/sci/ - Science & Math » Thread #13054193

179KiB, 879x1200, EWXBZONY.jpg

View Same Google iqdb SauceNAO

Principal component in genetics

Anonymous Fri 30 Apr 07:39:32 2021 No.13054193 View Reply Original Report

Quoted By: >>13054220 >>13054224 >>13054388

Can someone good at genetics and statistics explain to me in simple words, what is principal component here and why it's reliable parameter to clusterize races?

> https://www.researchgate.net/publication/45274375_Toward_a_more_Uniform_Sampling_of_Human_Genetic_Diversity_A_Survey_of_Worldwide_Populations_by_High-density_Genotyping

Anonymous

Anonymous Fri 30 Apr 2021 07:50:48 No.13054220 Report

Quoted By: >>13054244

>>13054193
pca is a dimensionality reduction technique that orientates data according the variability so the axis with the greatest variation is the first pc and the rest are orthogonal to it. it's not the same as clustering, that is separate but you can use it in clustering. clustering should be in all available dimensions and is done with algorithms like k-means and structure. then you use pca to plot the labelled clusters in 2 or 3 d.

Anonymous

Anonymous Fri 30 Apr 2021 07:52:06 No.13054224 Report

Quoted By: >>13054244

>>13054193
Principle component analysis is just a statistical way of creating clusters. So the genetic information is multidimensional, there are a lot of features per sample. The first principle component finds the linear combination of these samples that has the highest variatiance or spread. The second PC has the second highest variance. This just creates a new kower dimensional dataset that is ideal for clustering.

Anonymous

Anonymous Fri 30 Apr 2021 07:58:10 No.13054244 Report

Quoted By: >>13054256 >>13054277

>>13054220
>>13054224
Great explanation, thanks! What features are used in this study to form PCs?

Anonymous

Anonymous Fri 30 Apr 2021 08:04:01 No.13054256 Report

Quoted By: >>13054277

>>13054244
it depends, you can represent each dimension in many different ways with genetic data. say you have a dna strand. you can have 1, 2 or 4 dimensions for each position or it can be based on alleles. there is so many ways to go about this. the pca and clustering you just apply that data to, just make sure it's numeric.

Anonymous

Anonymous Fri 30 Apr 2021 08:11:50 No.13054277 Report

Quoted By: >>13054295 >>13054297

>>13054244
Im not a geneticist, but they mention SNP data in the abstract. Wiki yells me these are variations of single nucleotides at a certain location. So I reckon the data looks like a big array. The rows differentiate different samples, while the columns signify these SNP positions, the value represents the nucleotide at that position.
>>13054256 this anon is correct in that the data should be numeric so I guess they represent the nucleotide by a number. Another method ciuld be one-hot encoding, but that would quadruple the amount of columns used and I dont think PCA would work so well in that case.

Anonymous

Anonymous Fri 30 Apr 2021 08:20:57 No.13054295 Report

Quoted By: >>13054302

>>13054277
To come back to this, I am really not sure what form the data would have. I think it would more logical if the columna repreaent the SNP location with a certain nucleotide and the value represent how many times they have counted this in the sample corresponding to the row

Anonymous

Anonymous Fri 30 Apr 2021 08:21:39 No.13054297 Report

Quoted By:

>>13054277
yeah i was suggesting one-hot for the 4 dimensional per position. it's actually quite common to remove the bias from something like 1, 2, 3, 4 for the different bases.

Anonymous

Anonymous Fri 30 Apr 2021 08:25:10 No.13054302 Report

Quoted By:

>>13054295
you could do something like for each position you have 1 dimension and put a value mapped from a=1, c=2, g=3, and t=4. so you have as many dimensions as there are bases in the person's genome. each data point is one genome.

Anonymous

Anonymous Fri 30 Apr 2021 09:11:49 No.13054388 Report

Quoted By:

>>13054193
So pajeets and pakis end up in the same cluster as turks. That's kind of funny

Capcode	All Only User Posts Only Moderator Posts Only Admin Posts Only Developer Posts
Show Posts	All Only With Images Only Without Images
Deleted Posts	All Only Deleted Posts Only Non-Deleted Posts
Ghost Posts	All Only Ghost Posts Only Non-Ghost Posts
Post Type	All Only Sticky Threads Only Opening Posts Only Reply Posts
Results	All Grouped By Threads
Order	Latest Posts First Oldest Posts First

Your latest searches