SIBIS

SIBIS identifies inconsistencies and errors in protein sequences by analyzing multiple sequence alignments to improve the accuracy of protein sequence databases.

Key Features:

• Multiple sequence alignment analysis: Leverages evolutionary information in multiple sequence alignments to assess residue consistency across homologs.
• Bayesian Dirichlet mixture modeling: Applies a Bayesian framework integrated with Dirichlet mixture models to model amino-acid substitution patterns.
• Amino-acid probability estimation: Estimates the probability of observing specific amino acids at each alignment position.
• Inconsistency detection: Detects inconsistent or erroneous segments that may result from natural genetic variants or errors introduced during sequence prediction.
• Performance validation: Demonstrated higher sensitivity with minimal loss of specificity against a reference set of protein sequences with experimentally validated errors.
• Large-scale UniProt analysis: Applied to human sequences from UniProt, identifying evidence of inconsistency in 48% of previously uncharacterized sequences.
• Implementation: Implemented in C for Linux computational environments.

Scientific Applications:

• Protein database quality control: Identifies and flags inconsistent sequence segments to improve the accuracy of protein sequence databases.
• Gene prediction validation: Supports evaluation of predicted protein-coding genes and exon structures by detecting prediction-induced sequence errors.
• Structural, functional, and phylogenetic inference: Improves downstream inference of protein structure, function, and phylogeny by filtering erroneous sequences.
• Benchmarking and validation: Serves for benchmarking methods using reference sets of experimentally validated sequence errors.
• Large-scale screening of UniProt: Enables large-scale screening of UniProt human sequences to characterize the prevalence of inconsistencies.

Methodology:

Uses multiple sequence alignments and evolutionary information within a Bayesian framework combined with Dirichlet mixture models to estimate amino-acid probabilities at alignment positions and detect inconsistent segments; implemented in C on Linux.