Spurio

Spurio identifies spurious protein predictions in prokaryotic sequences by searching queries with tblastn against a prokaryotic nucleotide database and scoring evidence of degraded homologs using a Gaussian process model.

Key Features:

• tblastn search: Performs tblastn searches of query protein sequences against a prokaryotic nucleotide database to locate homologous DNA sequences.
• Feature extraction from matches: Derives features from tblastn matches with particular emphasis on the presence of stop codons within homologous DNA sequences presumed to translate into proteins.
• Gaussian process scoring: Applies a Gaussian process model to score the likelihood that a query sequence is spurious based on features derived from tblastn matches.
• Transposon protein handling: Notes that transposon proteins are frequently misclassified due to the prevalence of degraded homologs in sequence databases.
• Benchmarking versus AntiFam and UniProtKB: Shows superior performance relative to AntiFam, estimates that under 1% of UniProtKB entries may be spurious, and reports identification of over 60 times more such proteins than earlier methods.

Scientific Applications:

• Improving prokaryotic proteome quality: Identifies and filters spurious protein predictions from prokaryotic sequence databases.
• Evaluating gene prediction outputs: Assesses authenticity of protein sequences produced by gene prediction tools lacking experimental validation.
• Genome annotation curation: Supports curation of prokaryotic genome annotations by flagging likely erroneous protein entries.
• Estimating database error rates: Provides estimates of spurious-protein prevalence in reference databases such as UniProtKB.

Methodology:

Query protein sequences are searched with tblastn against a prokaryotic nucleotide database, features including stop codon occurrences are extracted from tblastn matches, and a Gaussian process model is used to score the likelihood that a query sequence is spurious.