Mind the gaps: overlooking inaccessible regions confounds statistical testing in genome analysis

Area of Science:

• Genomics
• Bioinformatics
• Statistical Genetics

Background:

• Reference genome assemblies contain gaps (stretches of Ns) and targeted sequencing assays omit certain regions, creating "inaccessible regions" depleted of experimental data.
• These inaccessible regions are often ignored in statistical analyses, potentially compromising the reliability of findings.

Purpose of the Study:

• To investigate whether ignoring inaccessible regions in the null model inflates false findings in statistical tests for genomic feature colocalization.
• To evaluate the impact of assembly gaps on the statistical significance of colocalization analyses.

Main Methods:

• Exploratory analysis of public genomic tracks against human reference genomes (hg19, hg38) to assess overlap with assembly gaps.
• Simulation of synthetic genomic tracks to test the hypothesis that excluding inaccessible regions from the null model leads to spurious inflation of statistical significance.
• Comparison of Monte Carlo permutation tests for colocalization, contrasting models that either include or exclude assembly gaps in the null distribution.

Main Results:

• Public genomic tracks show minimal overlap with assembly gaps, primarily at gap boundaries.
• Statistical tests that did not account for assembly gaps in the null model exhibited a right-shifted test statistic distribution and a left-shifted p-value distribution, indicating inflated significance.
• This inflation of significance was observed consistently across both hg19 and hg38 reference genomes.

Conclusions:

• Inaccessible genomic regions, even if a small fraction of the genome, can cause substantial false positives in colocalization analyses if not properly handled.
• Accounting for assembly gaps within the null model is essential for accurate statistical testing of genomic feature colocalization.