Isoform ratio investigation in RNA-Seq

[mknut]

Given:
2 samples with bio replicates, along with 1 control
FPKM values for each gene
FPKM values for each isoform of each gene

I am trying to find if proportional alteration of specific isoforms of one particular gene has an effect on proportional alteration of isoforms in other genes.

Example:
Gene A has two isoforms: a, b
In control the ratio of a/b is 5/1.
In sample 1 the ratio is 1/10, with b isoform dominating.
In sample 2 the ratio is 50/1, with a isoform dominating.
I want to find if altered a/b ratio of that one particular gene has an effect on proportional isoform distribution in other genes. Example:
Gene B, with isoforms x, y has x/y ratio of 10/1 in controls. In sample 1, y isoform takes proportionally much more, with x/y being 1/10, in line with the trend for gene A in the same sample. In sample 2, x/y follows the trend for Gene A again, with the ratio being 100/1.

I am trying to figure out a way to look for genes like gene B from the example. I am not interested in changes in gene expression levels, just isoform ratios.

In order to do this, I am considering using k-means clustering of ratios. I would have to calculate a ratio of two major isoforms for each gene in each sample, and simply run k-means on the sets. This would be very limiting, as I would be able to use only 2 isoforms of every gene I am trying to cluster, and would restrict myself to using two major isoforms and their ratios.

Do you have any suggestions? Any software with in-built functions for something like this? Or maybe I am missing something obvious?

I know that CummeRbund has similar functionality, but it is for gene level, not isoform level.

[santhilalsubhash]

Given:
2 samples with bio replicates, along with 1 control
FPKM values for each gene
FPKM values for each isoform of each gene

I am trying to find if proportional alteration of specific isoforms of one particular gene has an effect on proportional alteration of isoforms in other genes.

Example:
Gene A has two isoforms: a, b
In control the ratio of a/b is 5/1.
In sample 1 the ratio is 1/10, with b isoform dominating.
In sample 2 the ratio is 50/1, with a isoform dominating.
I want to find if altered a/b ratio of that one particular gene has an effect on proportional isoform distribution in other genes. Example:
Gene B, with isoforms x, y has x/y ratio of 10/1 in controls. In sample 1, y isoform takes proportionally much more, with x/y being 1/10, in line with the trend for gene A in the same sample. In sample 2, x/y follows the trend for Gene A again, with the ratio being 100/1.

I am trying to figure out a way to look for genes like gene B from the example. I am not interested in changes in gene expression levels, just isoform ratios.

In order to do this, I am considering using k-means clustering of ratios. I would have to calculate a ratio of two major isoforms for each gene in each sample, and simply run k-means on the sets. This would be very limiting, as I would be able to use only 2 isoforms of every gene I am trying to cluster, and would restrict myself to using two major isoforms and their ratios.

Do you have any suggestions? Any software with in-built functions for something like this? Or maybe I am missing something obvious?

I know that CummeRbund has similar functionality, but it is for gene level, not isoform level.

Hello mknut, I have not used this tool but this should solve your problem.

http://genes.mit.edu/burgelab/miso/

http://www.ncbi.nlm.nih.gov/pubmed/?term=21057496

[sphil]

Given:


In order to do this, I am considering using k-means clustering of ratios. I would have to calculate a ratio of two major isoforms for each gene in each sample, and simply run k-means on the sets. This would be very limiting, as I would be able to use only 2 isoforms of every gene I am trying to cluster, and would restrict myself to using two major isoforms and their ratios.

Why would you only be able to use 2 isoforms? You might declare one as the 'standard isoform' which is ok doing by random since you will always use this isoform. Calculate every ratio of interest against that reference 'standard isoform'. Now you can go on with your k-means clustering. However, you have to keep in mind that k-means has a fixed amount of clusters at some point. Therefore you can only cluster genes with the same amount of isoforms present at the same run.
Hope that helps..