The latest version of Franklin is live, and we’re excited to announce several new prediction updates, as well as other feature enhancements.
Updated prediction guidelines
In order to shed some light on ACMG classification rules based on in silico predictors, the ClinGen Sequence Variant Interpretation (SVI) working group team has provided evidence-based revisions for the application of the PP3 and BP4 criteria.
As a result, several changes have already been implemented on Franklin, based on the new recommendations of the SVI group.
Continue reading to learn more about the modified guidelines as well as Franklin’s updates.
The background
Since their inclusion in the ACMG classification guidelines published in 2015, in silico prediction tools have become a standard step in the variant assessment process. These algorithms are designed to predict whether a variant has a disruptive effect on the final gene product, and can be particularly useful in circumstances with little to no genetic or functional evidence.
The original recommendations state that if “multiple lines of computational evidence” support either pathogenic or benign classification, then the corresponding variant can be assigned the lowest level of evidence, PP3 (Supporting Pathogenic) or BP4 (Supporting Benign), respectively. As with any other Supporting criteria, these must be combined with several other lines of evidence to tilt the classification into a Pathogenic or Benign one. Hence, while in silico predictors alone are not capable of determining the pathogenicity of a variant, they can provide a useful contribution to the overall classification.
However, there are several challenges to applying PP3 and BP4 rules in practice, which can lead to overstating or understating the strength of computational evidence. Firstly, the PP3 and BP4 criteria refer to multiple independent algorithms that agree on classification, but several of the most commonly used predictive tools overlap in terms of methods and training sets, leading to circularities. On the other hand, some of the models were developed as theoretical tools, using training datasets consisting of well-studied genes, and should be validated before they can be considered generalizable and applicable in clinical settings. Overall, the 2015 guidelines did not provide standardized practices to assess the strength of computational tools for pathogenicity classification, resulting in inconsistent and even incorrect classification.
The new recommendations
A new publication by the ClinGen Sequence Variant Interpretation (SVI) working group provides recommendations regarding the application of PP3 and BP4, after calibrating thresholds for 13 of the most common in silico predictors for pathogenicity classification of missense variants. The study aims to establish a new standard that converts tool scores to evidence strengths, including levels beyond the original ACMG/AMP recommendation of Supporting, so that computational evidence could be used consistently and appropriately across different diagnostic centers.
The predictors studied by Pejaver et al. were CADD, Evolutionary Action, FATHMM, GERP++, MPC, MutPred2, PhyloP, PolyPhen-2, PrimateAI, SIFT, and VEST4, and the meta-predictors BayesDel and REVEL.
After performing comprehensive statistical analyses using carefully assembled independent data sets of missense variants from gnomAD and ClinVar, the research group estimated thresholds for all the in silico tools mentioned, corresponding to the eight ACMG strength levels, as shown in the image below.
As it can be seen from the chart, the researchers identified thresholds for Supporting and Moderate levels of evidence for pathogenicity and benignity for all tools, except for GERP++, which did not yield Supporting evidence for PP3, and MPC, which did not yield Supporting evidence for BP4. Moreover, the table shows that several tools could provide Strong evidence for pathogenicity (BayesDel, VEST4), benignity (CADD), or both (MutPred2, REVEL). The meta-predictor REVEL showed the widest range of performance, reaching Strong levels for PP3 and even Very Strong levels for BP4. Interestingly, the thresholds found by the SVI working group did not always align with the scores recommended by the tools’ developers, even for Supporting evidence strengths.
Based on these results, the scientific team provided the following recommendations for evidence-based revisions of the ACMG/AMP criteria when applying the PP3 and BP4 rules to missense variants:
The changes in Franklin
At Genoox, our Franklin curation specialists work tirelessly to keep the platform updated to match the latest clinical guidelines. Specifically for the ACMG classification, we make sure to swiftly incorporate all new ClinGen recommendations as soon as possible. Therefore, we implemented some changes based on the new recommendations of the SVI group. The resulting main changes in Franklin are:
Adjusted thresholds for all prediction tools mentioned in the study to match the new recommendations. In addition, Franklin’s aggregated prediction has been modified accordingly.
Changes in the implementation of PP3 and BP4 rules. Franklin currently automatically supports higher strength of evidence for the PP3 rule according to the recommendations. In contrast, and as a step of caution, the default strength will remain as Supporting for BP4, and the user may change its strength manually based on their judgment.
Franklin now uses REVEL as its default tool for missense variants, as it achieved the best results among all tools. Franklin's cautious approach will only apply the BP4 rule if the score is less than 0.15, and will give it a Supporting evidence strength.
Franklin will continue to use other prediction tools that are specifically developed for different scenarios, including:
APOGEE and MitoTip for mitochondrial variants.
CardioBoost for gene-specific predictions, when available.
SpliceAI and dbscSNV for variants overlapping splice regions.
Considerations when combining the PP3 rule with the PM1 criterion, limiting the strength of the computational prediction to Supporting. The PM1 rule is applied when the variant is located in a functional domain or mutational hotspot. To avoid double-counting, the consortia recommended that the total strength of the PM1 and the PP3 criteria combined be Strong at maximum. However, users can manually upgrade the PP3 evidence strength level to Moderate if considered appropriate.
If you’d like to learn more about how Franklin applies PP3 and BP4 rules, head to our Help Center to find a more detailed explanation, or contact our team to set up a meeting.
Other improvements
Drafts of somatic variant classifications. This new feature enables users to submit a classification draft, while an additional reviewer can approve. It allows organizations to control and supervise the classification process by providing different users permissions levels and authorizations.
A new feature provides suggested classification for your future cases. When there is already an approved classification in your org for the same variant, it will appear as a suggested classification on the variant tile. All you need to do is add the variant to the report and it will get the approved classification.
The RefSeq transcripts were updated to the latest version
Streamline case creation from ION Reporter is now supported
Complete repeats support on WGS
All SpliceAI scores are now available, while previously Franklin would show only the more significant ones
We hope you find these updates useful! We encourage you to give them a try and share your thoughts. We appreciate your feedback.
If you’re interested in learning more about these new features, or want to explore our Franklin Premium version, feel free to schedule a call with us!
- The Franklin Team |