A new estimation of protein-level false discovery rate

Area of Science:

• Proteomics
• Bioinformatics
• Statistical analysis

Background:

• Protein identification in mass spectrometry-based proteomics is crucial.
• Accurate statistical evaluation, particularly false discovery rate (FDR) control, is vital for reliable proteomics results.
• Current FDR estimation methods often rely on specific assumptions or involve complex multi-stage calculations.

Purpose of the Study:

• To develop a novel, non-parametric protein-level FDR estimation framework.
• To overcome limitations of existing methods, such as reliance on assumptions and two-stage calculations.
• To improve the efficiency and accuracy of FDR estimation in proteomics.

Main Methods:

• Proposed a new protein-level FDR estimation framework.
• Introduced the Permutation+BH (Benjamini-Hochberg) method for non-parametric FDR estimation.
• Developed a logistic regression-based method for efficient null distribution inference.

Main Results:

• The Permutation+BH method generates null distributions without specific assumptions, enabling direct p-value calculation and FDR control.
• The logistic regression model effectively infers null distributions for new samples, addressing the inefficiency of the permutation method for online identification.
• Experimental validation on public datasets demonstrated superior performance compared to the benchmark method (MAYU).

Conclusions:

• The proposed Permutation+BH method provides a robust and assumption-light approach to protein-level FDR estimation.
• The logistic regression model offers an efficient way to infer null distributions, enhancing practical applicability.
• The new framework consistently outperforms existing methods in controlling FDR in proteomics studies.