Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Towards Unified AI Drug Discovery with Multiple Knowledge Modalities

Yizhen Luo, Xing Yi Liu, Kai Yang, Kui Huang, Massimo Hong, Jiahuan Zhang, Yushuai Wu, Zaiqing Nie

TL;DR

KEDD introduces a unified end-to-end multimodal framework that jointly leverages molecular structure, structured knowledge from knowledge graphs, and unstructured biomedical text to advance AI-driven drug discovery. It addresses missing modality with a sparse-attention mechanism over knowledge graphs and a modality-masking strategy to robustly learn from incomplete data. Across DTI, DP, DDI, and PPI tasks, KEDD achieves state-of-the-art performance and demonstrates practical potential via a case study on ACE2, illustrating its applicability to real-world drug discovery and repurposing. The work provides strong evidence that integrating diverse biomedical knowledge sources can yield richer molecular representations and more accurate predictions than modality-specific approaches.

Abstract

In recent years, AI models that mine intrinsic patterns from molecular structures and protein sequences have shown promise in accelerating drug discovery. However, these methods partly lag behind real-world pharmaceutical approaches of human experts that additionally grasp structured knowledge from knowledge bases and unstructured knowledge from biomedical literature. To bridge this gap, we propose KEDD, a unified, end-to-end, and multimodal deep learning framework that optimally incorporates both structured and unstructured knowledge for vast AI drug discovery tasks. The framework first extracts underlying characteristics from heterogeneous inputs, and then applies multimodal fusion for accurate prediction. To mitigate the problem of missing modalities, we leverage multi-head sparse attention and a modality masking mechanism to extract relevant information robustly. Benefiting from integrated knowledge, our framework achieves a deeper understanding of molecule entities, brings significant improvements over state-of-the-art methods on a wide range of tasks and benchmarks, and reveals its promising potential in assisting real-world drug discovery.

Towards Unified AI Drug Discovery with Multiple Knowledge Modalities

TL;DR

KEDD introduces a unified end-to-end multimodal framework that jointly leverages molecular structure, structured knowledge from knowledge graphs, and unstructured biomedical text to advance AI-driven drug discovery. It addresses missing modality with a sparse-attention mechanism over knowledge graphs and a modality-masking strategy to robustly learn from incomplete data. Across DTI, DP, DDI, and PPI tasks, KEDD achieves state-of-the-art performance and demonstrates practical potential via a case study on ACE2, illustrating its applicability to real-world drug discovery and repurposing. The work provides strong evidence that integrating diverse biomedical knowledge sources can yield richer molecular representations and more accurate predictions than modality-specific approaches.

Abstract

In recent years, AI models that mine intrinsic patterns from molecular structures and protein sequences have shown promise in accelerating drug discovery. However, these methods partly lag behind real-world pharmaceutical approaches of human experts that additionally grasp structured knowledge from knowledge bases and unstructured knowledge from biomedical literature. To bridge this gap, we propose KEDD, a unified, end-to-end, and multimodal deep learning framework that optimally incorporates both structured and unstructured knowledge for vast AI drug discovery tasks. The framework first extracts underlying characteristics from heterogeneous inputs, and then applies multimodal fusion for accurate prediction. To mitigate the problem of missing modalities, we leverage multi-head sparse attention and a modality masking mechanism to extract relevant information robustly. Benefiting from integrated knowledge, our framework achieves a deeper understanding of molecule entities, brings significant improvements over state-of-the-art methods on a wide range of tasks and benchmarks, and reveals its promising potential in assisting real-world drug discovery.
Paper Structure (23 sections, 3 equations, 6 figures, 5 tables)

This paper contains 23 sections, 3 equations, 6 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Knowledge-empowered deep learning in AI drug discovery.
  4. Missing modality in multimodal learning.
  5. Method
  6. Preliminaries
  7. KEDD Architecture
  8. Mitigating Missing Modality with Sparse Attention and Modality Masking
  9. Experiments and Results
  10. Data preparation
  11. Implementation Details
  12. Performance Evaluation on Downstream Tasks
  13. DTI.
  14. DP.
  15. DDI.
  16. ...and 8 more sections

Figures (6)

  • Figure 1: (a) The KEDD framework. GraphMVP and MCNN can both serve as molecular structure encoders A and/or B, depending on the task. The "B" branches may also remain unused in the case of DP prediction. SK: structured knowledge; UK: unstructured knowledge. (b) Sparse attention pipeline for obtaining structured knowledge embeddings if a certain molecule is not found in BMKG.
  • Figure 2: AUROC on the BMKG-DTI dataset.
  • Figure 3: AUROC on the Yamanishi08 dataset.
  • Figure 4: Relationships between performance gain of sparse attention and the ratio of molecules without structured knowledge. Each dot represents the result on dataset, colored by its corresponding task.
  • Figure 5: A drug repurposing example for ACE2. (a) Details of ACE2, a protein targeted by KEDD. (b) Top 5 drug candidates proposed by KEDD and the heterogeneous information for each.
  • ...and 1 more figures