Saturday, May 16, 2009

Why is it so difficult to find the "autism gene"?

There was media coverage recently about the discovery of the "first common" set of autism gene variants. The alleles are apparently found in 65% of autistic people. What's interesting is that they also occur in about 50% of non-autistic people. (Some sources say 60%.)

That doesn't sound like a finding that, by itself, could be practically applied to the genetic screening of autistic people. I'm not too worried about that. Additionally, autism-related studies of this sort don't replicate a lot of times.

It seems to be a very difficult problem. Why is that? I'm sure different people have different views about this.

What I was wondering is whether experts dealing with a similarly elusive problem could provide some insights of note. The problem I'm referring to is that of finding the "race gene." The following are the recommendations regarding race and genetics by the National Human Genome Center of Howard University.

1. When the human species is viewed as a whole, underlying genetic variation and expressed physical traits exhibit gradients of differentiation, not discrete units. Therefore, modern extant humans do not fracture into races (subspecies) based on the modern phylogenetic criteria of molecular systematics.
2. The biological “boundaries” between any human divisions (groups, populations, nationalities) are circumstantial and largely dependent on what traits are chosen for emphasis.
3. The demographic units of human societies (and of the U.S. census) are the products of social or political rules, not the forces of biological evolution. The names and characteristics of demographic groups can change and have changed over time.
4. Group differences in health parameters are not encoded in the human genome as part of an evolutionary pattern of divergence. Thus, differences in health or disease cannot be treated as causally related to ethnoancestral groups.
5. Genotype-environment interactions are more important in explaining group differences in health than genotype, environment, or a factor called “race”.
6. The non-existence of human races (subspecies) does not mean the non-existence of racism. Racism is the structured systematic oppression against individuals and groups defined based on physical traits that reflect an extremely limited fraction of the human genome. Racism must be addressed.
7. Individuals cannot be treated as representative for all those who physically resemble them, or have some of the same ethnohistorical ancestry. Ancestries of individuals and groups should be ascertained in order to evaluate differential expression of genetic effects.


(source)

I thought that sounded quite pertinent.

I'm not saying that it won't ever be possible to fairly accurately distinguish an autistic person from a non-autistic one by simply looking at a genome sequence. As someone with a Computer Science background, I can theoretically speculate that someone will figure out a method eventually. I just don't think it will get done by simply looking for alleles that represent statistically significant "risk" factors.

See also: Race and Genetics at Wikipedia.

Friday, May 08, 2009

Another Way to Tell the "Epidemic" in California is Bogus

There's been some renewed California "epidemic" talk recently because of a report released by California DDS (this one) despite the usual DDS disclaimer to the effect that report numbers don't represent epidemiological counts of all autistic persons in the state. Kristina discussed it and there's been some media coverage as well, with the usual Rick Rollens scare-mongering about an upcoming surge of autistic adults that the state of California allegedly doesn't have any experience dealing with. There are a number of inaccuracies and faulty assumptions in the media coverage of the report, to be sure, but I wanted to focus on one particular claim found here:

The percentage of people with both autism and mental retardation has dropped significantly, a trend that may provide clues for those trying to solve the autism puzzle.


That's not stated accurately, but in fact there has been a gradual drop in the prevalence of mental retardation within the population of recognized autistics, at least for the last 17 years. See page 20 of the report.

This is a well known fact, or at least I'd like to think it is. I've discussed it several times previously (here, for example.) It is mentioned in Gernsbacher et al. (2005) as well.

I'm sure broadening ascertainment denialists can find ways to rationalize this finding. But what if I told you that the phenomenon is not only a time-based phenomenon? It can also be observed when you compare one regional center to another. That is, regional centers with a higher administrative prevalence of autism will tend to have a lower proportion of autistics who also have a classification of mental retardation.

You know a chart is coming, but as usual I'd like to be clear as to where the data comes from, so anyone reading can double-check if they so wish. I'm using a file provided by California DDS upon request, named CDERQtrData.zip. It only goes up to January, 2006, but it contains more autism-specific information for each regional center than you normally find in the regular report. I don't have population data for each regional center (and gathering that would be a bit much for a blog post) but I will use the Autism-Epilepsy ratio as a proxy of administrative prevalence. The administrative prevalence of epilepsy in California is roughly stable (just below 0.1%) so dividing the autism caseload by the epilepsy caseload of a regional center should provide us with an adequate proxy of the administrative prevalence of autism.



The figure demonstrates an inverse association between the Autism-Epilepsy ratio and the proportion (%) of autistics who also have an MR (or unkown MR) classification. The downward trend is statistically significant with 99.6% confidence.

Now, you'll note the distribution of the dots in the chart is fairly random. I wondered why that might be, especially why the Central Valley regional center would have a low administrative prevalence of autism and a low proportion of autistics with MR.

The Central Valley RC has 9,284 clients with mental retardation as of January, 2006. That's comparable to other big RCs. Of all individuals with MR, only 247 (2.7%) also have an autism classification. This is ridiculously low. I'm sure that if someone went to the Central Valley RC and screened the individuals with an MR classification, they would find that a lot more than 2.7% of them are also autistic.

It's generally understood that there's been increasing recognition of autism in the population without mental retardation. This is what they call HFA. What's not so intuitive is that there's also been increasing recognition of autism in the population with mental retardation. Clearly, regional centers have different levels of recognition across both populations. In average, 7.7% of persons with mental retardation will have a classification of autism in California (which is still rather low.)

Let's look at what the result would be if we were to adjust the autism numbers under the assumption that every regional center should have a level of recognition of 7.7%.



That's a lot more clear, isn't it? This suggests that differences in the recognition of autism in the population with MR across regional centers are an artifact.

Thought Experiment

Imagine there's a hypothetical regional center where recognition of autism has gotten out of hand, to the point where every person who resides in the area served by the RC is classified as autistic by it. In this case the Autism-Epilepsy ratio would be about 1,000.

You'll note the first chart above has a power regression model (the formula on the upper right.) The type of model is theoretically justifiable, and I can discuss that on another occasion.

y = 32.235 x-0.4525


In the model, x is the Autism-Epilepsy ratio and y is the proportion of autistics with MR.

So what if x is 1,000? This model predicts that the proportion of autistics who have MR would then be 1.41%. This is basically what you'd expect for the general population. In other words, DDS data is entirely consistent with a cultural explanation of the differences in administrative prevalence of autism between regions.