REVEL Score in "Extra Annotations"

[alexejknaus]

The HumanGenetics department uses the meta REVEL score (https://www.sciencedirect.com/science/article/pii/S0002929716303706) to classify rare missense variants.

Addition of REVEL scores to missense variants in the Extra Annotations view per variant.

REVEL (rare exome variant ensemble learner) is an ensemble method for predicting the pathogenicity of missense variants on the basis of individual tools: MutPred, FATHMM, VEST, PolyPhen, SIFT, PROVEAN, MutationAssessor, MutationTaster, LRT, GERP, SiPhy, phyloP, and phastCons...
...When applied to two independent test sets, REVEL had the best overall performance (p < 10−12) as compared to any individual tool and seven ensemble methods: MetaSVM, MetaLR, KGGSeq, Condel, CADD, DANN, and Eigen. Importantly, REVEL also had the best performance for distinguishing pathogenic from rare neutral variants with allele frequencies <0.5%.

[holtgrewe]

Some questions:

1. Is this something for SNVs only or also for indels?
2. Is pre-annotation of all SNVs / known indels available?
3. Is a binary / source code available?

[alexejknaus]

This is for missense mutations only.
REVEL scores are freely available for non-commercial use. For other uses, please contact Weiva Sieh.

https://sites.google.com/site/revelgenomics/downloads

[NandithaMallesh]

Usecase: Make REVEL score available for the filtered variants and prioritize variants on REVEL score
Attached is the implementation proposal:

REVEL_preliminary_implementation.docx

[alexejknaus]

Use case 1: User opens a VCF file in Varfish for filtering.

• in prioritization tab "Pathogenicity Prioritization" he sees besides CADD score REVEL score

[alexejknaus]

Use case 2:

• Analyst filters VCF file without setting prioritization for REVEL Score
• Score is displayed in "Extra annotations" per missense variant
• for non missense variants no score is displayed

[alexejknaus]

Use case 3:

• Analyst filters VCF file sets prioritization for REVEL Score
• Variant list is sorted by REVEL score.
• Variants without revel score are at the bottom
• Score is displayed in "Extra annotations" per missense variant

[alexejknaus]

Optional Use case 4:

• Filtering with REVEL score
• user sets threshold
• only variants remain above set threshold

[holtgrewe]

Use Case 5:

• if the user downloads the filtration results as TSV/Excel then the revel score shall be visible there as well

[xiamaz]

Developer Meeting Results 2021-10-08

Current implementations for CADD utilizes multiple methods (extra annotations if added as precomputed CADD values or as a separate service called after variant filtering for display purposes)

varfish-annotator currently primarily adds population frequencies and information for initial variant filtering.

varfish-db-downloader can add information into an extra annotations field, that can be displayed for individiual variants.

Different implementations might be possible depending on whether the score needs to be recomputed for novel variants and whether that operation is especially computationally expensive or complex.

Three major ways for implementation of additonal scores were discussed.

1. info-field and addition to varfish-annotator

The info-field is a postgres jsonb field, which is currently not indexed. Implementation would require changes to add the metainformation into the field, either in varfish-annotator or in tsv postprocessing. The info-field would need to be indexed and queries adapted for using that field.

2. extra-annotations field

Currently variable number of extra-annotations are supported for variants and examples of this can be found in varfish-db-downloader. Adding a new score to extra-annotations does not require any changes to varfish-server but only addition of a field to the tsv file and the extra-annotations column with is a json-array.

Currently display of that information is only supported in a dropdown box for each variant. Support for table export and table display would need to be implemented. Additionally possiblity for usage in filtering would require adding the extra-annotations field into the query engine. Depending on the number of variants to be processed, this operation might become very expensive.

The query engine found in variants/queries.py is rather complex and while it uses optimized query plans, preformance regression tests currently do not exist. Adding these variable format queries might have unforeseen performance impacts, that might need to be monitored with additional testing.

An advantage of following that approach would be the possiblity to apply that to all currently existing scores inside of extra annotations.

2b. Separate REVEL score table

A few scores are currently implemented in a separate table. This would make later score updates much easier, as different scores are kept in separate places. Main concern would be future scalability of that approach.

3. Separate Service and API fetch

If only display in the end results were needed the REVEL score could be kept in a separate service which communicates with the varfish interface via API calls. While this would allow for good abstraction, it does not scale well for performance and could only be used for at maximum 100s of variants.

Preferred method for computationally complex scores and scores that might need to be recalculated for novel variants.

Conclusion

The need for a proper plugin infrastructure is brought up. There is currently no planned work on that.

[holtgrewe]

Thanks for summarizing this.

Re: Preamble

1. varfish-db-downloader Snakemake workflow downloads files from public sources and generates TSV files fit for
   • import into varfish-server as background database, including gnomAD, KEGG, NCBI entrez gene information etc
   • import into varfish-annotator (H2) database via varfish-annotator init-db (actually called in varfish-db-downloader Snakemake workflow)
2. varfish-annotator annotate takes H2 database and small variant/SV VCF file and annotates them and writes out TSV files fit for import the case into varfish-server
   • the fields in small variant TSV file correspond to the SmallVariant model in VarFish (there is one such record for each alternative allele and each affected gene)
   • the fields in the SV TSV files correspond to StructuralVariant and StructuralVariantAnnotation where the first defines the variant and the second defines the affected genes

Now, (1) also creates a table group "extra annotations" with two TSV files ExtraAnno.tsv and ExtraAnnoField.tsv. These are imported into VarFish as background database. They are not imported into VarFish annotator. This file can be replaced (or re-imported with another file, overwriting the previous data) by a file with different fields. Currently, the ExtraAnnoFields.tsv file looks as follows:

field   label
1       SpliceAI-acc-gain
2       SpliceAI-acc-loss
3       SpliceAI-don-gain
4       SpliceAI-don-loss
5       MMSp_acceptorIntron
6       MMSp_acceptor
7       MMSp_exon
8       MMSp_donor
9       MMSp_donorIntron
10      dbscSNV-ada_score
11      dbscSNV-rf_score
12      CADD-PHRED

When importing a modified version, the values would be displayed automatically in the per-variant fold-out.

As for (2), the output TSV files could of course be preprocessed before being imported into VarFish. E.g., they contain an info column that could be used to get arbitrary data into the VarFish database. They would not be displayed anything.

[holtgrewe]

A few minor comments on the rest:

while it uses optimized query plans

It generates large queries that are thrown at postgres which optimizes it. In the case that we discovered issues, we went to look at the query optimizer results in postgres, the generated queries and rearranged things and added indices where needed.

• Code formatters are useful, e.g., https://sqlformat.darold.net/
• As are query plan visualizers
  • https://explain.dalibo.com/
  • https://tatiyants.com/pev/#/plans/plan_1633719627206
  • https://explain.depesz.com/

Main concern would be future scalability of that approach.

"Scalability" here means ability to extend code here and/or limit technical debt.

[snesic]

Regarding 1) and a table group "extra annotations" with two TSV files ExtraAnno.tsv and ExtraAnnoField.tsv:

How about we create another snakemake script (similar to extra_annos.smk), which will output two new files ExtraAnno_revel.tsv and ExtraAnnoField_revel.tsv and then we simply import them along with the original ExtraAnno.tsv? We could also keep the same filename but place them in a different folder (extra_annos_revel/date/ExtraAnno.tsv)?

[holtgrewe]

Actually, I would like to keep the REVEL score out of varfish-db-downloader at the moment. What about you creating a standalone repository with a script that augments a "vanilla" ExtraAnno.tsv and ExtraAnnoField.tsv with the REVEL annotation?

[holtgrewe]

Has been converted to issue #242. Discussion should continue there.