Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Learning Multi-view Molecular Representations with Structured and Unstructured Knowledge

Yizhen Luo, Kai Yang, Massimo Hong, Xing Yi Liu, Zikun Nie, Hao Zhou, Zaiqing Nie

TL;DR

Mol is presented, a molecular representation learning model that harvests multi-view molecular expertise from chemical structures, unstructured knowledge from biomedical texts, and structured knowledge from knowledge graphs that provides improved representations that substantially benefit molecular property prediction.

Abstract

Capturing molecular knowledge with representation learning approaches holds significant potential in vast scientific fields such as chemistry and life science. An effective and generalizable molecular representation is expected to capture the consensus and complementary molecular expertise from diverse views and perspectives. However, existing works fall short in learning multi-view molecular representations, due to challenges in explicitly incorporating view information and handling molecular knowledge from heterogeneous sources. To address these issues, we present MV-Mol, a molecular representation learning model that harvests multi-view molecular expertise from chemical structures, unstructured knowledge from biomedical texts, and structured knowledge from knowledge graphs. We utilize text prompts to model view information and design a fusion architecture to extract view-based molecular representations. We develop a two-stage pre-training procedure, exploiting heterogeneous data of varying quality and quantity. Through extensive experiments, we show that MV-Mol provides improved representations that substantially benefit molecular property prediction. Additionally, MV-Mol exhibits state-of-the-art performance in multi-modal comprehension of molecular structures and texts. Code and data are available at https://github.com/PharMolix/OpenBioMed.

Learning Multi-view Molecular Representations with Structured and Unstructured Knowledge

TL;DR

Mol is presented, a molecular representation learning model that harvests multi-view molecular expertise from chemical structures, unstructured knowledge from biomedical texts, and structured knowledge from knowledge graphs that provides improved representations that substantially benefit molecular property prediction.

Abstract

Capturing molecular knowledge with representation learning approaches holds significant potential in vast scientific fields such as chemistry and life science. An effective and generalizable molecular representation is expected to capture the consensus and complementary molecular expertise from diverse views and perspectives. However, existing works fall short in learning multi-view molecular representations, due to challenges in explicitly incorporating view information and handling molecular knowledge from heterogeneous sources. To address these issues, we present MV-Mol, a molecular representation learning model that harvests multi-view molecular expertise from chemical structures, unstructured knowledge from biomedical texts, and structured knowledge from knowledge graphs. We utilize text prompts to model view information and design a fusion architecture to extract view-based molecular representations. We develop a two-stage pre-training procedure, exploiting heterogeneous data of varying quality and quantity. Through extensive experiments, we show that MV-Mol provides improved representations that substantially benefit molecular property prediction. Additionally, MV-Mol exhibits state-of-the-art performance in multi-modal comprehension of molecular structures and texts. Code and data are available at https://github.com/PharMolix/OpenBioMed.
Paper Structure (25 sections, 15 equations, 4 figures, 7 tables)

This paper contains 25 sections, 15 equations, 4 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Related work
  3. Multi-view Molecular Representation Learning
  4. Molecular Representation Learning with Heterogeneous Sources
  5. Method
  6. A Knowledge-enhanced Pre-training Perspective for Multi-view MRL
  7. Model Architecture for MV-Mol
  8. Modality Alignment with Large-scale Unstructured Knowledge
  9. Multi-view Knowledge Incorporation with Heterogeneous Triplets
  10. Experiments
  11. Pre-training Setup
  12. Molecular Property Prediction (Q1)
  13. Zero-shot Cross-modal Retrieval (Q2)
  14. Cross-modal Generation (Q2)
  15. A Deeper Analysis of View-based Molecular Representations (Q3)
  16. ...and 10 more sections

Figures (4)

  • Figure 1: An overview of molecular expertise. Molecular expertise covers diverse views that share consensus and complementary information. It resides within heterogeneous sources that vary in quality and quantity.
  • Figure 2: Model architecture and pre-training pipeline of MV-Mol. (a) MV-Mol is composed of a view-based molecule encoder and a multi-modal decoder. (b) The molecule branch of the view-based molecule encoder. (c) The text branch of the view-based molecule encoder. (d) We perform cross-modal contrastive and cross-modal matching for modality alignment. (c) We model relation as a textual prompt that constrains molecular knowledge from a specific view, and design knowledge graph embedding and knowledge graph completion objectives for multi-view knowledge incorporation. Both branches are activated when the head entity refers to a molecule.
  • Figure 3: Visualization of view-based molecular representations. We show the molecular representations from chemical, physical, and pharmacokinetic views. We also highlight three molecules and the representations of their textual descriptions from each view.
  • Figure :