Drug repurposing with network reinforcement

Yonghyun Nam, Myungjun Kim, Hang-Seok Chang, Hyunjung Shin

TL;DR

This paper tackles the challenge of drug repurposing by building a more comprehensive drug network. They achieve this by reinforcing an initial drug-target protein network with drug-drug interaction data, leading to improved drug recommendation accuracy for identifying new uses of existing drugs.

What Problem Does This Paper Address?

Drug discovery is an incredibly expensive and time-consuming process, often taking over a decade and costing hundreds of millions of dollars. Drug repurposing, the idea of finding new therapeutic uses for existing drugs, offers a promising alternative to de novo drug discovery. However, identifying these new uses computationally (in silico) is still challenging. Existing in silico methods often rely on sparse networks of drug information, limiting their effectiveness. This paper aims to overcome this limitation by creating a richer, more interconnected drug network that can lead to more reliable drug repurposing predictions.

Methods

The core of this work is a three-step approach:

  1. Drug Network Construction: They begin by building an initial drug network where drugs are nodes and edges represent similarities derived from shared drug-target protein information. This is a standard approach, but it often results in a sparse network because not all drug-target relationships are known.
  2. Network Reinforcement: To address the sparsity, they enhance the initial network using external data sources. This involves:
  3. Complementary Linkage with Anchoring and Scoring (CLASH): This algorithm is used to connect previously unconnected drugs. It works by "anchoring" isolated drugs to existing ones based on external knowledge and then scoring potential connections to confirm them.
  4. External Data Sources: They integrate drug-drug interaction knowledge from sources like PubChem and PubMed, specifically focusing on bioactivity and medication information. This adds crucial connections that were missing in the initial drug-target network.
  5. Drug Scoring: Once the enhanced network is built, they use a graph-based semi-supervised learning (SSL) algorithm to score drugs for a specific target disease. SSL is chosen because it can leverage both labeled (known drug-disease associations) and unlabeled data, making it suitable for sparse interaction data. The scoring algorithm assigns a priority rank to each drug, with higher scores indicating a greater potential for repurposing.

Datasets

The study utilizes several datasets:

  • Drug and Disease Lists: 5442 drugs and 1938 diseases were obtained from PubChem and the National Library of Medicine's Medical Subject Headings (MeSH).
  • Drug-Target Protein Information: 91,450 drug-target protein associations were used.
  • Disease-Drug Association: 61,794 disease-drug associations were used for validation.
  • Drug-Drug Interaction Data: 77,729 drug-drug interactions from PubChem and PubMed were used for network reinforcement.

The paper does not introduce new benchmark datasets but rather leverages existing, publicly available ones.

Key Insights and Ideas

  • Network Sparsity is a Major Bottleneck: The paper clearly demonstrates that relying solely on drug-target protein information leads to sparse networks, limiting the ability to find meaningful connections for drug repurposing.
  • External Data is Crucial for Enrichment: Integrating diverse sources of drug-drug interaction data significantly enriches the network, making it more comprehensive and informative.
  • CLASH Algorithm's Effectiveness: The CLASH algorithm proves effective in connecting previously isolated drugs, increasing drug coverage and network density.
  • SSL for Robust Drug Scoring: Semi-supervised learning is a powerful tool for drug scoring, especially when dealing with incomplete interaction data, as it can infer relationships from limited information.
  • Reinforcement Improves Prediction Reliability: The enhanced network not only increases the number of connections but also leads to more reliable drug recommendations, as evidenced by improved performance metrics.

Key Results

  • Increased Network Coverage: The enhanced network increased drug coverage from 4738 to 5442 drugs and drug-drug associations from 808,752 to 982,361.
  • Improved Prediction Performance: The Area Under the ROC Curve (AUC) for drug repurposing prediction improved from 0.79 (original network) to 0.89 (enhanced network).
  • Statistical Significance: The performance difference between the original and enhanced networks was statistically significant (p-values of 0.0003 and 0.0005 for comparisons with original and data fusion networks, respectively).
  • Example Case Study (Vascular Dementia): The method successfully identified 11 candidate drugs for vascular dementia, including Amantadine and Conotoxin GV, with high confidence scores. The paper highlights how Amantadine's repurposing potential was revealed through its connection with memantine via drug-drug interaction data.

Limitations and Future Work

  • Negative Correlation Ignored: The authors acknowledge that they did not consider the negative correlation of newly added connections, which could represent antagonistic properties. Future work could explore incorporating this.
  • Personalization Potential: They suggest that personalizing the network with patient-specific genetic information could further enhance its utility for precision medicine.
  • Expanding Knowledge Sources: The paper notes that expanding complementary knowledge sources to include drug side-effects and drug chemical structures could further improve the model.

Summary updated

Abstract

Background Drug repurposing has been motivated to ameliorate low probability of success in drug discovery. For the recent decade, many in silico attempts have received primary attention as a first step to alleviate the high cost and longevity. Such study has taken benefits of abundance, variety, and easy accessibility of pharmaceutical and biomedical data. Utilizing the research friendly environment, in this study, we propose a network-based machine learning algorithm for drug repurposing. Particularly, we show a framework on how to construct a drug network, and how to strengthen the network by employing multiple/heterogeneous types of data. Results The proposed method consists of three steps. First, we construct a drug network from drug-target protein information. Then, the drug network is reinforced by utilizing drug-drug interaction knowledge on bioactivity and/or medication from literature databases. Through the enhancement, the number of connected nodes and the number of edges between them become more abundant and informative, which can lead to a higher probability of success of in silico drug repurposing. The enhanced network recommends candidate drugs for repurposing through drug scoring. The scoring process utilizes graph-based semi-supervised learning to determine the priority of recommendations. Conclusions The drug network is reinforced in terms of the coverage and connections of drugs: the drug coverage increases from 4738 to 5442, and the drug-drug associations as well from 808,752 to 982,361. Along with the network enhancement, drug recommendation becomes more reliable: AUC of 0.89 was achieved lifted from 0.79. For typical cases, 11 recommended drugs were shown for vascular dementia: amantadine, conotoxin GV, tenocyclidine, cycloeucine, etc.
BibTeX
@article{nam2019drug,
  author = {Yonghyun Nam and Myungjun Kim and Hang-Seok Chang and Hyunjung Shin},
  title = {{Drug repurposing with network reinforcement}},
  journal = {BMC Bioinformatics},
  year = {2019},
  doi = {10.1186/s12859-019-2858-6},
}
Figure