Phylogeny-aware identification and correction of taxonomically mislabeled sequences

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Public molecular sequence databases often contain taxonomic mislabels due to unvalidated author annotations.
• These errors propagate, affecting downstream analyses and biasing metagenetic studies.
• Manual curation is labor-intensive, leading to low correction rates.

Purpose of the Study:

• To develop and validate an automated method for identifying and correcting taxonomically mislabeled sequences.
• To assess the prevalence of mislabels in major microbial 16S rRNA gene reference databases.
• To evaluate taxonomic classifications using a phylogeny-aware approach.

Main Methods:

• Utilized the Evolutionary Placement Algorithm (EPA) to detect phylogenetic signal inconsistencies.
• Developed SATIVA, a phylogeny-aware method employing statistical models of evolution.
• Applied SATIVA to simulated data and real-world microbial 16S rRNA gene databases.

Main Results:

• SATIVA achieved high accuracy in identifying (96.9% sensitivity, 91.7% precision) and correcting (94.9% sensitivity, 89.9% precision) mislabeled sequences.
• Detected 0.2%–2.5% mislabels across Greengenes, LTP, RDP, and SILVA databases.
• Performed in-depth taxonomic evaluation for Cyanobacteria.

Conclusions:

• SATIVA offers an accurate and efficient solution for detecting and correcting taxonomic mislabels in sequence databases.
• The prevalence of mislabels in widely used databases highlights the need for automated quality control.
• Phylogeny-aware methods are crucial for reliable taxonomic assignment in molecular data.