Science skeptics have recently reviewed these services, and found plenty of quackery. [Science Based Medicine] [Skeptical Raptor]
However, there is some research to back up the idea that genetic testing can provide insights into metabolic disorders. There may be as many as 200 SNPs for which there are proven metabolic effects, and only a subset of these alter nutrient requirements in a significant portion of the population [e.g., the rs1801133 MTHFR SNP and folate requirement in 15–30% of the population (Solis et al., 2008) and the rs12325817 PEMT SNP and choline requirement in 20–45% of the population (da Costa et al., 2006)].
What do genetic testing services measure?
The human genome consists of about 3 billion nuceleotide bases, each of which is either A,C, T, or G. A reference genome based on the similarities of all genotyped humans has been assembled, along with a corresponding reference database of all of the point "mutations" where individuals differ from that baseline. So far, scientists have documented about 150 million of these single nucleotide polymorphisms, called SNPs. The average person doesn't have all of these differences, however; most people have about 3 millions SNPs that differentiate them from the reference human genome. Since 3 million is about 0.1% of the 3 billion base pairs, most humans differ from each other by about 0.1% of our DNA. Ancestry genomic testing services like 23andme test for a few hundred thousand SNPs for $100-$200. For $1000-10,000 a complete genomic sequence can be obtained. (there are about 3 billion bases total.
Examples: MTHFR – metabolic pathways and nutrigenomics
In humans, SNPs in the gene MTHFD1 increase the demand for betaine as a methyl-donor, thereby increasing the dietary requirement for choline. Another SNP in the gene PEMT prevents the activation of this gene by estrogen, thereby decreasing endogenous production of phosphatidylcholine (a source of choline) in the liver and increasing the dietary requirement for choline. [Choline: Critical Role During Fetal Development and Dietary Requirements in Adults. Ziesel]
But note that it is not so simple. There are several forms of the MTHFD1 gene, for example MTHFD1L and MTHFD2. If MTHFD1 is commonly mutated, it may be a pseudogene.
Complexities interpreting SNPs
No simple test can unravel the intricacies of the human genome, and consumers should be suspicious of anyone claiming to be able to interpret measurements of tens of thousands of genes, with millions of genetic variations, some of which have effects on hundreds or thousands of the small molecules of metabolism (and perhaps on thousands of peptides or proteins involved in metabolism).
[A grand challenge for nutrigenomics. Steven Zeisel. 2010.]
Mistakes in genomics data
Note that 23andme data, like any large genome scan, can have mistakes in it. For example, the Enlis genomics blog found more than 500 likely mistakes in a sample of 23andme raw data! (Enlis)
Solution: Metabolic Testing
There is a genetic test for MTHFR variations. But there’s also a cheaper and more accurate way to test for whetherMTHFR variations are causing disease. We simply check the levels of homocysteine in the blood...In other words, the homocysteine levels determine our actions, not the MTHFR test results.[Cleveland Clinic]