Variable locus length in the human genome leads to ascertainment bias in functional inference for non-coding elements

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Functional gene annotation databases infer biological function by identifying over- and underrepresented attributes.
• This approach is problematic for non-coding DNA due to variable sequence lengths, leading to biased functional conclusions based on neighboring genes.

Purpose of the Study:

• To assess systematic bias in Gene Ontology (GO) categories when inferring function for non-coding elements based on nearest gene annotations.
• To develop a correction method to eliminate ascertainment bias in non-coding DNA functional inference.

Main Methods:

• Systematic bias in GO categories was assessed using the hypergeometric test.
• Non-coding elements were randomly sampled from the human genome, and their function was inferred from the closest genes.
• Correction coefficients were developed to adjust GO category probabilities based on non-coding DNA length variability.

Main Results:

• Certain GO categories ('cell adhesion', 'nervous system development', 'transcription factor activities') were systematically overrepresented, while others ('olfactory receptor activity') were underrepresented in non-coding elements.
• A novel set of correction coefficients was introduced to account for non-coding DNA length variability.
• The proposed correction effectively eliminated ascertainment bias in functional characterization.

Conclusions:

• Functional inference for non-coding elements using standard gene annotation databases requires specific corrections.
• The developed correction coefficients accurately adjust for ascertainment bias, enabling more reliable functional characterization of non-coding DNA.
• This approach is generalizable to other gene annotation databases.