molecule shapes (ligand-based). In Figure (c) where only protein pockets are given, we heuristically sample reasonable molecular shapes from the pockets (pocket-based).
  • Figure 4: A 2D illustration of sampling molecular shapes from pockets. The 2D molecule in the figure represents a potential drug that fits the molecular shape.
  • Figure 5: The architecture of Shape2Mol.
  • ...and 6 more figures