如何在R中检测基因组数据中的dist和kNN？

Question

I have genomic data with missing values and I want to calculate the distance between the expression levels of each pair of genes by using the available values. Then i want to discover the K nearest neighbors to fill the gaps? How I can do that in R?

gene sample 1   sample 2    sample 3    sample 4
1      5555        NA          2151       5484    
2      5564        NA            NA        NA
3      4544       4656         14546       45455   
4      NA         54654           NA        NA

... How I can calculate the eucledian distance? I need to use a just one row at the time?

Sorry I´m new with genomic data and I can´t find this information anywhere.

Thanks.

Answer 1

I guess what you are trying to do is knn-imputation for the missing values, not knn-classification. There is a ready made function for this called impute.knn from the impute package on the bioconductor . Read the helpfile closely before use.

source("http://bioconductor.org/biocLite.R")
biocLite("impute")
require(impute)

x <- rnorm(1000, 50, 5)  # 1000 random samples
x[sample(1:1000, 50)] <- NA  # 50 are randomly made NA
x <- matrix(x, nrow = 10)  # make a matrix
impute.knn(x)

Answer 2

Googling for R k nearest neighbor leads me to the knn function in the class package. In regard to your second question, calculating the euclidian distance is simply:

sqrt((sample1_x - sample1_y)^2 + ... + (sample4_x - sample4_y)^2)

where x and y are the indices of the rows you want to calculate the distance between. However, you have a lot of NA's in your data, I'm not sure how you need to deal with that as the euclidean distance is undefined when there are NA's involved.