How long will it take for authors of the 5P study to offer a prenatal gene test for autism, a common genetic variant found in 60% of the entire population. After all, the lead author of the 5P study has claimed that if we could wipe out the 5P mutation, 15% of all autism would just disappear.

Did you lose track of the blog post that started this discussion? That is essentially Kev’s thesis. He even got the quote correct (25% in the quote)

THe Nature paper did not have realistic male/female ratios. The authors published the supplementary data used in the study:

Which only matters if there is a difference between the males and females in the control group.

If the goal was to understand the paper either you or passionlessdrone would have contacted the author and posted his answer here.

I’ll save you the trouble (from the contact author):

...this locus we reported on 5p14.1 contributes to both male and female autism and it makes no difference if we use males or females as controls (we loose some power if we drop out 50% of the females but the locus remains significant; the replication is also performed in trios where this has not impact but the bottom line is that MAF is the same in both sexes of healthy controls

They could remove many of the females from their control group. It would change the statistical power but not the conclusion.

I’m lost on what you are getting to here. Why would 1/3 of autistics “have” to have a6? Or are you saying it has been already found?

It was a hypothetical model, if that wasn’t clear. Briefly, if there are many complex genotypes that express as autism, it’s not practical to find single alleles that associate statistically with autism.

Basically, difficulties in finding straightforward genotype-phenotype associations are not evidence of lack of heritability. They evidence lack of problem solving technology. There are other ways to see if a phenotype is heritable.

My previous post read:
The study controlled only for gender (white people of European ethnicity).

Should have read:
The study controlled only for ethnicity (white people of European ethnicity).

The design criteria for AGRE admittance excluded Fragile X patients from inclusion. By definition, the AGRE data contains a subset of mentally retardated subjects with autistic-type features. The AGRE data set does not contain IQ scores thus cannot differentiate a candidate autism gene from a mental retardation candidate gene.

Remember last years discovery of a rare genetic variation ‘16P’ that the authors claimed represent at least 1% of all autism cases:

http://content.nejm.org/cgi/co.....JMoa075974

The authors made the same over exaggerated claims about their discovery that the authors of the 5P14 common genetic variant discovery.

Several studies published since the discovery have not replicated the Boston Children’s Hospital groups discovery:

http://www.ncbi.nlm.nih.gov/pubmed/19306953?

Revisiting the original study, the authors claim that the 16P discovery contained in th AGRE data was replicated in in a second group selected from the Boston Children’s Hospital.

The Boston Hospital Children’s Group did not replicate the AGRE findings since the Boston Children’s Hospital Group contained chidren selected in the author’s own words as:

“we observed the identical deletion in 5 of 512 children referred to Children’s Hospital Boston for developmental delay, mental retardation, or suspected autism spectrum disorder”

Still, the Boston Children’s Hospital is now selling a ‘genetic test for autism’ based on the hyper exaggerated discovery that 16P genetic variations account for ‘at least 1% of all autism cases’.

How long will it take for authors of the 5P study to offer a prenatal gene test for autism, a common genetic variant found in 60% of the entire population. After all, the lead author of the 5P study has claimed that if we could wipe out the 5P mutation, 15% of all autism would just disappear.

What is going on here?

THe Nature paper did not have realistic male/female ratios. The authors published the supplementary data used in the study:

http://www.nature.com/nature/j.....999-s1.pdf

The AGRE data used in the study:
Males = 83%
Females = 17%

The control group used in the study:
Males = 52%
Females= 48%

The study controlled only for gender (white people of European ethnicity). While they recognize that ethnicity can produce different rates of common genetic variants (ie sickle cell disease), they fail to consider the fact that gender (Type 2 Diabetes) may also produce different rates of common genetic variants.

The authors did not even segregate their findings by gender.

A few sceptics in the genetic science community consider genome wide association studies to be ‘fruitless’:

http://www.nytimes.com/2008/09.....wanted=all

Another way of looking at this would be, if autism affected one hundred males for every female, do you think it would a good idea to have a control group at fifty/fifty? Or what about a control group of ninety nine/one? The fact that our known ratio, four to one is much smaller serves only to make your range of making an error proportionally smaller.

You are making a basic error. If there is a genetic difference between males and females in locations of interest for either the study or control groups, there is no proper gender ratio. In those cases, the proper comparison would be to break out males and females separately. I.e. compare male controls to male subjects and, likewise, females to females.

If there is a genetic difference in the regions of interest for the study group only, that is a result and should be reported. I.e. if there were, say, a higher rate of the genetic difference amongst males than females in the study group, that would be a big deal and would be worth its own discussion.

I don’t have access to the paper right now—if/when I do I’ll check if they discuss that.

you keep acting as though I don’t understand your point. I do.

The point I made is very clear. You keep avoiding it.

I’m lost on what you are getting to here. Why would 1/3 of autistics “have” to have a6? Or are you saying it has been already found? [sorry]

Imagine what might happen if there are hundreds or thousands of ways to be autistic. There’s no chance a statistical scan will find much of note. It’s a computer science problem, rather than a statistics problem.

I suspect you may be right in that there are a great number available genotypes and consequent problems with identifying them.

But this is what led me to my original question on this thread, we seem to have already identified some genotypes that raise autism risk far more significantly than what was reported in this widely heralded study. In these cases, the authors seemingly took a backwards approach from these studies; targeted gene scans based on clinical findings, i.e., we know this and this are measured abnormally in autism, what happens when we scan for mutations in genes known to have impacts on this area. And indeed, it seems that combinations of alleles may confer much more risk than a single allele.

For example, Aberrations in folate metabolic pathway and altered susceptibility to autism was published a month ago.

RESULTS: MTHFR 677T-allele frequency was found to be higher in autistic children compared with nonautistic children (16.3 vs. 6.5%) with 2.79-fold increased risk for autism [95% confidence interval (CI): 1.58-4.93]. The frequencies of MTRR 66A allele (12.7 vs. 21.0%) and SHMT 1420T allele (27.9 vs. 45.3%) were lower in autistic group compared with nonautistic group with odds ratios 0.55 (95% CI: 0.35-0.86) and 0.44 (95% CI: 0.31-0.62), respectively, indicating reduced risk. MTHFR 1298C-allele frequency was similar in both the groups (53.3 vs. 53.6%) and hence individually not associated with any risk. However, this allele was found to act additively in the presence of MTHFR 677T allele as evidenced by 8.11-fold (95% CI: 2.84-22.92) risk associated with MTHFR 677CT+TT/1298AC+CC genotypes cumulatively.

The greatest relative risk observed in the Nature paper, if I remember correctly, was more like 1.18. Here we have a different group that reported a nearly eight times greater risk of autism based on targeted evaluations. Funny enough, it was only having a1 and a2 that increased risk; a1 wasn’t differentially transmitted. [Haven’t read the entire paper, so I can’t tell yet if they had realistic male/female ratios.]

What is going on here?

– pD

If there is no difference in the results of the tests on males and females, there is no systematic error introduced.

One of the most basic facets of having appropriate control groups is that they match, as closely as possible, the group you are trying to study. The goal is to minimize differences between the groups, otherwise, how can you tell if differences are the result of something you are trying to study, or something else?

But the problem is, what if being male is part of the problem at a genetic level? How can you tell if differences, or no differences, in the results of the test are random, or real? And lets say there was a difference found? In order to determine if it was real or not, do you recommend another massive geocoding, with a more representative control group to account for this? If that study comes back with different results, does that study get the press this one did?

Another way of looking at this would be, if autism affected one hundred males for every female, do you think it would a good idea to have a control group at fifty/fifty? Or what about a control group of ninety nine/one? The fact that our known ratio, four to one is much smaller serves only to make your range of making an error proportionally smaller.

– pD