Need suggestions about pathogenicity prediction of gdc level 3 SNV file

Question

[SUMIT]

Hi, I am trying to figure out which tool is most accurate in terms of pathogenicity prediction of TCGA SNVs level 3 data. TCGA offers SIFT, PolyPhen, and IMPACT scores for different kinds of mutations. SIFT, and PolyPhen cover mainly "Missense Mutation", while IMPACT categorizes every kind of mutation into MODERATE", "HIGH", "MODERATE", "MODIFIER".

I am ok with the IMPACT classification, however, I want numerical values rather than subjective classification like "MODERATE", "HIGH", "MODERATE", "MODIFIER". I explore some other tools like PROVEN, CAPICE, etc but all of them use GrCH37 rather than NCBI-built GrCh38 since later is used in the TCGA data.

So I am confused as to what tool I should use to find out the pathogenicity level of Missense Mutation, Silent, Frame Shift, INDEL etc.

Sample FIle from gdc portal for reference

[German.M.Demidov]

A small remark: numerical values does not prove pathogenicity, ACMG criteria are currently used for the evaluation.

Usually HIGH impact in important gene and absence in population is enough to say it is a pathogenic variant. MODERATE are way more tricky.

If you want to predict cancer driverness of these mutations (since there are many of them per tumor), you may use some computational prediction.

[SUMIT]

numerical values does not prove pathogenicity

I agree but I needed numerical values to fit it with other datasets. Would you like to suggest some computational tools that are based on GrCh38?

Usually HIGH impact in important gene and absence in population is enough to say it is a pathogenic variant. MODERATE are way more tricky. Exactly I found little help at Ensembl but I donot think it is not sufficient.

I found VEP and ANNOVAR, which may work, I am going to try them and get back to you.

May I ask one question, suppose I predicated some score using xyz tool, then how would I verify that my prediction is correct? Thank you in advance

[German.M.Demidov]

I predicated some score using xyz tool, then how would I verify that my prediction is correct?

The easiest answer would be you evaluate your variant with ACMG criteria https://www.acmg.net/docs/Standards_Guidelines_for_the_Interpretation_of_Sequence_Variants.pdf - even though it is not perfect

I found VEP and ANNOVAR, which may work, I am going to try them and get back to you.

Yeap, we also use VEP for annotation.

This is the paper you may find of interest: https://www.cell.com/cell/fulltext/S0092-8674(18)30237-X?utm_campaign=STMJ_1522958526_SC&utm_channel=WEB&utm_source=WEB&dgcid=STMJ_1522958526_SC

[SUMIT]

Dear German.M.Demidov Thank you so much for your time and kind assistance. Definitely, both papers look good, I will read and follow them.

Thank you

Sumit