In silico prediction of potential miRNA‐disease association using an integrative bioinformatics approach based on kernel fusion

Na‐Na Guan, Chunchun Wang, Li Zhang, Huang Li, Jianqiang Li, Xue PiaoJournal of Cellular and Molecular Medicine 2019 · DOI

TL;DR

This paper introduces KFRLSMDA, a novel computational model that integrates multiple data sources to predict associations between microRNAs (miRNAs) and diseases. By employing a kernel fusion technique and regularized least squares, KFRLSMDA significantly outperforms existing methods, achieving high accuracy in predicting these crucial biological links. This advancement holds promise for accelerating the discovery of disease-related miRNAs, aiding in diagnostics and therapeutic development.

What Problem Does This Paper Address?

MicroRNAs (miRNAs) are small RNA molecules that play critical roles in regulating gene expression and are increasingly implicated in various human diseases, particularly cancers. Identifying the specific miRNAs associated with particular diseases is vital for understanding disease mechanisms and developing targeted therapies. However, experimental methods for discovering these associations are time-consuming and expensive. While computational approaches have emerged to predict these links, they often struggle with accuracy, particularly when dealing with limited or noisy data, and some methods fail to predict associations for novel diseases. This paper aims to address these limitations by developing a more robust and accurate computational model for miRNA-disease association prediction.

Methods

The core of this work is the KFRLSMDA (Kernel Fusion-based Regularized Least Squares for MiRNA-Disease Association prediction) model. KFRLSMDA is an ensemble-based, semi-supervised learning approach. Here's a breakdown of its key components:

  • Kernel Fusion: Instead of simply averaging similarity matrices, KFRLSMDA uses a kernel fusion technique to combine different types of similarity information. This is applied separately in the miRNA space and the disease space.
  • Input Data Integration: The model integrates three primary data sources:
  • Known miRNA-disease associations: This forms the foundation for learning.
  • miRNA functional similarity: This captures how similar miRNAs are in terms of the diseases they regulate.
  • Disease semantic similarity: This leverages the hierarchical relationships between diseases (e.g., using MeSH descriptors and directed acyclic graphs).
  • Gaussian Interaction Profile (GIP) Kernel Similarity: A GIP kernel is computed for both miRNAs and diseases. This kernel captures interaction patterns, assuming that similar diseases are associated with miRNAs that have similar functional profiles.
  • Regularized Least Squares (RLS) Classifiers: After kernel fusion, RLS classifiers are employed in both the miRNA and disease spaces. RLS is a machine learning technique that balances fitting the data with regularization to prevent overfitting.
  • Ensemble Learning: The predictions from the miRNA space classifier and the disease space classifier are combined using a weighted average to produce the final association score.

The model is designed to be semi-supervised, meaning it can learn from unlabeled data (miRNA-disease pairs without confirmed associations) rather than requiring explicitly negative samples.

Datasets

The study utilized several publicly available databases for training and validation:

  • HMDD v2.0 database: This was the primary dataset used for training KFRLSMDA and for cross-validation. It contains a substantial number of known miRNA-disease associations.
  • miR2Disease database: Used as an evidence database for validating predictions in case studies.
  • dbDEMC database: Also used as an evidence database for validating predictions in case studies. This database focuses on differentially expressed miRNAs in human cancers.
  • HMDD v1.0 database: An older version of the HMDD database was used for a specific case study on lymphoma prediction.

The paper notes that while HMDD v2.0 contains 5430 associations, 495 miRNAs, and 383 diseases, the authors believe the current HMDD database is still insufficient for a truly comprehensive analysis.

Key Insights and Ideas

  • Kernel Fusion is More Than Averaging: The paper highlights that simply averaging similarity matrices is less effective than using a more sophisticated kernel fusion technique to integrate diverse data sources. This allows for a more nuanced capture of relationships.
  • Leveraging Both Global and Local Information: The model's success is attributed to its ability to integrate both global (e.g., overall similarity) and local (e.g., nearest neighbor interactions) information through the use of diffusion-like iterative updates and local similarity matrices.
  • No Need for Explicit Negative Samples: The semi-supervised nature of KFRLSMDA, which uses unlabeled data instead of requiring confirmed negative miRNA-disease pairs, simplifies the data preparation process and potentially improves robustness.
  • Predicting for Novel Diseases: The model's performance in predicting associations for diseases where known miRNA associations were removed (e.g., hepatocellular cancer) demonstrates its potential for identifying novel miRNA-disease links.

Key Results

KFRLSMDA demonstrated superior performance compared to 10 other state-of-the-art computational models:

  • Cross-Validation Performance:
    • Global LOOCV: Achieved an Area Under the Curve (AUC) of 0.9246.
  • Local LOOCV: Achieved an AUC of 0.8243.
  • 5-fold Cross-Validation: Achieved an average AUC of 0.9175 ± 0.0008.
  • Case Study Validation:
    • Breast Neoplasms: 96% of the top 50 predicted miRNAs were confirmed by experimental evidence.
  • Colon Neoplasms: 100% of the top 50 predicted miRNAs were confirmed.
  • Oesophageal Neoplasms: 90% of the top 50 predicted miRNAs were confirmed.
  • Hepatocellular Cancer (predicting for a disease with known associations removed): 98% of the top 50 predicted miRNAs were verified.
  • Lymphoma (using HMDD v1.0): 80% of the top 50 predicted miRNAs were confirmed.

The statistical analysis of prediction scores for confirmed versus unconfirmed miRNAs showed a significant difference, further supporting the model's reliability.

Limitations and Future Work

The authors acknowledge several limitations and suggest future research directions:

  • Parameter Optimization: KFRLSMDA has several parameters, and the optimal values are not yet definitively determined. Future work could focus on developing methods to directly obtain optimal parameter values.
  • Database Size: While HMDD v2.0 is substantial, the authors believe a larger database of known miRNA-disease associations would further enhance the model's accuracy and comprehensiveness.
  • Potential Bias: The model might exhibit bias towards miRNAs or diseases with a disproportionately large number of known associations.
  • Focus on Cancers: The current case studies primarily focused on cancers due to their clinical significance and the availability of related data. Future work could expand to include a wider range of disease types.
  • Incorporating Additional Data: The authors suggest exploring the inclusion of tissue-specific expression of miRNAs to further refine the model.

Summary updated

Abstract

Accumulating experimental evidence has demonstrated that microRNAs (miRNAs) have a huge impact on numerous critical biological processes and they are associated with different complex human diseases. Nevertheless, the task to predict potential miRNAs related to diseases remains difficult. In this paper, we developed a Kernel Fusion-based Regularized Least Squares for MiRNA-Disease Association prediction model (KFRLSMDA), which applied kernel fusion technique to fuse similarity matrices and then utilized regularized least squares to predict potential miRNA-disease associations. To prove the effectiveness of KFRLSMDA, we adopted leave-one-out cross-validation (LOOCV) and 5-fold cross-validation and then compared KFRLSMDA with 10 previous computational models (MaxFlow, MiRAI, MIDP, RKNNMDA, MCMDA, HGIMDA, RLSMDA, HDMP, WBSMDA and RWRMDA). Outperforming other models, KFRLSMDA achieved AUCs of 0.9246 in global LOOCV, 0.8243 in local LOOCV and average AUC of 0.9175 ± 0.0008 in 5-fold cross-validation. In addition, respectively, 96%, 100% and 90% of the top 50 potential miRNAs for breast neoplasms, colon neoplasms and oesophageal neoplasms were confirmed by experimental discoveries. We also predicted potential miRNAs related to hepatocellular cancer by removing all known related miRNAs of this cancer and 98% of the top 50 potential miRNAs were verified. Furthermore, we predicted potential miRNAs related to lymphoma using the data set in the old version of the HMDD database and 80% of the top 50 potential miRNAs were confirmed. Therefore, it can be concluded that KFRLSMDA has reliable prediction performance.
BibTeX
@article{guan2019silico,
  author = {Na‐Na Guan and Chunchun Wang and Li Zhang and Huang Li and Jianqiang Li and Xue Piao},
  title = {{In silico prediction of potential miRNA‐disease association using an integrative bioinformatics approach based on kernel fusion}},
  journal = {Journal of Cellular and Molecular Medicine},
  year = {2019},
  doi = {10.1111/jcmm.14765},
}
Figure