Question 1

How should caller performance be benchmarked?

Accepted Answer

Run against a reference sample with a published truth set and report sensitivity and precision separately for substitutions and for small insertions and losses, since performance differs sharply between them. State the confident region used, because figures computed inside high confidence intervals are not comparable to genome wide numbers.

Question 2

Why does transcript choice change the result?

Accepted Answer

Because a variant sits in different positions relative to coding sequence depending on which transcript is used, so the predicted consequence can shift between benign and severe. Pick a defined transcript set, prefer the clinically recognised transcript for each gene where one exists, and record both the set and its release in the methods.

Question 3

How are population frequency filters set safely?

Accepted Answer

Derive the threshold from disease prevalence, inheritance mode, and penetrance rather than using a fixed cutoff. Check frequency in ancestry matched cohorts, because reference databases skew towards some populations and a variant common in an underrepresented group can appear rare. Report the filter and the cohorts consulted.

Question 4

Can classification be fully automated?

Accepted Answer

Not at present. Automated tools apply many evidence codes consistently and save considerable time, but functional evidence, segregation data, and case level context still require human curation. Treat automated output as a first pass, and benchmark it against a set of expert classified cases to measure where it agrees and where it does not.

Question 5

How should variants of uncertain significance be handled?

Accepted Answer

They are the common outcome, not a failure. Record what evidence is missing so the case can be revisited when new data arrive, and consider whether functional assay work or family segregation could resolve it. A thesis that characterises why a class of variants remains uncertain is a legitimate contribution.

Question 6

What governance applies to human genomic data?

Accepted Answer

Consent terms, institutional ethics approval, and data access agreements all bind this work, and controlled access repositories add their own conditions. Plan storage, access control, and retention before requesting data. Where incidental findings are possible, agree the reporting policy with the clinical team in advance rather than when one arises.

Bioinformatics

Bioinformatics

Research Frontiers in Genomic Variant Interpretation Pipelines