Genesis of the 1500 IQ Humanoids

[J11]

Anyone want a genetic IQ test?

It would only account for a small percent of the variance, perhaps 2%, though it would get you ready for the time in the not too distant future when such a SNP based IQ test would be much more informative.

I have 120 perfect proxies for the 246 SNP from the recent IQ GWAS that are on my 23andme gene chip. I also wrote a program that could be downloaded for free that automatically extracts the information from a 23andme SNP file.

I will post this after I double check everything.

[J11]

The first column has the SNP as reported in the recent
GWAS along with a zscore, the allele frequency, the genotype with
the effect. The next column has a SNP that is a perfect proxy
for the SNP in column 1 in my 23andme file (v3) along with
its chromosome number and position.

The Zscore is a scaling of standard deviation with
roughly 300 Zscores = 1 SD

So, 300 Zscores would equal about 15 IQ points.
The perfect score here would be all positive alleles and no negative alleles
for a score of 800 Zscores = 2.5 SD.

However, the expected Zscore would be about 0.05 SD.
The distribution of IQ scores for these SNPs could be calculated.

Anyone who might be interested in double checking, please
report back. This is only an initial list. There were other SNPs that
were excluded though they had high reliability scores and there were
also alternative SNPs that could be tried.

rs7546297 -6.958 0.378426 a rs2271933 1 32092525
rs12035012 -8.262 0.217016 a rs12032756 1 41761429
rs501299 -5.858 0.640745 a rs501299 1 44051834
rs2420551 -6.549 0.887065 a rs1022013 1 69432044
rs3101338 8.902 0.802695 a rs3101338 1 72750353
rs34701878 7.423 0.521774 a rs6678734 1 96176563
rs11804556 6.317 0.069156 a rs11804556 1 98498441
rs1473474 5.762 0.472328 t rs1473474 1 98588885
rs6681533 6.626 0.410132 t rs1415345 1 103615938
rs72694234 5.998 0.69284 t rs4612664 1 153822673
rs34320898 5.982 0.183646 c rs17650310 1 171757204
rs6681390 -5.851 0.522221 a rs6670617 1 174246712
rs12470949 -5.829 0.284247 t rs12470949 2 23934816
rs967569 -6.051 0.674007 t rs1508120 2 41643650
rs2955280 -5.569 0.527964 t rs2955280 2 44116836
rs17049085 5.542 0.880182 t rs17049085 2 57948003
rs10189857 6.937 0.562889 a rs10189857 2 60713235
rs56202165 5.46 0.104681 a rs1011407 2 60665768
rs58593843 -5.915 0.096982 a rs907574 2 60503230
rs11898362 -6.023 0.303372 a rs10191517 2 73558403
rs2309812 7.93 0.358192 t rs13010010 2 100852734
rs4850921 5.594 0.360594 t rs11123814 2 100767082
rs6718450 5.704 0.381414 c rs1437721 2 117692893
rs10189912 -6.778 0.646474 a rs13405986 2 144161734
rs3106666 -6.006 0.411241 a rs3106666 2 155547853
rs6436555 6.826 0.508023 a rs6436555 2 157488277
rs10192369 -5.759 0.512673 a rs10192369 2 161380888
rs17221725 5.714 0.086464 t rs12052592 2 161724866
rs3749034 -6.01 0.217632 a rs3749034 2 171673475
rs1401112 5.798 0.809778 t rs1401108 2 180991780
rs62198803 5.629 0.236434 a rs17765227 2 186209495
rs7573001 -5.872 0.379905 c rs11679484 2 198921604
rs17355973 -5.502 0.253573 t rs17277107 2 213869354
rs908639 5.546 0.206221 t rs1400724 2 226326642
rs1523048 6.367 0.365815 t rs1523048 3 35527409
rs13096357 5.669 0.114444 a rs2276850 3 48669648
rs1540293 6.321 0.097351 t rs1540293 3 50463935
rs4485754 5.713 0.788497 a rs4485754 3 54234722
rs11720523 6.876 0.409655 a rs11720523 3 71545170
rs6770622 -6.853 0.041746 a rs6774581 3 85194054
rs12646225 6.086 0.119864 t rs12646225 4 696848
rs2295499 -6.254 0.434878 t rs2295499 4 2717690
rs4484297 5.64 0.25077 c rs975777 4 16318066
rs13107325 -9.736 0.065445 t rs13107325 4 103188709
rs2726491 -9.779 0.350416 a rs2726491 4 106212562
rs6840804 -5.953 0.695996 a rs4835349 4 147766118
rs75973558 5.71 0.884555 a rs1479679 5 26882085
rs414976 -5.81 0.41563 a rs414976 5 60965517
rs6882046 -7.402 0.715183 a rs6882046 5 87968864
rs166820 6.738 0.173576 a rs157566 5 89378726
rs4308464 -6.724 0.375177 c rs4342312 5 92477890
rs354670 -5.566 0.644734 t rs354670 5 101110298
rs7731260 5.78 0.48494 a rs7731260 5 107781712
rs1438660 5.684 0.656837 a rs3797696 5 109062626
rs1145123 7.434 0.516307 t rs1145123 5 111026598
rs1991228 -5.852 0.296751 c rs686349 5 113957246
rs753280 6.315 0.569849 t rs753280 5 140024042
rs566237 -6.449 0.683261 a rs538827 6 11509323
rs6456379 5.742 0.424638 a rs6456379 6 20893247
rs6903716 5.898 0.700015 a rs6903716 6 21956404
rs35433030 5.485 0.082738 a rs13195509 6 26463660
rs1280049 5.486 0.475485 a rs1280049 6 76536333
rs1906252 11.639 0.465953 a rs1906252 6 98550289
rs3823036 -6.548 0.681029 t rs3823036 6 99284532
rs9384679 -9.345 0.398784 t rs6911407 6 108867031
rs799444 6.6 0.450816 t rs4720475 7 44769869
rs12698810 -6.239 0.114105 t rs12698810 7 69300968
rs13223152 6.497 0.591423 a rs13232100 7 69945340
rs4731365 -7.043 0.393086 a rs4731365 7 127082497
rs13253386 -7.28 0.54033 t rs13253386 8 14002020
rs1473634 -5.984 0.697706 a rs1473634 8 20915316
rs13259607 5.628 0.110502 t rs13259607 8 71352047
rs990747 7.541 0.765091 t rs990747 8 93173000
rs12554512 -10.159 0.594364 t rs11794152 9 23345347
rs7871404 -5.822 0.81235 a rs7871404 9 99262296
rs1003346 5.481 0.419988 a rs1003346 9 111815340
rs913264 6.354 0.286634 t rs913264 9 131944138
rs7069887 5.977 0.852171 a rs7069887 10 29569272
rs2393967 -6.09 0.691484 a rs2393967 10 65133156
rs1408579 6.23 0.445704 t rs2862954 10 101912064
rs7921305 5.825 0.23956 a rs7085239 10 133750141
rs10896632 -5.473 0.265553 a rs10896632 11 57390302
rs7941785 5.583 0.36911 a rs7941785 11 63861317
rs2885208 5.522 0.807669 t rs10891503 11 112964139
rs3758927 -5.562 0.803927 c rs3802924 11 133827733
rs4937860 5.511 0.897351 a rs4937860 11 133814713
rs1054442 -7.541 0.619957 a rs1054442 12 49389320
rs11171739 -5.516 0.573529 t rs11171739 12 56470625
rs6539284 -6.947 0.612997 t rs6539284 12 79592680
rs1727307 5.819 0.288605 a rs10848428 12 123572495
rs3843954 -6.658 0.274161 c rs9537888 13 58551353
rs2478286 -8.338 0.745704 c rs2478288 13 106637428
rs4983183 6.293 0.319865 a rs4983183 14 27170907
rs7149600 6.709 0.842532 a rs7149600 14 29566302
rs971681 -5.861 0.383539 t rs2181179 14 30073056
rs2239647 -7.75 0.447783 a rs2239647 14 33292743
rs11622558 -6.223 0.388482 t rs11628131 14 36994609
rs1007934 5.715 0.379905 a rs1007934 14 73463479
rs10145335 -6.013 0.243825 a rs965770 14 98583045
rs2071407 -7.832 0.362458 t rs2071407 14 103987140
rs11634187 5.58 0.850447 t rs2289328 15 40705417
rs7184911 6.502 0.558192 a rs7184911 16 7670326
rs72774059 6.012 0.099384 a rs4780784 16 10176088
rs276626 5.594 0.812411 a rs276626 16 13156649
rs4781499 5.842 0.847659 a rs4781499 16 13630944
rs8054299 -8.004 0.682476 c rs17800727 16 53481010
rs12446238 5.628 0.458516 a rs12446238 16 62075138
rs8051038 6.33 0.749415 a rs8044342 16 71869532
rs6505191 -5.611 0.530213 a rs3829594 17 28842538
rs17698176 -5.772 0.802202 t rs17698176 17 44819595
rs11079849 5.747 0.313397 t rs11079849 17 47090785
rs12450712 -6.237 0.734955 t rs7210376 17 50566353
rs17002025 5.78 0.12239 a rs17002025 19 12530177
rs2072490 6.397 0.509332 t rs8108738 19 18255359
rs384114 6.773 0.600539 t rs407470 19 31937986
rs1862499 -5.512 0.398845 a rs1862499 19 47549791
rs78084033 -5.7 0.86563 a rs6058296 20 34269225
rs6019535 8.289 0.305297 a rs6095360 20 47532536
rs2154556 -5.545 0.704466 t rs2154556 21 39198887


The positive Zscore from these 120 SNPs is 400.032.
The negative Zscore from these 120 SNPs is -361.775.

The total positive Zscore for the full 246 is 801.783.
The total negative Zscore for the full 246 is -749.601.

[J11]

This is encouraging!

I just went to the dbsnp site and tried a batch submit of rs numbers.
dbsnp provides the minor allele frequency and the reference and alternative alleles.
Usually the alternative allele is the minor allele, though sometimes when it is close to
0.50 thing flip around. If I can trust the other alleles are the alternatives, then my whole
procedure that generated all these SNPs could be made into a solid pipeline without any
manual effort of looking for SNPs etc. involved.

[J11]

Great news!

The Fasta file from dbsnp has the needed information.
It would have been a hassle to check to determine which of the alleles was minor or major.

The Fasta format will allow me to write a program to automatically assess that.

[J11]

We have now upped the ante!!!
Stunning!

This research is called Compressed Sensing; and when they say Compressed : they mean it.

With a large set of linear equations with upwards of 99% of the columns with a value of 0, it should not
be overly surprising that the solution requires much less data input then would be otherwise expected.

Bottom line: For many traits/diseases (IQ, AD, ALS, ...), ALL SNPs should be detectable at a
GWAS sample of 1 million or so. GWAS samples are already moving to such a size and validation
of this research hypothesis is now happening in other traits.

This is profoundly important.
Any trait could use this basic idea (for example, AD).

All that is necessary is to move below the phase boundary and then increase sample size until median p-values descrease
(black line on page 20 Figure 1B url below) To move below the black line all you need to do is move down the vertical axis;
rho= s/n where s is the number of SNPs involved in the trait (for AD: perhaps 8,000-12,000) and
n of IGAP was about 80,000. As sample size continues to increase below the black line, median p-values will then make a
transition and then all the SNPs can be found Basically, increase the sample size and travel down and to the right, which brings you below the black line continuing to increase the sample size will lead to a complete list of SNPs.

Notice that changing heritability from 1.0 in Figure 1A (left) to 0.5 in Figure 1B (right) does not change the
black line. The phase boundary does not change, though the accuracy of the beta effects is much lower
on the right. What this would mean is you would not have a good estimate of the effect sizes of the SNPs
in the red zone. This should not be of overriding importance. It is true that some very rare SNPs might not
show up, though the increase in knowledge of AD would still be very large. With imputing, it would not be
unexpected that even rare SNPs would be detectable.

This would be awesome!

You have a complete list of 10,000 AD SNPs, so what?

Finding a rare AD variant for anyone with a family history of dementia would then only require
a simple search of the variants against a whole genome sequence or imputed gene chip results.

Funders have been cautious about stepping up with larger and larger AD GWAS because they likely
thought it would then be necessary to continue to endlessly increasing sample sizes to discover
the information needed (i.e., find out the actual effect sizes of each SNP etc.). Yet, apparently it should
be much easier and cheaper to find all the AD SNPs then had been imagined.

Lobbying to move under the transition boundaries would be in the self interests of the AD and other
advocacy communities. That there is a compelling cost-benefit rationale would make such lobbying all
the easier.

https://arxiv.org/pdf/1310.2264.pdf

[J11]

Wow!!!

They have unlocked a polygenic trait: height, using Compressed Sensing.
This has never been done before.

Article below mentions that other traits such as AD could now follow.
Considering that AD has higher heritability perhaps ~0.80 versus height ~0.5 there
should be a firmer transition boundary and, thus, it should be easier to solve AD genetics.

Advocating for a mega AD GWAS to fully resolve dementia genetics now seems very sensible.
As the article notes: "The potential public health benefits are potentially enormous."

http://www.biorxiv.org/content/biorxiv/ ... 4.full.pdf

[Stavia]

what is GWAS? I'm totally lost J11. I don't understand any of this.

Sent from my SM-G930F using Tapatalk

[J11]

Stavia, this is very startling!

GWAS-- Genome Wide Association Study

They gene chip people (e.g., 23andme) and then they ask about presence or absence of traits/illnesses such as AD.
With 1 million comparisons, you need a genome wide p-value of .05/1,000,000 = 5*10-8.
For about ten years they had no luck replicating results.

However, we have now crossed a critical point in GWAS.
We have now crossed over the phase boundary of GWAS research.
Things should really start to happen now.

In the height GWAS, they took 500,000 at the UK Biobank and they gene chipped and phenotyped them.
GWAS have been going on for almost 20 years now and not much has happened: when you
have 1 million markers and 5 thousand people in the study it was hopeless.

Look at the figure below.
The panel on the left shows a simulation in which heritability is 1. There is no noise.
Not many traits have a heritability near 1, perhaps autism goes up to 0.8+.
The coloring is the normal error and the black line is the phase boundary.
When you move below the boundary you rapidly pick up true SNPs.
At the white circle, 100% of the SNPs have been found.

Phase 1.PNG

The panel on the right has a heritability of 0.5.
It is modeling height, though there are many others with h2 in this range, for example intelligence.
The black line is still the phase boundary, though when you push through the boundary not much happens
The left panel had a hard phase shift, the right panel has a soft phase shift.
You need to move down to the white circle on the right to capture all the SNPs and even
then there will be a fair amount of error. Yet, this is a remarkable achievement.
You have captured all of the SNPs.

The rule of thumb is all the SNPs can be captured with a sample size of 30s -- that is 30 times the number of SNPs at h2=0.5.
They really nailed this one it looks like about 20,000 SNPs * 30 = 600,000 sample size.
UK Biobank is 500,000 + and they added in other samples.

The y-axis in the figures is rho = s/n, where s is the number of non-zero SNPs and n is the sample size
The x-axis is delta= n/p, where n is as above and p is the number of genotypes on the gene chip.

Basically the game plan is that you want to move downwards on the figure to hit into the blue area.
If your trait/illness is sparse, not many non-zero effect sizes for the SNPs, then you are at an advantage.
You are also at an advantage if your heritability looks more like the left than the right.
AD should be somewhere in between.

The great part of all this is we now have a real world proof that the theory works.
All that is needed is to move past the phase boundary hit into the blue and the entire AD genome will unlock.
This might give us 10,000-20,000 SNPs.

Up till this point the researchers have been totally unaware of where they were in the rho-delta plane.
When you look at the figures, you start to have the quivers.
If there were 10,000 AD SNPs that means s=10,000 and you have an AD IGAP GWAS size of 100,000, then n=100,000
so rho = s/n= 0.1 means you are right about the height of the white and red dots of the left panel.
And you have n=100,000 and the p is 1-5 million; n/p= ~ 0.1.
This places you right near the white dot on the left panel.

This is very exciting!
It boils down to what is s and what does the phase diagram actually look like.
This can all be found empirically.

If they found that the median p-value of non-zero betas was downshifting, then this could be quite invigorating.
The payoff would be overwhelmingly in favor of anteing up.

AD heritability is not 1, it might be 0.8, might be as low as 0.5. If we are sitting right above
the blue area then we could simply push down into it and the entire AD genome would unlock. The theory of the article cited above used minimum of 30s which for AD might be 300,000, though that was assuming h2=0.5; so AD should be less.

The phase transition of AD (add in a triangle of deep blue on the right panel that goes roughly up to 0.2 on the y-axis on the right hand side) should be intermediate between the left and right panels. Knowing precisely the transition boundary would be
very helpful. If we could run an AD GWAS that pushed us below the black line and into the deep blue, then we would have a hard boundary and all the AD SNPs would quickly pop out of the regression and we would have very good estimates of the effect sizes of the SNPs.

Apparently the GWAS research community doesn't speak phase boundary language, though given the current results it is hard to
see how they can avoid doing so.

The GWAS up to this point has been mostly noise.
Yet with L1 regression, they have now been able to essentially find all the SNPs.

Pretty much anyone and everyone will now be marching in the streets to have a GWAS fill in the final pieces.
The AD community probably should get on their marching shoes too.
Having the entire set of genes and SNPs that combine to cause AD would be invaluable.
The current research has found that basically a complete list can be found.

AD should have a good transition boundary. This transition boundary research was reported about
10 years ago and it relates to questions such as how you can transmit a picture over the internet
using the fewest bits: compression. Since genes chips have up to a million data points and many
traits such as AD only involve roughly 10,000, the matrix is so called sparse and this leads to the interesting
result that at a certain size of GWAS, the entire list of trait SNPs will just pop out of the regression.

What has now happened with the height study is they have finally crossed the transition boundary.
They have essentially maxed out the heritability from common SNPs; correlation of 0.64.
This is what the nature nurture debate has been about for over a century.

The current research using only gene chip results can predict a typical person's height within a few centimeters.

The scary part of all this is that we are within sight of predicting a lot about people from a simple DNA sample.
For example, 23andme could soon be able to make reasonably precise estimations about people's IQ, income,
AD risk, behavioral profile, height, weight, BMI, everything.

[Stavia]

J11 you lost me at paragraph 2. Can you use small words? I dont understand phase boundary or phase shift or anything past paragraph 1.
If it's too hard to explain dont worry. But seriously I don't understand any of this...

Sent from my SM-G930F using Tapatalk

[circular]

J11 can you find some online videos that teach all this in lay people's terms?