Detecting and correcting misclassified sequences in the large-scale public databases.

Area of Science:

• Bioinformatics
• Genomics
• Computational Biology

Background:

• Public sequence repositories like the non-redundant (NR) database are growing rapidly.
• User-submitted metadata is prone to errors, leading to potential misclassification of biological sequences.
• Existing databases lack robust methods for identifying and correcting metadata errors, risking error propagation.

Purpose of the Study:

• To develop and validate a heuristic method for detecting potentially misclassified taxonomic assignments in the NR database.
• To quantify the extent of taxonomic misclassification within the entire NR database.

Main Methods:

• Implemented a curation technique and quality control for identifying probable taxonomic assignments.
• Incorporated data provenance, annotation frequency from diverse sources, and sequence similarity clustering (95%) into the detection method.

Main Results:

• Identified over two million potentially taxonomically misclassified proteins in the NR database.
• Achieved high precision (97%) and recall (87%) in detecting misclassified proteins using simulated data.

Conclusions:

• The developed heuristic method effectively detects taxonomic misclassifications in large sequence databases.
• The findings highlight significant data quality issues in public repositories, necessitating improved curation strategies.
• The methodology is adaptable for application to other biological databases.