HetNetEX: Exact Asymptotic Inference in Heterogeneous Biomedical Knowledge Graphs

Area of Science:

• Biomedical Informatics
• Network Science
• Computational Biology

Background:

• Heterogeneous biomedical knowledge networks (hetnets) integrate diverse data like drugs, genes, and diseases.
• Existing methods like XSwap use permutations to assess connectivity significance but face scalability issues in large networks.
• These limitations include finite p-value resolution, high computational cost, and inaccurate variance modeling.

Purpose of the Study:

• To introduce HetNetEX, an analytical method for computing the null distribution of degree-weighted path counts (DWPC) in hetnets.
• To address the practical constraints of permutation-based methods like XSwap for large-scale network analysis.
• To provide an exact inference approach that is computationally efficient and offers analytical p-values.

Main Methods:

• Developed HetNetEX for analytical computation of null DWPC distribution using the configuration model.
• Utilized degree sequences for efficient calculation with O(Ln) runtime.
• Compared HetNetEX against XSwap through simulations with P=200 and path lengths L=1-4.

Main Results:

• HetNetEX achieved high concordance (Spearman ρ > 0.96) with XSwap rankings in simulations.
• HetNetEX was over 10,000 times faster than XSwap.
• Analytical p-values from HetNetEX avoid the resolution ceiling inherent in permutation methods and show lower sampling error for high-degree pairs.

Conclusions:

• HetNetEX offers a computationally efficient and accurate alternative to permutation-based methods for analyzing hetnets.
• The analytical approach overcomes key limitations of XSwap, particularly for large networks and high-degree node pairs.
• HetNetEX enables more precise and scalable assessment of connectivity significance in complex biomedical networks.