AGMT common mistakes
AGMT · Understand the traps and how to avoid them.
Reporting a variant as pathogenic on the basis of a single in-silico prediction or a single affected proband, without applying the full ACMG/AMP evidence framework and combining criteria into a classification.
Candidates memorize the criteria list but never practice weighting and combining them, so they treat any supporting line of evidence as sufficient and skip the strength hierarchy (very strong/strong/moderate/supporting) and the combination rules.
What to do instead
Practice classifying a set of variants end-to-end using the ACMG/AMP combining rules, writing out each code used, its strength, and the justified final class; learn which criteria can never stand alone (e.g., PP3, BP4) and which combinations are required for pathogenic/likely pathogenic.
Conflating analytic sensitivity/specificity with clinical sensitivity/specificity and positive predictive value when interpreting or comparing test performance.
These are taught together and candidates default to the familiar 2x2 table without asking whether the denominator is 'variants detected' or 'patients with disease', which changes the metric entirely.
What to do instead
For every performance question, explicitly define the reference standard and the population before computing anything; practice converting between analytic and clinical metrics on the same scenario and state which one answers the clinical question asked.
Choosing the wrong cytogenetic or molecular test for a given clinical scenario (e.g., ordering a karyotype when a microdeletion or a single-gene point variant is the likely cause, or ordering a panel when aneuploidy detection is the goal).
Candidates learn test technologies in isolation and never build a resolution-versus-indication map, so they pick the test they know best rather than the one whose resolution and target match the phenotype.
What to do instead
Build and memorize a resolution ladder (karyotype ~5-10 Mb, FISH locus-specific, CMA kb-scale copy number only, sequencing base-pair level) and rehearse first-tier test selection for common indications: developmental delay, recurrent pregnancy loss, suspected aneuploidy, known familial variant, and prenatal screening.
Misinterpreting a variant of uncertain significance as actionable or as equivalent to a negative result, and failing to describe the correct follow-up (segregation, functional studies, reclassification, family studies).
Candidates want a binary answer and treat VUS as a communication problem rather than a distinct evidence category, so they either over-act or dismiss it.
What to do instead
Practice writing VUS results with explicit next steps: do not use for clinical management, offer parental testing to phase, pursue segregation in affected relatives, set a reanalysis/recontact plan, and document the evidence that would move the classification.
Applying population genetics formulas incorrectly: using allele frequency as genotype frequency, forgetting that carrier frequency is approximately 2q, or mishandling X-linked and consanguinity adjustments in risk calculations.
Candidates memorize the Hardy-Weinberg equations but not their assumptions or the mapping between disease incidence, allele frequency, and carrier frequency, so they plug numbers into the wrong slot.
What to do instead
Work a fixed set of archetypes (autosomal recessive carrier risk from incidence, Bayesian risk with test results, X-linked carrier risk with affected sons) and always state the assumption (HWE, no consanguinity, disease frequency vs allele frequency) before computing.
Failing to apply Bayesian reasoning when updating carrier or recurrence risk with test results, new information, or family history.
Candidates treat risks as fixed numbers and do not recognize when prior probability must be revised by conditional probabilities, so they ignore the informative power of a negative test in an at-risk pedigree.
What to do instead
Practice constructing explicit prior/conditional/joint/posterior tables for carrier-risk and recurrence-risk scenarios, and always ask 'what is the prior, what is the new information, and how likely is that information under each hypothesis?'
Confusing coverage, depth, and read-level metrics with variant quality, and accepting or rejecting calls without checking mapping quality, strand bias, allele balance, or orthogonal confirmation.
Candidates conflate 'the base was sequenced' with 'the variant is real', and pipeline output is treated as ground truth rather than as a set of calls requiring QC filters.
What to do instead
Learn the specific filters (MQ, BQ, DP, AF/allele balance, strand bias, repeat context) and the rule that clinically reported variants require orthogonal confirmation by a second method; practice reviewing a variant call's supporting metrics and deciding pass/fail.
Ignoring quality-management fundamentals: not distinguishing validation from verification, not knowing when proficiency testing or a new lot/lot-to-lot comparison is required, and misapplying CAP/CLIA requirements to a test's lifecycle.
Regulatory content is memorized as acronyms rather than as a workflow, so candidates cannot place a given activity at the correct stage (development, validation, verification, ongoing monitoring).
What to do instead
Map each activity to its lifecycle stage: method validation before clinical use, verification when introducing an approved test or new instrument, ongoing QC and PT after launch, and documented change control for any modification; rehearse which changes require revalidation.
Treating an incidental or secondary finding as a routine result, or disclosing/withholding genetic information without addressing consent, confidentiality, duty to warn, or insurance/genetic-discrimination protections.
Candidates focus on the science and underweight the consent and disclosure framework, so they answer the clinical question and miss the ethical/legal layer the item is testing.
What to do instead
For any scenario involving family members or third parties, run a checklist: was informed consent obtained and for what scope, what is the patient's preference on secondary findings, is there a duty to warn and to whom, and which statute (e.g., genetic information nondiscrimination protections) applies.
Misreading variant nomenclature and reference transcript context: not checking the transcript/version, misapplying HGVS rules (c. vs p., intronic offsets, duplications vs insertions), and reporting a variant against the wrong reference.
Candidates treat the variant string as self-explanatory and skip the step of confirming transcript, genome build, and HGVS syntax, which shifts the meaning of the variant.
What to do instead
Before interpreting any variant, verify gene, transcript and version, genome build, and HGVS correctness; practice converting between genomic, coding, and protein descriptions and checking that the stated consequence matches the transcript used.