0
$\begingroup$

I have genes from four species of bacteria. I constructed an undirected graph from the sequence similarity between all the genes from these bacteria. Each node in the graph is a gene and each edge is the similarity between two genes. I partitioned this graph into subgraph clusters of similar genes using the MCL software.

The four species of bacteria I am using are split into two groups (A & B) based on their response to an environmental condition. I would like to identify the clusters of genes from my graph that are producing this response. To put it another way I would like to create a measure to find the clusters which discriminate A from B. Generally each cluster has genes from all four species.

What would be the best way to do this?

One idea I had was to determine the mean weight between all A-type nodes in a cluster and then divide by the mean edge weight of the cluster. Values greater than 1 would indicate the A-type genes are more closely related to each other.

$\endgroup$
  • $\begingroup$ "I would like to create a measure to find the clusters which discriminate A from B." Obviously and plainly, it is the ratio between numbers of As and Bs in a cluster. If As or Bs prevail in a cluster then that cluster discriminates them well. Your idea of ratio of mean similarity between As in a cluster to mean similarity of all points in the cluster is questionable. What if you have just 2 As (very close to each other) and 20 Bs in a cluster. The ratio will be >1. What does it tell you? $\endgroup$ – ttnphns Oct 4 '11 at 3:55
  • $\begingroup$ The four bacterial genomes are closely related and the clusters I am examining exhibit no large differences in number of nodes from either group A or B. Therefore as my next step, given that there is an observable experimental between A and B, I assume there may be still be a genetic basis and that this may be exhibited in the relationships between A/B nodes in the clusters. $\endgroup$ – Michael Barton Oct 4 '11 at 14:07

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.