FIND: a software tool for identifying population-enriched pathogenic variants in gnomAD

Hidden in Plain Sight: Uncovering the Genetic Signatures of Ancestry

Genetic medicine often struggles with a fundamental imbalance. Our understanding of disease is heavily skewed toward populations that have been extensively sequenced. Many other ancestries remain underrepresented. This leaves their specific disease burdens obscured. This is particularly problematic when dealing with founder mutations. These are variants that arose in a single ancestor. They become highly prevalent within a descendant population due to historical bottlenecks or endogamy (the practice of marrying within a specific social or ethnic group).

Identifying these mutations is clinically transformative. When a physician knows a specific mutation is enriched in a patient's ancestral group, they can move from broad screening to targeted diagnostics. This can motivate patients to pursue cascade testing (testing family members who may carry the same mutation). However, these population-specific signals often remain buried in massive repositories like gnomAD (a global database of human genetic variation).

The difficulty of mining population-specific signals

The challenge is not a lack of data. The challenge is a lack of signal clarity. In a massive dataset like gnomAD, a rare mutation might appear in a few individuals from an underrepresented group. It is difficult to distinguish a genuine population-specific trend from mere statistical noise.

Current approaches to finding these variants typically require bespoke analytic pipelines. These require significant programming expertise. In many cases, they require privileged access to secure biobanks. This creates a barrier for clinicians and bench scientists. They may recognize a clinical pattern in their patients but lack the computational infrastructure to prove it across a global dataset. Furthermore, distinguishing a true "founder" from a "candidate founder" is theoretically complex. Confirming a founder mutation strictly requires haplotype analysis (examining the specific patterns of DNA surrounding a mutation to trace it back to a single ancestor). This step is not always possible with aggregate data.

Automating the search for enrichment

To bridge this gap, the authors developed FIND (Founder candidates hidden IN Data). This is a Python-based web tool. It automates the systematic screening of gnomAD v4.1 for population-enriched variants. Instead of requiring users to write complex queries, FIND allows researchers to input up to ten genes. It then provides a structured report of potentially significant mutations.

The core mechanism of FIND relies on a disciplined filtering architecture. This separates meaningful biological signals from random occurrences. As illustrated in the conceptual logic of the tool, the process follows several rigorous stages:

1. Pathogenicity Filtering: The tool restricts its search to variants classified as pathogenic or likely pathogenic in ClinVar (a public archive of clinical interpretations). It also includes predicted loss-of-function (pLoF) variants. These are mutations that disrupt the protein product, such as frameshifts or premature stop codons.
2. Noise Reduction: To prevent sparse, low-count variants from creating false signals, FIND employs a "zeroing" threshold. Any population where a variant is observed in four or fewer individuals is treated as having a frequency of zero.
3. Stringent Enrichment Criteria: The tool identifies a variant as enriched only if its frequency in one specific ancestry group is at least tenfold higher than in any other group. It also requires a minimum frequency threshold of 0.00008 in the enriched population.

By applying these mathematical constraints, FIND ignores the background noise of rare, sporadic mutations. It focuses on the peaks where a variant is disproportionately common in a single group.

Validating the signal with known founders

The utility of FIND was tested by querying five actionable genes: FLNC, TMEM127, MYH7, BRCA1, and BRCA2. The authors report that the tool successfully recovered twelve well-known founder or population-enriched mutations. This provided immediate proof of concept.

More importantly, the tool uncovered variants that had escaped previous systematic scrutiny. The researchers found five variants documented as "recurrent" in specific populations. However, these had never undergone a formal cross-population frequency comparison. Additionally, the tool surfaced three variants that had not been previously reported as population-enriched at all.

As shown in the results interface, the output provides a clear breakdown of allele frequencies across different ancestry groups.

To ensure these findings were not artifacts of the gnomAD dataset, the authors validated the enrichment of variants in African American and Admixed American (Latino) populations using the All of Us database. Because All of Us deliberately oversamples these historically underrepresented groups, the replication of FIND's findings in an independent dataset provides strong evidence for the signals.

Constraints of the algorithmic approach

While FIND represents a significant step, it is not a definitive tool for establishing evolutionary history. The authors are transparent about several critical limitations:

• Dependency on Existing Classifications: FIND relies on ClinVar for pathogenicity labels. It will inherently miss variants that are clinically significant but not yet formally classified. This is a notable drawback for understudied populations. In these groups, the classification of "variants of unknown significance" (VUS) often lags behind.
• The "Tenfold" Blind Spot: The strict requirement for a tenfold difference in frequency limits the tool. FIND cannot detect variants that act as founders in two different, unrelated populations. It also misses mutations that arose before major population divergences occurred.
• Discrete vs. Continuous Ancestry: The tool treats ancestry as a set of discrete buckets. Human genetic ancestry is actually a continuous spectrum. This approximation can sometimes mask subtle variations or lead to misclassification in highly admixed populations.

A tool for clinical democratization

The verdict is clear: FIND is an effective discovery engine for candidate founder mutations. It does not replace the specialized haplotype analysis required to prove a mutation's evolutionary origin. Instead, it serves as a powerful filter. It points clinicians and researchers toward the most promising targets.

By automating a tedious and error-prone manual process, FIND moves us closer to a reality where genetic screening is more equitable. It helps identify clinically relevant variants in understudied groups. The source code is openly available on GitHub under an MIT license. A functional web interface can be accessed at ethnic-variant-mutation-finder.onrender.com.