Dear all,
talking about the pathogenicity predictors on cancer mutations, what algorithms or meta-predictors would you recommend to use ? Among possible choices : CADD, MutationTaster, FATHMM, CHASM, Condel CanDrA , or any other predictors/meta-predictors.
thank you,
bogdan
Take your pick...
This is not a complete listing, as there are many more.
[uses various modelling algorithms and produces PDB files, which can be loaded into protein viewers like Jmol]
For further reading:
I would recommend my recent method CHASMplus for missense mutations (see https://www.cell.com/cell-systems/fulltext/S2405-4712(19)30154-1 ). It performs better in benchmarking than other methods (even against meta-preditctors), is cancer type-specific, and you can get scores through an easy to use graphical user interface (see https://chasmplus.readthedocs.io/en/latest/quickstart_opencravat.html#install-opencravat-app ). You only need to specify your variants in a simple tab-delimited format or as VCF.
There already exist consensus predictors such as PredictSNP, which combine multiple individual pathogenicity predictors. In my experience, even installing these tools locally is a gigantic pain, and given the diversity of factors and methods used to predict pathogenicity, getting a consensus result out of them is a larger pain. Fact of the matter is, most of these tools are good in predicting benign changes, but overly cautious in predicting pathogenic/deleterious results, resulting in a whole lot of false positives.
Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
version 2.3.6
this just came out today: https://genomebiology.biomedcentral.com/articles/10.1186/s13059-017-1353-5
Great timing!
thank you gentlemen !
If you are additionally interested in creating PDB files from novel/mutated amino acid sequences, and then checking how protein conformation may have changed due to the mutation, then look at the Protein Model Portal. I have added this to my list below.
Do you know if there is a paper that assesses the performance of this approach on somatic mutations? Analyzing mutational clustering in protein structures has shown to perform well, but I'm not aware of successful methods taking a pure biophysical/protein conformation approach for cancer.
please take a look here : http://bioinformatics.burnham.org/pages/publication.html;
especially : https://www.ncbi.nlm.nih.gov/pubmed/28714987
or : https://academic.oup.com/nar/article/43/D1/D968/2438384
I tweaked the wording of my reply so it is less ambiguous. I was actually talking about the approach Kevin suggested by analyzing protein conformation changes when the actual amino acid is substituted in the protein structure. I actually know Eduard personally (the first author on the papers you linked), and I developed HotMAPS which looks at mutational clustering in protein structures (https://www.ncbi.nlm.nih.gov/pubmed/27197156 ).
Thank you for the link to HotMAPS. We have been doing some whole genome sequencing analysis and we hope to link at some moment the mutation to the changes in the protein conformation.
Hi Collin - great work. I will take a read.
Are you interested in somatic mutations or germline mutations? The answer depends on your intended use.
Thank you Collin for your question : we would primarily be interested in somatic mutations.
The top 4 I would recommend for missense mutations would be CHASM, CanDrA (version "plus", with "cancer-in-general"), FATHMM cancer, or ParsSNP. From examining prior benchmarks and my own benchmarks, these seem to perform better. Some methods which are designed for germline mutations are decent (eg., VEST3 and REVEL), but generally the cancer focused methods are better.
thank you Collin. For Cancer Somatic mutations, could we also use some pathogenicity predictors like CADD and MCAP ? (that initially have been designed for germline mutations) .
I've personally aggregated a set of 8 benchmarks for missense mutations comprising in vitro experiments, in vivo experiments, and literature curated databases (OncoKB). CADD and MCAP didn't perform as well.
Thank you. I will look into : CHASM, CanDrA, FATHMM cancer, or ParsSNP. Talking about REVEL -- does it do a good work on somatic mutations ?
It does the best that I've seen for methods not tailored to cancer/somatic mutations. I'd recommend to stick with the cancer specific predictors unless you need to assess some other type of alteration that is not missense.
Is it correct to use SurfR to analyze intronic variants (from an exome sequencing)?
Yes, I believe you can use it for these