Reproducibility Crossroads: Impact of Statistical Choices on Proteomics Functional Enrichment

Quantitative proteomics analyses rely on robust statistical methods for differential expression, impacting downstream pathway and functional enrichment. This meta-analysis investigated the influence of Hypothesis Testing Methods (HTMs) and Criteria for Biological Relevance (CBRs) on functional enrichment concordance. Five independent label-free quantitative proteomics datasets were reanalyzed using diverse frequentist (t-test, Limma, DEqMS, MSstats) and a Bayesian (rstanarm) approach. Concordance of enriched terms was assessed using Jaccard indices, categorized by four comparison types: Intra-HTM_FC_CBR, Intra-HTM_Bayes_CBR, Intra-CBR_Fixed_HTM, and Inter-HTM_Inter_CBR. Results showed highly significant differences in Jaccard similarity distributions among comparison types (Kruskal-Wallis p = 5e-04). “Intra-HTM_FC_CBR” exhibited the highest consistency, indicating minor HTM influence when using FC-based CBR. “Intra-CBR_Fixed_HTM” also maintained high concordance, suggesting robust agreement between FC and Bayesian CBRs when HTM is fixed. Conversely, “Intra-HTM_Bayes_CBR” and “Inter-HTM_Inter-CBR” showed the lowest consistency, highlighting the critical impact of Bayesian method choice and mixed comparisons on functional overlaps, particularly for Gene Ontology terms. KEGG pathways displayed more uniform, method-insensitive concordance. Sensitivity analysis confirmed the robustness of these findings. This study underscores that analytical choices profoundly influence functional enrichment outcomes, emphasizing the need for transparency and careful consideration in proteomics research to ensure reproducibility.