Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

S-MolSearch: 3D Semi-supervised Contrastive Learning for Bioactive Molecule Search

Gengmo Zhou, Zhen Wang, Feng Yu, Guolin Ke, Zhewei Wei, Zhifeng Gao

TL;DR

S-MolSearch is the first framework to the authors' knowledge, that leverages molecular 3D information and affinity information in semi-supervised contrastive learning for ligand-based virtual screening, and demonstrates superior performance on widely-used benchmarks LIT-PCBA and DUD-E.

Abstract

Virtual Screening is an essential technique in the early phases of drug discovery, aimed at identifying promising drug candidates from vast molecular libraries. Recently, ligand-based virtual screening has garnered significant attention due to its efficacy in conducting extensive database screenings without relying on specific protein-binding site information. Obtaining binding affinity data for complexes is highly expensive, resulting in a limited amount of available data that covers a relatively small chemical space. Moreover, these datasets contain a significant amount of inconsistent noise. It is challenging to identify an inductive bias that consistently maintains the integrity of molecular activity during data augmentation. To tackle these challenges, we propose S-MolSearch, the first framework to our knowledge, that leverages molecular 3D information and affinity information in semi-supervised contrastive learning for ligand-based virtual screening. Drawing on the principles of inverse optimal transport, S-MolSearch efficiently processes both labeled and unlabeled data, training molecular structural encoders while generating soft labels for the unlabeled data. This design allows S-MolSearch to adaptively utilize unlabeled data within the learning process. Empirically, S-MolSearch demonstrates superior performance on widely-used benchmarks LIT-PCBA and DUD-E. It surpasses both structure-based and ligand-based virtual screening methods for AUROC, BEDROC and EF.

S-MolSearch: 3D Semi-supervised Contrastive Learning for Bioactive Molecule Search

TL;DR

S-MolSearch is the first framework to the authors' knowledge, that leverages molecular 3D information and affinity information in semi-supervised contrastive learning for ligand-based virtual screening, and demonstrates superior performance on widely-used benchmarks LIT-PCBA and DUD-E.

Abstract

Virtual Screening is an essential technique in the early phases of drug discovery, aimed at identifying promising drug candidates from vast molecular libraries. Recently, ligand-based virtual screening has garnered significant attention due to its efficacy in conducting extensive database screenings without relying on specific protein-binding site information. Obtaining binding affinity data for complexes is highly expensive, resulting in a limited amount of available data that covers a relatively small chemical space. Moreover, these datasets contain a significant amount of inconsistent noise. It is challenging to identify an inductive bias that consistently maintains the integrity of molecular activity during data augmentation. To tackle these challenges, we propose S-MolSearch, the first framework to our knowledge, that leverages molecular 3D information and affinity information in semi-supervised contrastive learning for ligand-based virtual screening. Drawing on the principles of inverse optimal transport, S-MolSearch efficiently processes both labeled and unlabeled data, training molecular structural encoders while generating soft labels for the unlabeled data. This design allows S-MolSearch to adaptively utilize unlabeled data within the learning process. Empirically, S-MolSearch demonstrates superior performance on widely-used benchmarks LIT-PCBA and DUD-E. It surpasses both structure-based and ligand-based virtual screening methods for AUROC, BEDROC and EF.
Paper Structure (31 sections, 3 theorems, 16 equations, 4 figures, 5 tables)

This paper contains 31 sections, 3 theorems, 16 equations, 4 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related work
  3. Virtual Screening
  4. Optimal Transport and inverse optimal transport
  5. Semi-supervised learning
  6. Method
  7. Overview
  8. Pretraining Backbone of Molecular Encoder
  9. Training Strategy of Encoder on Labeled Dataset
  10. Training Strategy of Encoder on Full Dataset
  11. Regularization techniques
  12. Framework for S-Molsearch Induced by Inverse Optimal Transport
  13. Experiments
  14. Training Data
  15. Benchmarks
  16. ...and 16 more sections

Key Result

Proposition 1

Given encoder $f_{\theta}$ for labeled dataset $X_{sup}$ and $g_{\psi}$ for full dataset $X_{full}$, $x_{sup}$ represents the embeddings of labeled data from $f_{\theta}$ , while $x_{full}$ represents the embeddings of the full dataset from $g_{\psi}$. Semi-supervised contrastive learning is then fo where $KL(X||Y) = \sum\limits_{ij}x_{ij}log\frac{x_{ij}}{y_{ij}} - x_{ij} + y_{ij}$ represents the

Figures (4)

  • Figure 1: Overview of S-MolSearch Framework
  • Figure 2: t-SNE visualization of molecular representations learned by S-MolSearch versus pretrained checkpoint. Different colors represent different protein targets' active molecules.
  • Figure 3: Performance on DUD-E and LIT-PCBA with varying numbers of labeled data, while keeping unlabeled data fixed at 1m. The blue bars represent the results of encoder $g_{\psi}$, while the green bars represent the results of encoder $f_{\theta}$.
  • Figure 4: Qualitative examples of similarities for targets hdac2 and csf1r in DUD-E.

Theorems & Definitions (3)

  • Proposition 1
  • Proposition 2
  • Lemma 3