Hi All
Normalization is very important for clustering gene expression data. I am working on some cancer-related datasets and I came across this paper which does normalization of both rows and columns, and as far as I know we cannot normalize rows and columns simultaneously and also most people prefer to either normalize the row or the column.
But this paper stresses that the data used should be row- and column normalized (so that both rows and columns sum to 0 and have S.D of 1). Normalizing the row first and then performing column normalization on the row normalized data, did not add the rows and columns to zero!
Any idea.. Thanks! :)
Usually the only normalization i know is "Within-Array" and "Between-Array". But i should read the paper to be more accurate.
Best regards.
Hi Alberto, the paper is titled Consensus Clustering: A Resampling-Based Method for Class Discovery and Visualization of Gene
Expression Microarray Data by Monti et al.
yes Aparma, but for a very fast analysis i have read that only the hierarchical clustering (HC) has the row and column normalization.
Yeah, since consensus clustering uses many different algorithms to identify the number of clusters and to perform HC, they have performed row and column normalization, so in order to compare with other algorithms for clustering, the data should also be normalized in the similar way so that the comparison is more diligent.
I might have understood it not the right way, since I am reading more on it now, I am trying to make sense of what they have done! :)
why not! if you have many clusters you could normalize in the row and in the column way in order to compare the results.
But in order to be more accurate you could send a mail directly to Monti. I think he can answer all your doubt.
Best regards.
I mailed Monti in the first place since as you said, I also felt he might be able to answer my doubts clearly and explain why he chose to do row and column normalization, but I have not got any reply from him as yet. :(
yeah thats precisely what I would liek to do, Ian et al have created a package in R base don the technique developed by Monti. I am going to use that package to perform consensus clustering, but I need to normalize the data before I start using the package and run the data in it. And it is quite important,a s you know, that normalization plays an important role in the kind of results we can get and by using the wrong method, the results might be skewed.
Hi guys, I have been looking at a similar problem (yesterday) and have not found a reference yet.
At first sight, it seems non-trivial to normalize both row and colum simoultaneously, and I actually don't think there exists a closed-form solution (but I could be wrong). I have then implemented a simple iterative procedure which seems to work well (for 1000-by-500 matrices and accuracy 1e-6, convergence requires less than 10 iterations). It is attached as a Matlab m file (use ">> Mn = normalizationRC(M)" to normalize M) --- hopefully there is no error, I have not tested it much yet...
Let me know if you find any useful reference.
Found a reference
http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2868388/
also available at http://statistics.stanford.edu/~ckirby/techreports/BIO/BIO%20246.pdf
- Badges
- Science topic
- Similar topics
- Medicine
- Oncology
- Cancer Research
- Cancer Biology