Heterogeneous network propagation for herb target identification

Kuo Yang, Guangming Liu, Ning Wang, Runshun Zhang, Jian Yu, Jianxin Chen, Xuezhong ZhouBMC Medical Informatics Decis. Mak 2018 · OpenReview

TL;DR

This paper introduces heNetRW, a novel random walk algorithm that leverages heterogeneous networks of herbs, their targets, and related biological information to predict potential protein targets for herbal drugs. heNetRW significantly outperforms existing methods like PRINCE, offering a more robust and informative approach to understanding the molecular mechanisms of Traditional Chinese Medicine (TCM).

What Problem Does This Paper Address?

Identifying the specific protein targets of herbal medicines is a crucial but challenging step in understanding their pharmacological effects and developing new treatments. Traditional experimental methods are time-consuming and costly. While computational approaches exist, effectively integrating the complex, multi-faceted data available for herbs and their targets (e.g., herb-efficacy associations, protein-pathway relationships) remains a significant hurdle. This paper aims to bridge this gap by proposing a method that can better utilize this heterogeneous network information for more accurate herb target prediction.

Methods

The core of this work is the heNetRW (heterogeneous network random walk) algorithm. It constructs a heterogeneous herb-target network that integrates:

  • Herb-herb network: Based on shared efficacies between herbs.
  • Protein-protein network: Based on shared pathways between proteins.
  • Herb-target associations: Known links between herbs and their protein targets.

A random walk is then simulated on this network. The algorithm allows a "walker" to traverse between herbs and proteins, with probabilities influenced by the network's structure and the strength of connections (e.g., efficacy similarity, pathway similarity). The steady-state probability distribution of the walker on the network is used to infer candidate targets for a given herb. This approach differs from methods like PRINCE, which primarily operate on protein-protein interaction networks, by incorporating a broader range of biological information.

Datasets

The study utilizes several publicly available datasets:

  • HIT database: For herb-target associations (integrated 23,453 associations between 1016 herbs and 1214 targets).
  • Chinese Pharmacopoeia (CHPA): For herb-efficacy associations (3487 associations between 742 herbs and 360 efficacies).
  • KEGG database: For protein-pathway associations (16,162 associations between 4794 proteins and 244 pathways).

The paper does not introduce new benchmark datasets but rather integrates existing ones to create its heterogeneous network.

Key Insights and Ideas

  • Heterogeneity is Key: The success of heNetRW highlights that simply relying on protein-protein interaction networks is insufficient. Integrating diverse data sources like herb efficacies and protein pathways provides a richer context for target prediction.
  • Efficacy and Pathway Similarity Matter: The assumption that herbs with similar efficacies share targets, and proteins with similar pathways are involved in similar biological processes, is validated. This forms the basis for constructing the herb-herb and protein-protein networks.
  • Robustness of Random Walk: The random walk approach on a well-constructed heterogeneous network proves to be more robust than methods that might struggle with isolated nodes or limited information in a single network type.
  • Parameter Sensitivity: While the algorithm is generally robust, certain parameters (like the probability of jumping between herb and protein layers) have a more significant impact on performance, suggesting avenues for fine-tuning.

Key Results

  • Improved Prediction Performance: heNetRW consistently outperformed the baseline PRINCE algorithm across two simulated scenarios:
  • NoTarget: Improved F1-score by 0.08 and Hit@1 by 21.34%.
  • HalfTarget: Improved F1-score by 0.54 and Hit@1 by 69.08%.
  • Validation of Novel Targets: The method successfully predicted potential targets for Rhizoma Coptidis and Turmeric, some of which were later confirmed by recent literature, demonstrating its practical utility.
  • Network Structure: The herb-herb network had 741 nodes and 60,753 edges (average degree 163.98), while the protein-protein network had 4794 nodes and 656,681 edges (average degree 273.95). The combined heterogeneous network was sparse (density 0.03).

Limitations and Future Work

  • NoTarget Scenario: The paper notes that heNetRW's performance was not as strong in the "NoTarget" scenario when initial nodes were solely herb nodes. Future work aims to address this by including known targets of herb neighbors as initial nodes.
  • Unilateral Similarity: The current similarity calculations for herb pairs and protein pairs are based on single aspects (efficacy or pathway). The authors suggest future work could involve building more comprehensive heterogeneous networks incorporating additional information like indications, ingredients, GO terms, and interactions, and then applying network embedding techniques for multi-dimensional similarity measures.
  • Parameter Tuning: While parameters were tuned, further exploration of their optimal values and their interplay could refine performance.

Summary updated

Abstract

Identifying targets of herbs is a primary step for investigating pharmacological mechanisms of herbal drugs in Traditional Chinese medicine (TCM). Experimental targets identification of herbs is a difficult and time-consuming work. Computational method for identifying herb targets is an efficient approach. However, how to make full use of heterogeneous network data about herbs and targets to improve the performance of herb targets prediction is still a dilemma. In our study, a random walk algorithm on the heterogeneous herb-target network (named heNetRW) has been proposed to identify protein targets of herbs. By building a heterogeneous herb-target network involving herbs, targets and their interactions and simulating random walk algorithm on the network, the candidate targets of the given herb can be predicted. The experimental results on large-scale dataset showed that heNetRW had higher performance of targets prediction than PRINCE (improved F1-score by 0.08 and Hit@1 by 21.34% in one validation setting, and improved F1-score by 0.54 and Hit@1 by 69.08% in the other validation setting). Furthermore, we evaluated novel candidate targets of two herbs (rhizoma coptidis and turmeric), which showed our approach could generate potential targets that are valuable for further experimental investigations. Compared with PRINCE algorithm, heNetRW algorithm can fuse more known information (such as, known herb-target associations and pathway-based similarities of protein pairs) to improve prediction performance. Experimental results also indicated heNetRW had higher performance than PRINCE. The prediction results not only can be used to guide the selection of candidate targets of herbs, but also help to reveal the molecule mechanisms of herbal drugs.
BibTeX
@inproceedings{yang2018heterogeneous,
  author = {Kuo Yang and Guangming Liu and Ning Wang and Runshun Zhang and Jian Yu and Jianxin Chen and Xuezhong Zhou},
  title = {{Heterogeneous network propagation for herb target identification}},
  booktitle = {BMC Medical Informatics Decis. Mak},
  year = {2018},
  url = {https://openreview.net/forum?id=e4vZj2IVAv},
}
Figure