A Statistical Approach to Correcting Cross-Annotations in a Metagenomic Functional Profile Generated by Short Reads

Area of Science:

• Metagenomics
• Bioinformatics
• Computational Biology

Background:

• Functional profiling of metagenomic samples relies on categorizing protein-coding sequences by biochemical function.
• Relative abundances are determined by aligning sequencing reads to protein databases, but short reads cause cross-annotation errors.
• Current methods use empirical cutoffs or simple adjustments to address cross-annotation.

Purpose of the Study:

• To develop a method to correct cross-annotation errors in metagenomic functional profiling.
• To improve the accuracy of functional profiling using short sequencing reads.

Main Methods:

• Probabilistic Latent Semantic Analysis (PLSA) was employed to model read proportions assigned to functional families.
• The PLSA approach was applied to metagenomic samples with known or estimated functional families.
• The method was tested on simulated and real-world metagenomic data.

Main Results:

• The Probabilistic Latent Semantic Analysis approach successfully addressed cross-annotation issues.
• The method demonstrated effectiveness on in vitro-simulated, bioinformatics tool-simulated, and real-world metagenomic samples.
• Accurate functional family assignment was achieved.

Conclusions:

• Correcting cross-annotation significantly increases the accuracy of metagenomic functional profiling from short reads.
• Improved functional profiling benefits differential abundance analysis of metagenomic samples.
• This method enhances the reliability of metagenomic functional characterization.