Human drug-pathway association prediction based on network consistency projection

Area of Science:

• Computational biology
• Pharmacogenomics
• Bioinformatics

Background:

• Identifying drug-pathway associations is crucial for disease diagnosis, therapy, and drug development.
• Traditional biological studies and clinical trials are time-consuming and resource-intensive.
• Computational methods are increasingly used to predict drug-pathway interactions.

Purpose of the Study:

• To propose a novel computational method, Network Consistency Projection for Human Drug-Pathway Association (NCPHDPA), for predicting drug-pathway associations.
• To develop a parameter-less method that does not require negative samples and can predict associations for drugs with no known pathways.
• To evaluate the predictive performance of NCPHDPA against existing methods.

Main Methods:

• NCPHDPA utilizes drug-pathway target information, computing pathway-pathway interaction similarity and drug-drug interaction similarity using Jaccard similarity.
• The method integrates drug cosine similarity, pathway cosine resemblance, and known drug-pathway interaction networks.
• Leave-One-Out Cross-Validation (LOOCV) and 10-fold Cross-Validation (CV) were employed for performance evaluation.

Main Results:

• NCPHDPA achieved high predictive accuracy with an Area Under the ROC Curve (AUC) of 0.7479 (LOOCV) and 0.7566 (10-fold CV).
• The method outperformed existing approaches like SIMCCDA (AUC = 0.7364), LOMDA (AUC = 0.6729), and DMTHNDM (AUC = 0.5000).
• Case studies involving cancer pathways and hepatocellular carcinoma demonstrated NCPHDPA's robust predictive capability.

Conclusions:

• NCPHDPA offers a reliable and efficient computational approach for predicting drug-pathway associations.
• The method's performance suggests its utility in accelerating drug discovery and understanding disease mechanisms.
• Findings highlight the link between protein interactions, pathway alterations, and cancer onset.