Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

friends.test: rank-based method for feature selection in interaction matrices

Alexandra Suvorikova, Alexey Kroshnin, Dmirijs Lvovs, Vera Mukhina, Andrey Mironov, Elana J. Fertig, Ludmila Danilova, Alexander Favorov

TL;DR

The paper tackles the challenge of identifying informative, specific interactions within a bipartite interaction matrix when data are heterogeneous across sources. It introduces friends.test, a rank-based, scale-invariant approach that detects structural breaks in per-entity interaction profiles using a two-component mixture model and information-criterion-based model selection. Applied to head and neck cancer transcriptomics (GSE112026), the method yields a stable set of cancer-specific gene markers and enables downstream functional analysis and clustering, with strong statistical significance ($p<10^{-6}$) under permutation testing. The approach offers a scalable, unsupervised tool for feature selection and graph sparsification in heterogeneous bipartite data, with practical implications for identifying tissue- or condition-specific biomarkers and for elucidating functional relationships among markers.

Abstract

The analysis of the interaction matrix between two distinct sets is essential across diverse fields, from pharmacovigilance to transcriptomics. Not all interactions are equally informative: a marker gene associated with a few specific biological processes is more informative than a highly expressed non-specific gene associated with most observed processes. Identifying these interactions is challenging due to background connections. Furthermore, data heterogeneity across sources precludes universal identification criteria. To address this challenge, we introduce \textsf{friends.test}, a method for identifying specificity by detecting structural breaks in entity interactions. Rank-based representation of the interaction matrix ensures invariance to heterogeneous data and allows for integrating data from diverse sources. To automatically locate the boundary between specific interactions and background activity, we employ model fitting. We demonstrate the applicability of \textsf{friends.test} on the GSE112026 -- transnational data from head and neck cancer. A computationally efficient \textsf{R} implementation is available at https://github.com/favorov/friends.test.

friends.test: rank-based method for feature selection in interaction matrices

TL;DR

The paper tackles the challenge of identifying informative, specific interactions within a bipartite interaction matrix when data are heterogeneous across sources. It introduces friends.test, a rank-based, scale-invariant approach that detects structural breaks in per-entity interaction profiles using a two-component mixture model and information-criterion-based model selection. Applied to head and neck cancer transcriptomics (GSE112026), the method yields a stable set of cancer-specific gene markers and enables downstream functional analysis and clustering, with strong statistical significance () under permutation testing. The approach offers a scalable, unsupervised tool for feature selection and graph sparsification in heterogeneous bipartite data, with practical implications for identifying tissue- or condition-specific biomarkers and for elucidating functional relationships among markers.

Abstract

The analysis of the interaction matrix between two distinct sets is essential across diverse fields, from pharmacovigilance to transcriptomics. Not all interactions are equally informative: a marker gene associated with a few specific biological processes is more informative than a highly expressed non-specific gene associated with most observed processes. Identifying these interactions is challenging due to background connections. Furthermore, data heterogeneity across sources precludes universal identification criteria. To address this challenge, we introduce \textsf{friends.test}, a method for identifying specificity by detecting structural breaks in entity interactions. Rank-based representation of the interaction matrix ensures invariance to heterogeneous data and allows for integrating data from diverse sources. To automatically locate the boundary between specific interactions and background activity, we employ model fitting. We demonstrate the applicability of \textsf{friends.test} on the GSE112026 -- transnational data from head and neck cancer. A computationally efficient \textsf{R} implementation is available at https://github.com/favorov/friends.test.
Paper Structure (18 sections, 7 equations, 2 figures, 2 tables, 1 algorithm)

This paper contains 18 sections, 7 equations, 2 figures, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Accepted notations.
  3. Methodology
  4. Scale-invariant data representation
  5. The structural break model
  6. Detecting friendship
  7. Experimental results
  8. Friends' set similarity analysis
  9. Discussion
  10. Related works
  11. Limitations of the approach
  12. Possible applications and interpretation of the results
  13. Feature selection and graph sparsification.
  14. Functional similarity and clustering.
  15. "Anti-friends" and negative selection.
  16. ...and 3 more sections

Figures (2)

  • Figure 1: Validation of the friends.test method. (a) Frequency of gene identification across parallel runs, with the dashed line indicating the stability threshold (0.25). (b) The empirical null distribution shows that the number of identified markers in the real data significantly exceeds random noise-based results. The white bars represent the $10^6$ permutations; the gray bars correspond to all the reliability test runs; the dark-gray bar consists the run that was user for permutations.
  • Figure 2: Hierarchical tree based on Weighed Jaccard Similarity.