SDA: a semi-parametric differential abundance analysis method for metabolomics and proteomics data.

Area of Science:

• Biochemistry
• Bioinformatics
• Analytical Chemistry

Background:

• Metabolomics and proteomics studies frequently aim to identify differentially abundant features across experimental groups.
• Mass spectrometry (MS) data analysis presents challenges due to potential non-normality and a high proportion of zero values.
• Existing statistical methods often rely on data normality assumptions or lack efficiency.

Purpose of the Study:

• To introduce a novel semi-parametric differential abundance analysis (SDA) method tailored for mass spectrometry (MS) data.
• To address the challenges of non-normally distributed data and excessive zero values inherent in MS datasets.
• To provide an efficient and robust statistical tool for analyzing metabolomics and proteomics data.

Main Methods:

• The proposed SDA method employs a two-part model: logistic regression for zero proportions and a semi-parametric log-linear model for non-zero values.
• A kernel-smoothed likelihood approach is utilized for estimating model coefficients.
• A likelihood ratio test is developed for conducting differential abundance analyses.

Main Results:

• The SDA method effectively characterizes MS data, accommodating both non-normally distributed values and a substantial fraction of zeros.
• The method allows for the adjustment of covariates, enhancing its applicability.
• Simulations and real-world data analyses indicate that SDA surpasses the performance of current analytical methods.

Conclusions:

• The two-part semi-parametric model in SDA provides a robust framework for analyzing complex MS data.
• SDA offers improved accuracy and efficiency in identifying differentially abundant features compared to existing techniques.
• The SDAMS R package is available for implementing this advanced statistical method.