Hello Kejin,

this is a very difficult question. I have the same problem by planing my projects with genetic depending heart diseases. The first point should by thinking about such points:

1. Easy to sequence?

2. What is known about the mechanism of the protein? - many is known = maybe not nice to search for more

3. Did you have patients with problems in this proteins.

The basis for your research should be available samples. In other words, the measured biochemical markers, the level of protein is determined by pathology and so on. Based on this, try to select a group of genes, presumably involved in one of the mechanisms of pathology. For example, for heart disease - inflammation, lipid metabolism. Any study of this kind should not be just the definition of association, but have a certain logical biological basis.

Thanks Sevn and Konstantin. This is a difficult problem. Like the Sevn mentioned, we can use the sequence methods to scan the sample and find out the variants of patients. But, you will face a more terrible condition too many variants to find out, which is/ are true candidate genes? Although, given you find out one potential candidate gene within your samples, you just add another candidate gene up to current more than 100 candidate genes group. But how about for this disease's research? whether we really find out its pathogenic genes? Maybe! So, This is what I think about. when we are facing so many candidate genes of one disease, what can I do? How to verify them? to confirm which are true respond for disease, or others are not. At this point, Konstantin' idea may more reasonable and practical.

you can do the genetic polymorphism studies of very significantly associated genes of your interest disease.

You can perform whole genome sequencing against your samples using service at Complete Genomics. Then, you can filtering variants considering function by Ingenuity Variant Analysis (Variant filtering and genome comparison software). You can report selected variants within 100 candidate genes.

Hi Kejin,

I believe it is important to look at the genes that have the highest impact on the disease in question. You can also search online (for example on genecards.org) which genes have polymorphisms that are quite common in your population.

Thanks for all. The fact makes me come to believe that, almost no disease is single-genetic response for, let alone some human complex disease. So I'm still implicated some follow problems: 1) are there any genes causal for this disease? a single gene, several or too many genes? 2) how to use this existing condition (too many candidate genes for one certain disease), and the nest research work should be?

I performed a little experiment, to test more than 30 reported candidate genes among more 300 patients, almost half genes were found out in less than 30% patients. I still do not know how to arrange and elucidate this results. So, I am hesitated to design the protocol with my samples, just because I am not sure how to deal with the same condition. After all, my final work may be also to report a reported/ new candidate gene.

The problem is at least in part how YOU define and determine your candidate genes. From a linkage study, from genome wide association studies or from prior biological/biochemical knowledge. For example, any gene from a particular pathway where you know there is between-individual differences in the pathway's activity associated with the disease, will be a 'better' candidate than a gene associated with the general area of metabolism and linked to a QTL associated with the disease by GWA. Similarly, quantitative theory suggests that for any characteristic of an individual, variants will be segregating in a variable population at frequencies inversely related to their effects on the trait in question (assuming a population that is relatively well adapted to its environment). So for any potential candidate gene there are also potentially several variants segregating in different individuals within and across populations, but these variants are usually going to be mainly of small impact on the character being studied. Only rarely will you get lucky and hit a variant that has a significant effect of 1 or more sd of the environmental/background variation in the trait in the particular population of patients you are looking at. These rare cases are what we tend to mean when we talk of Mendelian, or single gene, traits. So in order to accurately assess the impact of a single particular variant at any particular candidate gene you need a fairly large cohort of patients and controls drawn from the same (sub)populations (in the order of 1-10k for relatively common QTLs). For characters like blood pressure there may be many hundreds of loci that influence the phenotype across all populations, but in any particular population maybe only 20 will have variants with effects large enough to detect with 300 patients. SO you need to ask yourself whether you really want to try to detect as many loci that affect the trait as possible, or whether you really just want to 'pick-off' the 1, 2 or a few most significant effects within your patient group. The design of your experiment then becomes one of considering the laws of diminishing return on research investment made. Where you draw the line though remains a judgment call based on available resources, time and personal inclination.

I have more or less the same problem. Obviously I cannot investigate everything at the same time but I have chosen a biochemical pathway and investigate the genes associated with it starting from one central.

Most of the gene associated studies report conflicting data due to small sample size, heterogenety of the cohort, ethnic and racial variations. It is difficult to predict your results but meta analyses help a lot.

Robert give us a permanent rule when to make research design. I feel that, if you choose the complex disease, and want to address its casual genetic bases, it means you choose a very difficult way. The diversity of disease's clinical diagnosis and its related critical are headache for us already. In addition, you have to face lots of candidate genes and still try to scanning the primary one for this disease. Also, in order to have a higher statistic power, you have to recruit a huge scale cohort.

In fact, we have adapted the biochemical pathway strategy several years ago. Several methods, including the case-control studies and the family-base associate test and linkage methods, were performed in our previous work. Unfortunately, it seem that, you always can capture some positive association between disease and genes given with a powerful sample size. It is no surprise, after all, they did have inextricably relationship. However, we also can not find out a more reasonable method to address this confusion. But, I think it maybe the time to find out other ways to research this complex relationship.

Hi Kejin,

I agree that many false-positive associations can be made in a sample size that has statistical power for making correlations. Mutations can be prioritized by type of mutation seen (Is the mutation deleterious? Is the gene haploinsufficient? Can protein prediction tools help?), family-associations, linkage tests, etc. Sometimes, looking at the associations of the published gene functions, results in GenomeRNAi, and/or experiments mouse knockout phenotypes with known disease phenotypes can help to narrow down the possibilities (Are there changes in any metabolic readouts in blood tests that could point to candidate genes?). The relationship of the gene to known genetic diseases with similar phenotypes in ClinVar may also provide clues. This often requires a "6th sense" from an experienced geneticist who can interpret this type of data (eg. reviewers of exome sequencing data in sequencing centers/companies).  Once the candidate gene list is finalized, the mutations still need downstream experimental validation. In a complex multi-driver disease such as cancer, multiple gene aberrations are tested concurrently for their interactions by mating various genetic mouse models together. Triple mutant mice are almost a "norm". I also expect that the new CRISPR/Cas9 system will enable us to validate multiple combinations of point mutations in these types of models.