On the Robustness of Graph-Based Clustering to Random Network Alterations
Bibliographic record
Abstract
Biological functions emerge from complex and dynamic networks of protein-protein interactions. Because these protein-protein interaction networks, or interactomes, represent pairwise connections within a hierarchically organized system, it is often useful to identify higher-order associations embedded within them, such as multimember protein complexes. Graph-based clustering techniques are widely used to accomplish this goal, and dozens of field-specific and general clustering algorithms exist. However, interactomes can be prone to errors, especially when inferred from high-throughput biochemical assays. Therefore, robustness to network-level noise is an important criterion. Here, we tested the robustness of a range of graph-based clustering algorithms in the presence of noise, including algorithms common across domains and those specific to protein networks. Strikingly, we found that all of the clustering algorithms tested here markedly amplified network-level noise. Randomly rewiring only 1% of network edges yielded more than a 50% change in clustering results. Moreover, we found the impact of network noise on individual clusters was not uniform: some clusters were consistently robust to injected noise, whereas others were not. Therefore we developed the clust.perturb R package and Shiny web application to measure the reproducibility of clusters by randomly perturbing the network. We show that clust.perturb results are predictive of real-world cluster stability: poorly reproducible clusters as identified by clust.perturb are significantly less likely to be reclustered across experiments. We conclude that graph-based clustering amplifies noise in protein interaction networks, but quantifying the robustness of a cluster to network noise can separate stable protein complexes from spurious associations.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.015 | 0.094 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.002 |
| Bibliometrics | 0.004 | 0.002 |
| Science and technology studies | 0.001 | 0.004 |
| Scholarly communication | 0.002 | 0.003 |
| Open science | 0.002 | 0.002 |
| Research integrity | 0.002 | 0.002 |
| Insufficient payload (model declined to judge) | 0.001 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".