Class‐paired Fuzzy SubNETs: A paired variant of the rank‐based network analysis family for feature selection based on protein complexes

Abstract

Identifying reproducible yet relevant protein features in proteomics data is a major challenge. Analysis at the level of protein complexes can resolve this issue and we have developed a suite of feature-selection methods collectively referred to as Rank-Based Network Analysis (RBNA). RBNAs differ in their individual statistical test setup but are similar in the sense that they deploy rank-defined weights amongst proteins per sample. This procedure is known as gene fuzzy scoring. Currently, no RBNA exists for paired-sample scenarios where both control and test tissues originate from the same source (e.g. same patient). It is expected that paired tests, when used appropriately, are more powerful than approaches intended for unpaired samples. We report that the class-paired RBNA, PPFSNET, dominates in both simulated and real data scenarios. Moreover, for the first time, we explicitly incorporate batch-effect resistance as an additional evaluation criterion for feature-selection approaches. Batch effects are class irrelevant variations arising from different handlers or processing times, and can obfuscate analysis. We demonstrate that PPFSNET and PFSNET, are particularly resistant against batch effects, and only select features strongly correlated with class but not batch. This article is protected by copyright. All rights reserved.

NetProt: Complex-based Feature Selection

“…17,41 There are currently four RBNAs: SubNET (SNET), 45 Fuzzy SNET (FSNET), and paired FSNET (PFSNET) 40,45 and class-paired PFSNET (PPFSNET). 21 SNET and FSNET use the same one-sided reciprocal statistical test (Figure 2) but differ in the upstream data transformation: SNET uses a simple binarization procedure, while FSNET uses GFS. PFSNET uses GFS upstream but differs from FSNET by swapping the one-sided test in favor of a single-sample test.…”

“…However, it goes one step further and defines a class-representation proportion (class weighting) per gene, given the rationale that a top-ranking gene frequently observed in the top n % of samples in its respective class is more likely a true-positive (Figure ). Indeed, RBNAs are extremely powerful, greatly improving signal-to-noise ratios over a series of clinical blood cancer genomics data sets and also display very high utility on proteomics data. , There are currently four RBNAs: SubNET (SNET), Fuzzy SNET (FSNET), and paired FSNET (PFSNET) , and class-paired PFSNET (PPFSNET) . SNET and FSNET use the same one-sided reciprocal statistical test (Figure ) but differ in the upstream data transformation: SNET uses a simple binarization procedure, while FSNET uses GFS.…”

“…Aside from high precision and sensitivity, protein complex-based feature selection (PCBFS) is surprisingly robust against technical bias, particularly batch effects. − Batch effects are nonbiological variation and arise from confounding sources such as different experiment times, handlers, reagents, and instruments . It is an extremely important problem for contemporary data analysis and is commonly resolved by batch effect-correction algorithms (BECAs) such as ComBat and SVA .…”