Hi Glynis Frans

There are several options for you.

Firstly you can use KEGG disease database

http://www.genome.jp/kegg/disease/

Also go through OMIM database

Online Mendelian Inheritance in Man

http://www.omim.org/

You can also filter your SNPs using Cosmic database which is related to cancer.

If you let me know which kind of patients you are investigating then I will be able to suggest you more precisely.

Thank you Ishtiaq! We are looking for mutations in genes that can cause Mendelian inherited primary immunodeficiencies. I already check our WES data for the known (= published) genes by filtering them out by a program that I created in Visual Basic. But some patients have a familial history of PID but still none of the known genes pop up, so we are also looking for new genetic defects. Unfortunately, at the moment we are having a hard time identifying possible genetic candidates...

Hi Glynis,

It seems you are up against the sort of problem those of us who were gene hunting 25 years ago came up against - how do you find a gene when you don't know what you're looking for?! If the genetic variation(s) responsible for the ID in your families have not been reported previously, then database trawling might not be the best way to proceed in your case.

You mention that you are investigating familial ID. Do you have DNA from any large families? If so, you could look for linkage between known or novel SNPs detected by your WES and the phenotype. Any genes in regions of SNP/phenotype co-segregation that are also expressed in lymphoid and/or myeloid lineages might be reasonable candidates. Otherwise, there are a number of programs (whose names escape me at the moment - alas!) written to detect associations between SNPs and phenotypes via transmission disequilibrium testing, affected sib-pair analysis etc.

If you like, I can dig out some info on where to find and how to use these programs.

HTH

Hi Glynis,

I really suggest you to use Exome variant database, EVS. (http://evs.gs.washington.edu/EVS/)

It includes exome data from more that 6000 individuals. Could be interesting to know if they identified your same genetic variants and the minor allele frequency in their population.

Hope this will help you!

Valeria

Hi Glynis,

We successfully identified a few new disease genes and I must say that ideally you want a more complete variant annotation than you described, which will integrate most of the annotations I mention below. Short of that, you can use some free online tools listed below. SIFT and Polyphen are not really useful when looking at exome data, and GO annotations can be useful but you want more thorough info on each gene.

RS numbers and 1000 genome frequencies are a good start, but I highly recommend using EVS variant frequency data, as Valeria said above. For recessive diseases, if a variant is not much rarer than a MAF of 0.01, it would be frequent enough to be a well known disease mutation. To get EVS frequency data, you can generate a VCF file for your BAM file using SNVerGUI, then annotate it using wANNOVAR. These are free tools and can be used on any laptop by anyone who doesn't know bioinformatics or code language.

You have to make sure you exclude known human diseases, that includes those where immunodeficiency might not be the primary manifestation. You can use Phenomizer for a thorough gene list. Also, plug your filtered list in the CCS Miami ROH tool to get links to all OMIM disorders. Then, you have to assume that for recessive diseases, you might be missing one of the two mutations (either not covered, or deep intronic) and thus consider carefully all recessive disease genes with a single hit. Once that is done, look at the mouse phenotype of your filtered list by using the MGI Batch query tool. Finally, to look at gene function, I like to use Genedistiller, which also gives some basic info on associated human and mouse phenotypes.

Once you have candidate variants, look at "control" BAMs from other diseases using IGV from the Broad Institute (with dbSNP track implemented) to make sure the variants are not technology-specific artefacts or misaligned indels. Then, do a lot of Sanger to see how the candidate variants track in your family.

Good luck and let me know if you need help using those.

Philippe

I am not sure if this helps, but this is a problem we encounter frequently with gene lists from GWAS and transcriptomics, and in essence the problem is the same for exome data. You have a long list of candidate genes, and you don't know where to start with best candidates for further empirical analysis...

If you have a list of genes that contain candidate variants, you could also use some of the computational disease gene prioritization tools that are out there - for example ENDEAVOUR at http://homes.esat.kuleuven.be/~bioiuser/endeavour/index.php. If you have a set of already known disease-causing genes, you can also look at the Gene Ontology functional annotations of known genes and see which of your candidate genes have similar functions (NIH DAVID website can help you with functional annotation enrichment using GO terms).

Pathway analysis can also be helpful, to see if any of your candidates from the exome analysis fall in pathways that are functionally appropriate or contain some of the already known disease genes. I prefer to use Ingenuity Pathway Analysis software (where you could also upload the actual variants rather than the gene lists) and see if any of the canonical pathways with relevant function are enriched with the variants from your analysis - but this is quite expensive software and it depends whether your institution has a license... but because the databases etc are highly curated, you can derive good biological context without necessarily needing bioinformatics skills so it is worth scouting around for access to IPA.

Also, do you know if the inheritance is autosomal dominant, recessive, sex-linked etc? Or are they de novo mutations? This can help you decide whether you can filter out heterozygous/homozygous variants.

I would also be cautious of using SIFT, PolyPHEN etc as an early filtering step, as our knowledge of variant function is very incomplete. I would save this as a later step, and use it for prioritising certain SNPs rather than actually excluding any SNPs or genes from the analysis.

Here are some references that might be of assistance in computational disease gene prioritisation:

Nat Rev Genet. 2012 Jul 3;13(8):523-36. doi: 10.1038/nrg3253.

Computational tools for prioritizing candidate genes: boosting disease gene discovery.

Methods Mol Biol. 2011;760:189-206. doi: 10.1007/978-1-61779-176-5_12.

Web tools for the prioritization of candidate disease genes.

Methods Mol Biol. 2011;760:175-87. doi: 10.1007/978-1-61779-176-5_11.

Conceptual thinking for in silico prioritization of candidate disease genes.

Also if you google "disease gene prioritisation" you may come up with some additional ideas... there is a small but interesting field of bioinformatics research focusing on exactly this problem!

Hope this helps with some additional ideas....

Best wishes for your analysis :)

Let us start by your model of inheritance. You are looking for monogenic disease-causing mutations, so the first question is whether it is dominant or recessive (X-linked narrows things down to one chromosome, so much less of a problem).

Your approach will be drastically different depending on model. In dominant, you are looking for heterozygous protein-altering (or copy-number, if you have enough depth to assess that) variants, of which each healthy individual carries hundreds. In return, dominant inheritance gives you large pedigrees that allow you to narrow down the variants co-segregating with the disease (i.e. good old linkage analysis). I would WES the proband and do high-resolution SNP profiling on the affected siblings, parents, grandparents, aunts, uncles and cousins to narrow the mutation down to the small part of the genome all of these individuals share on one chromosome (unaffecteds also help, if the pedigree suggests very high penetrance). Dominant inheritance also allows you to use a much more stringent filter cutoff for allele frequency in known variants.

In recessive disease, you are looking for genes with two such variants, so this alone narrows it down to a dozen or two variants, after the filtering for known diseases and variants that the previous discussants suggested. I strongly recommend against any filtering of protein-altering variants for computational prediction of harmlessness (SIFT, Polyphen, etc) and will automatically reject any paper __relying__ on it, submitted to my journal (J Med Genet, IF=5.7). These should be used only for prioritizing, not for filtering. Also, filtering for low-frequency in known variants requires some caution in recessive disease. A thousand-genome variant with carrier rate of 0.001 would cause disease in 1/4 million live births, rare enough to have escaped conventional "positional cloning." In recessive disease, the problem is that the only family history you get is affected siblings, usually one. This allows you to narrow it down to 1/4 of the variants, in a non-consanguineous offspring. This leaves you with a small handful of genes to test in an unrelated proband. In the absence of unrleated probands, there is no way to arrive at a genetic proof, even if you are lucky enough to get it down to one variant. Functional studies are great and now mandatory to get your WES paper published in a good journal but they are not a substitute for genetic proof, except in rare occasions (e.g. animal-model KO with a strikingly similar phenotype, or a gene in a known anzymatic pathway for a biochemical phenotype). In using unrelated probands be careful about genetic heterogeneity. If your disease has escaped "positional cloning" to date, chances are that it is genetically heterogeneous.

Finally, in the absence of robust inheritance pattern, think twice before assuming that your phenotype is monogenic. Two affected siblings do not prove monogenic inheritance except, perhaps, in consanguineous offspring.

Good luck!