Questions: Functional Annotation

Question 1

Why is annotation based on Pfam domain identification sometimes more informative than a simple BLAST best-hit approach?

• A. Pfam domains are always longer than BLAST alignments
• B. Pfam uses hidden Markov models that capture position-specific conservation patterns, detecting remote homologs that BLAST misses
• C. Pfam does not require a reference database
• D. BLAST cannot search protein sequences

Answer: (B)

Pfam domain models are profile HMMs built from curated multiple sequence alignments of known protein families. They encode position-specific amino acid preferences and insertion/deletion probabilities, making them far more sensitive than pairwise BLAST for detecting distant homologs. A protein may share only 15% sequence identity with known family members — too low for confident BLAST matches — but the HMM can still recognize it as a family member because the pattern of conserved and variable positions matches. Domain annotation also identifies multi-domain architectures that provide richer functional information than a single best BLAST hit.

Question 2

A gene's functional annotation based on sequence homology is always experimentally verified.

• T. True
• F. False

Answer: False

The vast majority of functional annotations are computational predictions transferred from experimentally characterized homologs, not direct experimental evidence for the gene in question. GO annotations carry evidence codes that distinguish experimental evidence (IDA, IMP, IGI) from computational prediction (ISS, IEA). In most genomes, fewer than 1% of genes have direct experimental functional evidence. This means annotation errors can propagate: if the original experimentally characterized gene was misannotated, all genes annotated by homology to it inherit the error. Annotation quality should always be evaluated using evidence codes.

Question 3

Explain the concept of annotation error propagation and why it is a significant problem in genomic databases.

Model answer: When a gene is functionally annotated by homology to another gene, any error in the source annotation is transferred to the new gene. As databases grow, these errors amplify: a misannotated gene serves as the basis for annotating hundreds of homologs across genomes, each of which may then serve as evidence for still more annotations. Studies estimate that 5-40% of functional annotations in public databases contain errors, depending on the organism and the specificity of the annotation. The problem is compounded because transitive annotation (A annotated from B, B annotated from C) can link genes through chains of decreasing reliability.

This is why experimental evidence codes matter and why curated databases like SwissProt (manually reviewed) are more reliable than TrEMBL (automatically annotated). It is also why careful orthology assignment is critical — annotating a gene based on a paralog rather than the true ortholog increases the risk of functional misprediction.