misannotations

Misannotations are erroneous assignments of information or labels to data objects. In biology, misannotations refer to incorrect functional, structural, or contextual annotations attached to genes, transcripts, or proteins, and in other domains to errors in labeling data such as texts or images. Misannotations can arise from automated pipelines, human error, or outdated knowledge and can propagate through databases and literature.

In genomics and proteomics, misannotation includes a gene model that is incorrectly predicted, a protein assigned a function without sufficient experimental evidence, incorrect enzyme class or subcellular localization, or wrong gene name. Common causes are overreliance on sequence similarity without corroborating experiments, transfer of annotation from related species, misinterpretation of domain architecture, and database versioning issues. Pseudogenes misannotated as coding genes and multi-domain proteins mistaken for single-function enzymes are notable examples.

Impact of misannotations includes misleading research, wasted resources, flawed comparative analyses, and cascading errors as annotations are reused in multiple resources. Inaccurate annotations can hinder replication and obscure true biology, affecting downstream experiments and interpretations.

Mitigation efforts focus on detection and correction. Approaches include manual curation, standardized evidence codes, experimental validation, and regular reannotation cycles. Databases employ cross-reference checks, orthology-based reconciliation, versioning, and community annotation efforts to reduce errors. In data science and related fields, best practices involve documenting provenance, uncertainty estimates, and using curated gold standards, along with inconsistency checks across databases to flag dubious annotations.