Posts Tagged ‘Statistics’

Dr. Tal Yarkoni: “Functional MRI in Health Psychology and beyond: A call for caution

In practice, the modal fMRI sample size of 15 – 20 subjects often provides little power to detect anything but very large effects (Yarkoni, 2009). For example, a one-sample t test performed on 20 subjects at a statistical threshold of p < .001 (the modal threshold in the fMRI literature) has only 40% power to detect even a canonically ‘large’ effect of d = 0.8. For a correlational analysis, the same sample size provides only 12% power to detect an extremely large correlation of r = 0.5.

How many LOW and UNDER-powered imaging genetic studies (where genetic variation is merely one of many variables correlated with individual differences in task or baseline BOLD responses) have I have covered in this blog?  Eeeek!  I’d better not go there … I’ll go to ENIGMA instead.

[pic cred]

Read Full Post »

Statistical pickles

Just a pointer to Dr. Ben Goldacre’s wonderful book and recent blog post on some widespread statistical flubs.  Readers of genetics and neuroscience media should be alert to conclusions of this ilk: “people with genotype AA respond differently to treatment (pre- vs. post-) than people with genotype aa (pre- vs. post-).”

You can say that there is a statistically significant effect for your chemical reducing the firing rate in the mutant cells. And you can say there is no such statistically significant effect in the normal cells. But you cannot say that mutant cells and mormal cells respond to the chemical differently.

Read Full Post »

I mean, how many people are really needed to run a sufficiently powered genome-wide association study?  Are there enough people on the planet?  Heather J. Cordell’s review, Detecting gene-gene interactions that underlie human diseases, seems optimistic, but, at this point, it seems a valid question … at least if you want to detect gene-gene interactions.

“The historical lack of success in genetic studies of complex disease can largely be attributed, not to ignored biological interactions, but rather to under-powered studies that surveyed only a fraction of genetic variation …”

thanks for the pic heckyeahart

Enhanced by Zemanta

Read Full Post »

An historic find has occurred in the quest (gold-rush, if you will) to link genome variation with brain structure-function variation.  This is the publication of the very first genome-wide (GWAS) analysis of individual voxels (voxels are akin to pixels in a photograph, but are rather 3D cubes of brain-image-space about 1mm on each side) of brain structure – Voxelwise genome-wide association study (vGWAS) [doi: 10.1016/j.neuroimage.2010.02.032] by Jason Stein and colleagues under the leadership of Paul M. Thompson, a  leader in the area of neuroimaging and genetics – well-known for his work on brain structure in twin and psychiatric patient populations.

In an effort to discover genes that contribute to individual differences in brain structure, the authors took on the task of statistically analyzing the some 31,622 voxels (per brain) obtained from high-resolution structural brain scans; with 448,293 Illumina SNP genotypes (per person) with minor allele frequencies greater than 0.1 (common variants); in 740 unrelated healthy caucasian adults.  When performed on a voxel-by-voxel basis, this amounts to some 14 billion statistical tests.

Yikes!  A statistical nightmare with plenty of room for false positive results, not to mention the recent disillusionment with the common-variant GWAS approach?  Certainly.  The authors describe these pitfalls and other scenarios wherein false data is likely to arise and most of the paper addresses the pros and cons of different statistical analysis strategies – some which are prohibitive in their computational demands.  Undaunted, the authors describe several approaches for establishing appropriate thresholds and then utilize a ‘winner take all’ analysis strategy wherein a single ‘most-associated winning snp’ is identified for each voxel, which when clustered together in hot spots (at P = 2 x 10e-10), can point to specific brain areas of interest.

Using this analytical approach, the authors report that 8,212 snps were identified as ‘winning, most-associated’ snps across the 31,622 voxels.  They note that there was not as much symmetry with respect to winning snps in the left hemispere and corresponding areas in the right hemisphere, as one might have expected.  The 2 most significant snps across the entire brain and genome were rs2132683 and rs713155 which were associated with white matter near the left posterior lateral ventricle.  Other notable findings were rs2429582 in the synaptic (and possible autism risk factor) CADPS2 gene which was associated with temporal lobe structure and rs9990343 which sits in an intergenic region but is associated with frontal lobe structure.  These and several other notable snps are reported and brain maps are provided that show where in the brain each snp is associated.

As in most genome-wide studies, one can imagine that the authors were initially bewildered by their unexpected findings.  None of the ‘usual suspects’ such as neurotransmitter receptors, transcription factors, etc. etc. that dominate the psychiatric genetics literature.  Bewildered, perhaps, but maybe thats part of the fun and excitement of discovery!  Very exciting stuff to come I’ll bet as this new era unfolds!

Reblog this post [with Zemanta]

Read Full Post »