How to Use Population Frequency Databases (gnomAD, 1000 Genomes) for Variant Interpretation: Thresholds and Best Practices

By Dr. Zubair Khalid, DVM, MS, PhD · 2026-08-21

Key Takeaways

• Population frequency databases like gnomAD and 1000 Genomes are critical for variant interpretation, informing criteria such as BA1 (Benign stand-alone), BS1 (Benign supporting), and PM2 (Pathogenic supporting) within the ACMG/AMP framework.
• Setting allele frequency thresholds requires calculating the "maximum credible AF" based on disorder prevalence, inheritance pattern (autosomal dominant, recessive, X-linked), penetrance, and genetic heterogeneity, rather than relying solely on fixed cutoffs.
• Accurate application necessitates rigorous quality control, including verifying adequate read depth (e.g., 30X for substitutions) and assessing filter status and population representation to avoid misclassifications due to poor data quality or under-representation of specific ancestries.
• Interpreters must document the database version, population subset, allele frequency, read depth, maximum credible AF calculation inputs, and rationale for criterion application to ensure reproducibility and facilitate re-evaluation with database updates.
• Conflicts between frequency data and other evidence (e.g., functional studies) require careful weighing of evidence strength and confidence, with high-confidence frequency data often taking precedence, particularly for BA1.
• Reliance on global allele frequencies can be misleading; using population-specific data that matches the patient's ancestry is crucial, especially for admixed populations where allele frequencies can differ significantly from global averages.

Population frequency databases such as gnomAD and 1000 Genomes provide reference data for interpreting germline sequence variants in clinical and research settings. The core problem for interpreters is determining whether a variant is rare enough to support pathogenicity or common enough to support benign classification under the ACMG/AMP framework. This article explains how to set allele frequency thresholds for criteria such as PM2, BA1, and BS1, how to account for population-specific data, and how to document decisions in a reproducible workflow. The guidance applies to biology students, researchers, laboratory professionals, and life-science practitioners who perform variant interpretation for Mendelian disorders, hereditary cancer syndromes, and other germline applications.

Context for Population Frequency Data in Variant Interpretation

Population frequency databases aggregate sequence data from large cohorts of individuals without a targeted disease phenotype. The two most widely used resources are gnomAD (Genome Aggregation Database) and 1000 Genomes. These databases allow interpreters to ask a fundamental question: how common is this allele in the general population? The answer informs whether a variant is likely to be disease-causing or likely to be benign.

The ACMG/AMP guidelines provide a structured framework for classifying variants into five categories: pathogenic, likely pathogenic, uncertain significance, likely benign, and benign. Population frequency data directly support several criteria within this framework. The BA1 criterion applies when a variant is found at too high a frequency to be causative for a dominant disorder. The BS1 criterion applies when the allele frequency is greater than expected for the disorder. The PM2 criterion applies when a variant is absent from or extremely rare in population databases, which supports a pathogenic classification when other evidence is present.

The practical challenge is that the guidelines do not specify exact allele frequency thresholds. Interpreters must set thresholds based on the disorder prevalence, inheritance pattern, genetic heterogeneity, and penetrance. This article provides a structured approach to making those decisions and documenting them for clinical or research use.

At a Glance: Frequency Criteria and Threshold Considerations

The table below summarizes the key population frequency criteria in the ACMG/AMP framework and the factors that influence threshold selection.

The application of these criteria requires careful attention to the source of allele frequency data. Exome-derived and genome-derived datasets within gnomAD can show different read depth for different gene exons, which affects variant detection reliability. A study of cancer syndrome genes found that a 30X read depth achieved acceptable precision and recall for detection of substitutions but poor recall for small insertions and deletions. The same study showed that exome-derived and genome-derived datasets exhibited low read depth for different gene exons, meaning that a variant may be reliably detected in one dataset but missed in another. Interpreters should assess read depth separately for population datasets sourced from different sequencing technologies before assigning a frequency-based classification code.

Core Principles for Setting Allele Frequency Thresholds

Maximum Credible Population Allele Frequency

The foundation of frequency-based variant interpretation is the maximum credible population allele frequency, often abbreviated as maximum credible AF. This value represents the highest allele frequency that a pathogenic variant could plausibly reach in the general population given the disorder characteristics. The calculation uses the disorder prevalence, inheritance pattern, penetrance, and genetic heterogeneity.

For an autosomal dominant disorder, the maximum credible AF is calculated using the formula that incorporates prevalence, penetrance, and the proportion of cases attributable to the gene of interest. For an autosomal recessive disorder, the calculation uses the square root of the prevalence adjusted for the proportion of cases attributable to the gene. The resulting value provides a ceiling above which a variant is unlikely to be pathogenic for that disorder.

The BA1 criterion applies when the observed allele frequency exceeds the maximum credible AF. The BS1 criterion applies when the observed allele frequency is greater than expected but does not necessarily exceed the maximum credible AF. Many laboratories set the BS1 threshold at a fraction of the maximum credible AF, often between 0.5 and 1.0 times that value, depending on the confidence in the prevalence estimate and the disorder characteristics.

Inheritance Pattern and Threshold Selection

The inheritance pattern directly influences the maximum credible AF calculation. For autosomal dominant disorders, pathogenic variants are typically rare because they are subject to purifying selection. A variant observed at a frequency of 1 percent in the general population would be highly unlikely to cause a severe autosomal dominant disorder with full penetrance. For autosomal recessive disorders, carrier frequencies can be substantially higher, particularly for founder variants in isolated populations.

A study of fibrillinopathies, which are autosomal dominant disorders, found 2653 likely pathogenic variants in 253 genes associated with autosomal dominant disorders within the gnomAD dataset. This finding demonstrates that variants classified as likely pathogenic can appear in population databases more frequently than expected. The authors noted that selection of appropriate frequency cutoffs represents a considerable challenge in variant interpretation and that neglecting this challenge may lead to incomplete or missed diagnoses.

For X-linked disorders, the threshold calculation differs because affected males are typically hemizygous and the disorder prevalence in males informs the calculation. The maximum credible AF for an X-linked disorder is typically lower than for an autosomal dominant disorder with similar prevalence because the variant is expressed in males with a single copy.

Genetic Heterogeneity and Penetrance Adjustments

Genetic heterogeneity refers to the proportion of cases of a disorder that are attributable to variants in a specific gene. If a disorder can be caused by variants in multiple genes, the maximum credible AF for any single gene must be adjusted downward. For example, if a disorder has a prevalence of 1 in 10,000 and variants in a specific gene account for 20 percent of cases, the maximum credible AF for that gene is lower than for a gene that accounts for 80 percent of cases.

Penetrance adjustments account for the probability that an individual carrying a pathogenic variant will express the disorder phenotype. Reduced penetrance allows pathogenic variants to reach higher population frequencies because not all carriers are affected and therefore are not subject to the same selective pressure. The maximum credible AF calculation should incorporate the penetrance estimate, with lower penetrance supporting a higher maximum credible AF.

Practical Workflow for Applying Frequency Criteria

Step 1: Select the Appropriate Population Database

The choice between gnomAD and 1000 Genomes depends on the specific question and the population being studied. gnomAD provides a larger aggregated dataset with exome and genome data from multiple ancestral populations. 1000 Genomes provides a smaller but well-characterized dataset with deep sequencing coverage. For most clinical variant interpretation, gnomAD is the primary resource because of its larger sample size and broader population representation.

The NCBI Data Resources provide access to multiple sequence resources and analysis services that can support variant interpretation workflows. The EMBL-EBI Training materials offer structured learning pathways for bioinformatics data resources and practical analysis education. These resources can help interpreters understand the strengths and limitations of each database.

Step 2: Assess Database Coverage and Read Depth

Before applying a frequency-based criterion, the interpreter must verify that the variant position is adequately covered in the population database. Poor coverage at a specific position can result in a false absence, leading to inappropriate application of PM2. Conversely, low-quality variant calls can produce spurious allele frequencies that incorrectly support benign classification.

The study of cancer syndrome genes demonstrated that exome-derived and genome-derived datasets can have low read depth for different gene exons. This finding means that a variant may be reliably detected in the exome dataset but missed in the genome dataset, or vice versa. Interpreters should check the read depth at the variant position in both datasets when available and should not apply PM2 if coverage is inadequate.

The same study identified a 30X read depth as the minimum acceptable threshold for detecting substitutions with acceptable precision and recall. Small insertions and deletions required higher read depth for reliable detection. Interpreters should consider the variant type when assessing whether the database coverage is sufficient to support a frequency-based classification.

Step 3: Extract and Evaluate Allele Frequency Data

The interpreter should extract the allele frequency for the variant of interest from the population database, noting the specific population subset and the filter status. The filter allele frequency, which excludes variants that fail quality filters, is often more appropriate for clinical interpretation than the raw allele frequency.

The study of admixed populations found that 43 percent of shared variants showed significantly different allele frequencies between a Brazilian cohort and gnomAD v4.1. Among these, 113 variants exhibited large effect sizes, including 39 variants of uncertain significance. Twenty of these variants had higher allele frequency in the Brazilian cohort and exceeded benignity thresholds while remaining rare in other populations. This finding demonstrates that population-specific allele frequency data can change variant classification and that reliance on a single global database may lead to inappropriate clinical management in diverse populations.

Step 4: Apply the Appropriate Frequency Criterion

Once the allele frequency is extracted and the database coverage is verified, the interpreter applies the relevant criterion based on the calculated maximum credible AF. The decision tree follows a logical sequence:

1. If the allele frequency exceeds the maximum credible AF, apply BA1 as a benign stand-alone criterion.
2. If the allele frequency is greater than expected for the disorder but does not exceed the maximum credible AF, apply BS1 as a benign supporting criterion.
3. If the allele frequency is zero or extremely low in a well-covered database, consider PM2 as a pathogenic supporting criterion, but only in combination with other evidence.
4. If the variant is observed in the homozygous state in healthy adults and the disorder is recessive with full penetrance, consider BS2 as a benign supporting criterion.

The application of PM2 requires particular caution. PM2 is a supporting criterion, meaning it is not sufficient on its own to classify a variant as pathogenic. The criterion should only be applied when the database coverage is adequate and the population representation is appropriate for the patient's ancestry.

Step 5: Document the Decision and Rationale

Reproducibility requires documentation of the database version, the population subset used, the allele frequency value, the filter status, and the coverage assessment. The documentation should include the calculated maximum credible AF and the rationale for the prevalence, penetrance, and genetic heterogeneity estimates used in the calculation.

The Galaxy Training Network provides accessible workflow training and analysis tutorials that emphasize reproducibility in genomic analysis. The nf-core Documentation describes community pipeline standards for reproducible workflow configuration. These resources support the development of documented, reproducible variant interpretation workflows.

Options and Tradeoffs in Frequency Threshold Selection

Fixed Thresholds versus Calculated Thresholds

Some laboratories use fixed allele frequency thresholds for all variants, such as 1 percent for BA1 and 0.1 percent for BS1. Fixed thresholds are simple to implement and provide consistency across variants. However, they do not account for disorder-specific characteristics and may be inappropriate for disorders with very low prevalence or for founder variants in isolated populations.

Calculated thresholds based on the maximum credible AF provide a more precise approach that accounts for disorder prevalence, inheritance pattern, penetrance, and genetic heterogeneity. The tradeoff is increased complexity and the need for reliable prevalence estimates. For disorders with uncertain prevalence, the calculated threshold may vary substantially depending on the estimate used.

Global Frequency versus Population-Specific Frequency

The choice between global allele frequency and population-specific allele frequency is critical for accurate variant interpretation. A variant that is rare globally may be common in a specific ancestral population due to founder effects. Conversely, a variant that is common globally may be rare in a specific population.

The study of admixed populations demonstrated that incorporating regional reference datasets can reduce uncertainty and resolve conflicting variant classifications. The authors argued for routine incorporation of regional reference datasets in diagnostic curation to reduce uncertainty and avoid inappropriate clinical management in diverse populations. Interpreters should use the population-specific allele frequency that matches the patient's ancestry when available, and should document the population subset used in the analysis.

Exome-Derived versus Genome-Derived Frequencies

gnomAD provides both exome-derived and genome-derived allele frequencies. These datasets can differ in coverage and variant detection sensitivity for specific genomic regions. The study of cancer syndrome genes found that exome-derived and genome-derived datasets exhibited low read depth for different gene exons. Interpreters should compare the allele frequency across both datasets when available and should investigate discrepancies before applying a frequency-based criterion.

The study also found that individual variants were mostly assigned to non-divergent frequency bins, with more than 95 percent of variants assigned to the same bin using allele frequency and more than 97 percent using filter allele frequency. However, two major bin divergences were resolved by applying the minimal acceptable read-depth threshold. This finding supports the practice of checking read depth before applying frequency-based criteria.

Observations and Measurements for Quality Control

Read Depth Verification

Read depth at the variant position is the primary quality metric for population frequency data. The interpreter should verify that the variant position has adequate read depth in the population database before applying PM2 or BS1. The minimum acceptable read depth depends on the variant type, with substitutions requiring lower depth than small insertions and deletions.

The cancer syndrome gene study identified 30X read depth as achieving acceptable precision and recall for substitutions but poor recall for small insertions and deletions. This finding suggests that interpreters should be more cautious when applying PM2 for small insertions and deletions, particularly if the database coverage at the position is marginal.

Filter Status Assessment

The filter status of a variant in the population database indicates whether the variant passed quality filters. Variants that fail quality filters should not be used to support frequency-based classification. The filter allele frequency, which excludes variants that fail quality filters, is often more appropriate for clinical interpretation.

Interpreters should record both the raw allele frequency and the filter allele frequency for each variant. Discrepancies between these values may indicate quality issues at the variant position that warrant further investigation.

Population Representation Assessment

The population representation in the database should match the patient's ancestry for accurate frequency interpretation. gnomAD provides allele frequencies for multiple ancestral populations, including African/African American, Amish, Ashkenazi Jewish, East Asian, Finnish, Latino/Admixed American, Middle Eastern, Non-Finnish European, and South Asian populations.

The study of admixed populations found that 43 percent of shared variants showed significantly different allele frequencies between a Brazilian cohort and gnomAD v4.1. This finding highlights the importance of using population-specific data when available and of recognizing that global databases may not adequately represent all populations.

Records and Documentation Standards

Variant Interpretation Record

Each variant interpretation should include a record of the population frequency evidence used in the classification. The record should document:

1. Database name and version (for example, gnomAD v4.1)
2. Population subset used for the allele frequency
3. Raw allele frequency and filter allele frequency
4. Read depth at the variant position
5. Coverage assessment for the variant type
6. Calculated maximum credible AF and the inputs used
7. Frequency criterion applied (BA1, BS1, PM2, or BS2)
8. Rationale for the threshold selection

The Carpentries Lessons provide foundational training in data management and reproducible analysis practices that support consistent documentation standards.

Threshold Calculation Record

The maximum credible AF calculation should be documented with the specific inputs used. The record should include the disorder prevalence estimate, the inheritance pattern, the penetrance estimate, and the genetic heterogeneity estimate. The source of each estimate should be cited, and the date of the calculation should be recorded.

For disorders with uncertain prevalence, the interpreter should document the range of plausible maximum credible AF values and the threshold selected. Sensitivity analysis, in which the threshold is varied across the plausible range, can help identify variants whose classification is robust to threshold uncertainty.

Common Failure Patterns in Frequency-Based Interpretation

Applying PM2 Without Coverage Verification

A common error is applying PM2 when the variant position has inadequate coverage in the population database. A variant that appears absent may simply be undetected due to low read depth. This error can lead to inappropriate pathogenic classification when PM2 is combined with other weak evidence.

The cancer syndrome gene study demonstrated that exome-derived and genome-derived datasets can have low read depth for different gene exons. Interpreters should verify coverage at the variant position in both datasets before applying PM2. If coverage is inadequate, PM2 should not be applied.

Using Global Frequency for Population-Specific Variants

Another common error is using the global allele frequency when a population-specific frequency is more appropriate. A variant that is rare globally may be common in a specific ancestral population due to founder effects. Using the global frequency may incorrectly support PM2 when the variant is actually common in the patient's ancestral population.

The study of admixed populations found that 20 variants of uncertain significance had higher allele frequency in the Brazilian cohort and exceeded benignity thresholds while remaining rare in other populations. This finding demonstrates that global frequency data can be misleading for variants that are common in specific populations.

Ignoring Filter Status

Variants that fail quality filters in the population database should not be used to support frequency-based classification. The raw allele frequency may include low-quality variant calls that inflate the apparent frequency. Interpreters should use the filter allele frequency for clinical interpretation and should investigate discrepancies between raw and filter allele frequencies.

Applying Frequency Criteria Without Calculating Maximum Credible AF

Some interpreters apply fixed frequency thresholds without calculating the maximum credible AF for the specific disorder. This approach can lead to inappropriate classification for disorders with very low prevalence or for genes with high genetic heterogeneity. The maximum credible AF calculation provides a disorder-specific threshold that accounts for the relevant factors.

The fibrillinopathy study found that likely pathogenic variants in genes associated with autosomal dominant disorders may be more frequent than expected in gnomAD. This finding suggests that fixed thresholds may not adequately account for the complexity of variant interpretation and that disorder-specific threshold calculations are preferable.

Limitations of Population Frequency Data

Database Representation and Ancestry Diversity

Population frequency databases have historically under-represented non-European populations. The study of admixed populations found that under-representation of non-European groups hinders accurate variant interpretation. This limitation is particularly relevant for hereditary cancer genes, where variants of uncertain significance are common and population-specific allele frequency data can change classification.

Interpreters should recognize that absence from a population database does not prove that a variant is absent from the population. The absence may reflect inadequate representation of the patient's ancestral population in the database. This limitation is particularly important when applying PM2 for patients from under-represented populations.

Variant Type Limitations

Population databases have different detection sensitivities for different variant types. The cancer syndrome gene study found that 30X read depth achieved acceptable precision and recall for substitutions but poor recall for small insertions and deletions. This finding suggests that PM2 should be applied with greater caution for small insertions and deletions than for substitutions.

The evaluation of in silico pathogenicity prediction tools for small in-frame indels found that tools performed well enough to aid clinical variant classification in a similar manner to missense prediction tools. However, the study also found that the area under the ROC curve decreased substantially when evaluating variants not present in the tools' training data. This finding highlights the importance of considering variant type when applying frequency-based criteria.

Founder Variants and Population-Specific Pathogenic Variants

Founder variants are pathogenic variants that are common in specific populations due to a common ancestor. These variants may exceed the maximum credible AF calculated for the general population but may still be pathogenic in the specific population. The Lesch-Nyhan syndrome study demonstrated the use of population frequency databases in filtering candidate pathogenic variants in a clinical setting. Interpreters should recognize that founder variants require population-specific interpretation and that global frequency thresholds may not be appropriate.

Database Version Differences

Population databases are updated regularly, and allele frequencies can change between versions. A variant that is absent from an older database version may be present in a newer version with expanded population representation. Interpreters should document the database version used and should re-evaluate variant classifications when new database versions are released.

The study of admixed populations used gnomAD v4.1, demonstrating the importance of using current database versions. Interpreters should establish a process for periodic re-evaluation of variant classifications when population databases are updated.

Safety and Regulatory Context

Clinical Reporting Requirements

Variant interpretation for clinical purposes must follow established guidelines and standards. The ACMG/AMP guidelines provide the framework for variant classification, and laboratory accreditation standards require documentation of the evidence used for classification. The Lesch-Nyhan syndrome study described the use of the ACMG Standards and Guidelines for the Interpretation of Sequence Variants (2015) for pathogenicity classification.

Interpreters should ensure that their variant interpretation workflow complies with applicable regulatory requirements and laboratory accreditation standards. The documentation should be sufficient to support the classification in a clinical report and should be reproducible by another interpreter.

Professional Escalation Criteria

Interpreters should escalate to a senior colleague or a variant curation expert when:

1. The variant classification changes based on the population frequency threshold selected
2. The variant is common in a specific population but rare globally
3. The variant position has inadequate coverage in the population database
4. The disorder prevalence or penetrance estimates are highly uncertain
5. The variant is a founder variant in a specific population
6. The variant classification has implications for clinical management

The ClinGen Variant Curation Expert Panels provide gene-specific or disease-specific variant curation rules that can support classification decisions. The study of admixed populations described the use of ClinGen VCEP rules to downgrade candidate variants of uncertain significance and resolve conflicting calls.

Practical Implementation Steps for Laboratories

Establish a Threshold Calculation Protocol

Laboratories should establish a written protocol for calculating maximum credible AF thresholds. The protocol should specify the formula to use, the sources for prevalence and penetrance estimates, and the process for documenting the calculation. The protocol should be reviewed and updated regularly to reflect current evidence.

Define Database Selection Criteria

Laboratories should define criteria for selecting the population database and the population subset for each variant interpretation. The criteria should specify when to use gnomAD versus 1000 Genomes, when to use exome-derived versus genome-derived frequencies, and when to use population-specific frequencies.

Implement Coverage Verification

Laboratories should implement a process for verifying coverage at the variant position before applying frequency-based criteria. The process should include checking read depth in both exome-derived and genome-derived datasets when available and should specify the minimum acceptable read depth for different variant types.

Document Population Frequency Evidence

Laboratories should document the population frequency evidence for each variant classification in a structured format. The documentation should include the database version, population subset, allele frequency values, filter status, read depth, and the threshold calculation. The documentation should be sufficient to support the classification in a clinical report.

Establish Re-Evaluation Processes

Laboratories should establish processes for re-evaluating variant classifications when population databases are updated. The re-evaluation process should prioritize variants whose classification is sensitive to allele frequency changes and variants in genes with high clinical actionability.

A Structured Decision Framework for Frequency-Based Variant Classification

Applying population frequency criteria in variant interpretation requires more than selecting a threshold and comparing it to an observed allele frequency. Interpreters must integrate multiple lines of evidence, account for database limitations, and document decisions in a way that supports reproducibility. This section provides a structured decision framework that moves from threshold calculation through evidence integration to final classification, with specific attention to the practical decisions that arise when frequency data conflict with other evidence.

The Frequency Evidence Integration Matrix

A practical approach to frequency-based classification uses a matrix that combines the calculated maximum credible allele frequency with the observed population frequency and the quality of the evidence. The matrix helps interpreters avoid the common error of applying a single criterion in isolation without considering how frequency evidence interacts with other ACMG/AMP criteria.

The matrix has three dimensions. The first dimension is the observed allele frequency relative to the maximum credible AF. The second dimension is the confidence in the frequency measurement, which depends on read depth, filter status, and population representation. The third dimension is the strength of other evidence for or against pathogenicity, including functional studies, segregation data, and in silico predictions.

For each combination of these dimensions, the matrix specifies a recommended action. When the observed allele frequency exceeds the maximum credible AF and the frequency measurement is high confidence, BA1 applies as a stand-alone benign criterion regardless of other evidence. When the observed allele frequency is moderately elevated but does not exceed the maximum credible AF, BS1 applies as supporting evidence for benign classification, but other evidence may still support pathogenicity if it is strong.

When the observed allele frequency is zero or extremely low and the frequency measurement is high confidence, PM2 applies as supporting evidence for pathogenicity. However, PM2 must be combined with at least one other pathogenic criterion to reach a likely pathogenic classification. When the frequency measurement is low confidence due to inadequate read depth or poor population representation, no frequency criterion should be applied, and the interpreter should document the reason for excluding frequency evidence.

Building a Frequency Evidence Worksheet

A frequency evidence worksheet provides a structured format for recording the information needed to apply frequency criteria consistently. The worksheet should be completed for every variant where population frequency data is considered, regardless of whether a frequency criterion is ultimately applied.

The worksheet begins with variant identification information, including the genomic position, reference and alternate alleles, and the gene name. The next section records the population database details, including the database name, version, and the population subset used for the frequency assessment. The interpreter records both the raw allele frequency and the filter allele frequency, along with the read depth at the variant position and the coverage assessment for the variant type.

The worksheet includes a section for the maximum credible AF calculation. This section records the disorder prevalence estimate, the inheritance pattern, the penetrance estimate, and the genetic heterogeneity estimate. The source of each estimate is documented, and the date of the calculation is recorded. The calculated maximum credible AF is recorded, along with any sensitivity analysis that was performed.

The final section of the worksheet records the frequency criterion applied and the rationale for the decision. If no frequency criterion is applied, the worksheet documents the reason, such as inadequate coverage or uncertain population representation. The worksheet serves as the primary record for the frequency evidence used in the variant classification and supports re-evaluation when databases are updated.

Handling Conflicting Frequency and Functional Evidence

A common challenge in variant interpretation is the conflict between frequency evidence and functional evidence. A variant may have a population frequency that exceeds the maximum credible AF, suggesting a benign classification, but functional studies may indicate a deleterious effect. Conversely, a variant may be absent from population databases, supporting PM2, but functional studies may show no effect on protein function.

The ACMG/AMP framework does not provide a simple algorithm for resolving these conflicts. The interpreter must weigh the strength and quality of each line of evidence. The frequency evidence integration matrix provides a structured approach by considering the confidence in the frequency measurement and the strength of the conflicting evidence.

When the frequency evidence is high confidence and the observed allele frequency clearly exceeds the maximum credible AF, BA1 applies as a stand-alone benign criterion. In this situation, the functional evidence should be re-examined for potential artifacts or misinterpretation. When the frequency evidence is moderate confidence and the observed allele frequency is near the maximum credible AF, the interpreter should consider whether the functional evidence is strong enough to override the frequency evidence.

The fibrillinopathy study demonstrated that likely pathogenic variants can appear in population databases more frequently than expected. The authors noted that selection of appropriate frequency cutoffs represents a considerable challenge in variant interpretation. This finding supports a cautious approach when frequency evidence conflicts with other evidence, particularly for variants in genes associated with autosomal dominant disorders.

Population-Specific Threshold Adjustments

The maximum credible AF calculation assumes a single threshold for all populations. However, population-specific factors can affect the appropriate threshold for a given variant. Founder variants, which are pathogenic variants that are common in specific populations due to a common ancestor, may exceed the maximum credible AF calculated for the general population but may still be pathogenic in the specific population.

The study of admixed populations found that 20 variants of uncertain significance had higher allele frequency in a Brazilian cohort and exceeded benignity thresholds while remaining rare in other populations. This finding demonstrates that population-specific allele frequency data can change variant classification and that reliance on a single global database may lead to inappropriate clinical management in diverse populations.

Interpreters should adjust the frequency threshold when the patient's ancestry is known and the population database provides data for that ancestral population. The adjustment should consider whether the variant is a known founder variant in the population and whether the disorder prevalence differs between populations. The frequency evidence worksheet should document the population-specific threshold and the rationale for the adjustment.

Re-Evaluation Triggers and Version Control

Population databases are updated regularly, and allele frequencies can change between versions. A variant that is absent from an older database version may be present in a newer version with expanded population representation. Interpreters should establish triggers for re-evaluating variant classifications when population databases are updated.

The primary trigger for re-evaluation is the release of a new database version with substantially expanded population representation or improved coverage. The study of admixed populations used gnomAD v4.1, demonstrating the importance of using current database versions. Laboratories should track database version releases and prioritize re-evaluation of variants whose classification is sensitive to allele frequency changes.

A second trigger for re-evaluation is a change in the disorder prevalence or penetrance estimates used in the maximum credible AF calculation. New epidemiological studies may provide more accurate prevalence estimates, which can change the calculated threshold and affect variant classification.

A third trigger is the identification of the variant as a founder variant in a specific population. Founder variant status may become known through population-specific studies or through the accumulation of clinical evidence. When a variant is identified as a founder variant, the frequency threshold should be adjusted to account for the population-specific prevalence.

The frequency evidence worksheet should include a version control section that records the database version, the date of the assessment, and the date of any re-evaluation. This documentation supports the reproducibility of the variant classification and provides a clear record of when and why classifications changed.

Troubleshooting Common Frequency Data Discrepancies

Interpreters frequently encounter discrepancies between different population databases or between different datasets within the same database. The cancer syndrome gene study found that exome-derived and genome-derived datasets can have low read depth for different gene exons. This finding means that a variant may be reliably detected in one dataset but missed in another.

When the allele frequency differs substantially between exome-derived and genome-derived datasets, the interpreter should investigate the cause before applying a frequency criterion. The first step is to check the read depth at the variant position in both datasets. If the read depth is inadequate in one dataset, the frequency from the higher-quality dataset should be used.

The second step is to check the filter status in both datasets. A variant that fails quality filters in one dataset but passes in another may have a genuine frequency difference or may reflect a quality issue. The interpreter should compare the raw and filter allele frequencies in both datasets to identify potential quality issues.

The third step is to check the population representation in both datasets. If the variant is common in a specific ancestral population, the frequency may differ between datasets because of differences in population composition. The interpreter should use the population-specific frequency that matches the patient's ancestry when available.

When discrepancies cannot be resolved through these steps, the interpreter should consider the variant position uninterpretable for frequency evidence and should document the reason. The frequency evidence worksheet should record the discrepancy and the investigation performed.

Escalation Criteria for Frequency Evidence Challenges

Certain situations warrant escalation to a senior colleague or a variant curation expert. The first situation is when the variant classification changes based on the frequency threshold selected. This situation indicates that the classification is sensitive to the threshold calculation and that the threshold should be reviewed by an expert.

The second situation is when the variant is common in a specific population but rare globally. This situation requires population-specific interpretation and may involve founder variant considerations that are beyond the scope of routine interpretation.

The third situation is when the variant position has inadequate coverage in the population database. This situation requires a decision about whether to apply PM2 based on the available evidence or to exclude frequency evidence entirely.

The fourth situation is when the disorder prevalence or penetrance estimates are highly uncertain. This situation requires a sensitivity analysis to determine whether the variant classification is robust to the uncertainty in the threshold calculation.

The fifth situation is when the variant classification has implications for clinical management. Variants that would change a clinical diagnosis, treatment recommendation, or reproductive counseling should be reviewed by an expert before the classification is finalized.

The ClinGen Variant Curation Expert Panels provide gene-specific or disease-specific variant curation rules that can support classification decisions. The study of admixed populations described the use of ClinGen VCEP rules to downgrade candidate variants of uncertain significance and resolve conflicting calls. Interpreters should consult these rules when available and should escalate to the relevant expert panel when the rules do not resolve the classification question.

Frequently Asked Questions

What is the difference between BA1 and BS1 in the ACMG/AMP framework?

BA1 is a benign stand-alone criterion that applies when the variant allele frequency exceeds the maximum credible population allele frequency for the disorder. BS1 is a benign supporting criterion that applies when the variant allele frequency is greater than expected for the disorder but does not necessarily exceed the maximum credible AF. BA1 alone is sufficient to classify a variant as benign, while BS1 must be combined with other evidence to support a benign classification.

How do I calculate the maximum credible allele frequency for a disorder?

The maximum credible AF is calculated using the disorder prevalence, inheritance pattern, penetrance, and genetic heterogeneity. For an autosomal dominant disorder, the calculation incorporates the prevalence, the proportion of cases attributable to the gene, and the penetrance. For an autosomal recessive disorder, the calculation uses the square root of the prevalence adjusted for the proportion of cases attributable to the gene. The resulting value represents the highest allele frequency that a pathogenic variant could plausibly reach in the general population.

When should I use population-specific allele frequency instead of global allele frequency?

Population-specific allele frequency should be used when the patient's ancestry is known and the population database provides data for that ancestral population. A variant that is rare globally may be common in a specific ancestral population due to founder effects. Using the global frequency may incorrectly support PM2 when the variant is actually common in the patient's ancestral population. The study of admixed populations found that 43 percent of shared variants showed significantly different allele frequencies between a Brazilian cohort and gnomAD v4.1.

How do I verify that a variant is truly absent from a population database before applying PM2?

Before applying PM2, verify that the variant position has adequate read depth in the population database. The cancer syndrome gene study identified 30X read depth as the minimum acceptable threshold for detecting substitutions with acceptable precision and recall. Small insertions and deletions required higher read depth for reliable detection. Check coverage in both exome-derived and genome-derived datasets when available, and do not apply PM2 if coverage is inadequate.

What is the role of filter allele frequency in variant interpretation?

The filter allele frequency excludes variants that fail quality filters in the population database. This value is often more appropriate for clinical interpretation than the raw allele frequency, which may include low-quality variant calls. The cancer syndrome gene study found that more than 97 percent of variants were assigned to non-divergent filter allele frequency bins, supporting the use of filter allele frequency for frequency-based classification.

How do founder variants affect frequency-based variant interpretation?

Founder variants are pathogenic variants that are common in specific populations due to a common ancestor. These variants may exceed the maximum credible AF calculated for the general population but may still be pathogenic in the specific population. Interpreters should recognize that founder variants require population-specific interpretation and that global frequency thresholds may not be appropriate. The Lesch-Nyhan syndrome study demonstrated the use of population frequency databases in filtering candidate pathogenic variants in a clinical setting.

What should I do when the variant classification changes based on the frequency threshold selected?

When the variant classification changes based on the frequency threshold selected, escalate to a senior colleague or a variant curation expert. Document the range of plausible maximum credible AF values and the threshold selected. Sensitivity analysis, in which the threshold is varied across the plausible range, can help identify variants whose classification is robust to threshold uncertainty. The ClinGen Variant Curation Expert Panels provide gene-specific or disease-specific variant curation rules that can support classification decisions.

How often should I re-evaluate variant classifications when population databases are updated?

Re-evaluate variant classifications when population databases are updated with substantially expanded population representation or improved coverage. Prioritize variants whose classification is sensitive to allele frequency changes and variants in genes with high clinical actionability. The study of admixed populations used gnomAD v4.1, demonstrating the importance of using current database versions. Establish a process for periodic re-evaluation and document the database version used for each classification.

Related Bioinformatics Guides

• Lipidomic Analysis: A Beginner's Guide to Workflows and Data Interpretation
• Longitudinal Microbiome Data Analysis: Methods and Best Practices
• Metagenomic Contamination Control: Best Practices for Clean Data
• Persistent Identifiers for Research Data: A Guide to Selection and Use
• RNA-Seq Databases: Accessing and Using Public RNA-Seq Data

Related Clinical & Scientific Guides

• A Practical Guide to Detecting Antimicrobial Resistance Genes in Shotgun Metagenomic Data
• Computational Immunology: Modeling the Immune System
• How to Set Hard Filters for Germline Variant Calling: A Practical Guide to GATK Best Practices

References and Further Reading

• NCBI Data Resources. National Center for Biotechnology Information.
• EMBL-EBI Training. European Bioinformatics Institute.
• Bioconductor. Bioconductor Project.
• Galaxy Training Network. Galaxy Project.
• nf-core Documentation. nf-core.
• The Carpentries Lessons. The Carpentries.
• Considerations for using population frequency data in germline variant interpretation: Cancer syndrome genes as a model.. Human mutation, 2021.
• Variant filtering, digenic variants, and other challenges in clinical sequencing: a lesson from fibrillinopathies.. Clinical genetics, 2020.
• Evaluation of in silico pathogenicity prediction tools for the classification of small in-frame indels.. BMC medical genomics, 2023.
• [[Clinical characteristics and treatment of two children with Lesch-Nyhan syndrome].](https://pubmed.ncbi.nlm.nih.gov/40763966). Zhonghua yi xue yi chuan xue za zhi = Zhonghua yixue yichuanxue zazhi = Chinese journal of medical genetics, 2025.
• Variant classification of hereditary cancer genes is affected by genomic underrepresentation of admixed populations.. Molecular genetics and genomics : MGG, 2025.

This article is educational and does not replace validated analysis plans, institutional policy, clinical interpretation, or specialist review.