Looking to get PATTERN CLUSTERING of gene expression data with multiple time points

Question

[Hushus]

Hello all,

This is the 3rd thread concerning this issue and we have made steady progress. I sincerely appreciate the inputs of cpad0112 and h.mon for their efforts

Have: * RNA expression data for 12 time points, 1 replicate * List of genes of interest and their expression data http://m.uploadedit.com/bbtc/1512924884525.txt [Sample data, does not contain all 100 genes]

Want: Expression pattern clustering like this where x axis is time and y axis is relative expression (or something like it)

Have tried: * Timecourse (R) which apparently does not accept if there is only 1 replicate [C: Cannot get result from TIMECOURSE (R PROGRAM)]

Question: Do you know of ANOTHER program that can achieve a similar result?

Possible solutions: 1) Take polynomial trend line of each line (gene) of expression data, cluster them accordingly manually. Problem: do not know how to get polynomial equation of each line of data using R.

Answer 1

[dariober]

A simple way to go about it may be k-means clustering, like

dat<- read.table('http://m.uploadedit.com/bbtc/1512924884525.txt')
kmeans(dat, centers= 3)

To get a reasonable guess for the number of clusters you could look at the between_SS / total_SS statistic, start with a small number and increase until the between_SS / total_SS stops increasing sharply.

Another useful method may be non-negative matrix factorization as implemented in e.g. the NMF package. The vignettes of NMF are quite useful. An example could be:

library(NMF)
dat<- read.table('http://m.uploadedit.com/bbtc/1512924884525.txt')
xnmf<- nmf(dat, rank= 4) # Again, you need to "guess" the number of "pseudogenes"
xnmf@fit@H

The H matrix essentially gives you a small set of "pesudogenes" that together describe well the (much larger) full set of genes.

[Hushus]

Oooo thank you for your detailed input. I will definitely try this out.

How would you plot the results?

Answer 2

[Hushus]

ANSWER: DOWNLOAD MeV SOFTWARE USE K-MEANS CLUSTERING, PEARSON.

YOU GET THE FOLLOWING:

GOOD ENOUGH FOR ME.