Domain-Based Clustering Of Protein Structures

Question

[Pals]

I am working with lysosomal proteins. I have made working set of structures which I want to further analyze. Clustering them into fewer groups would make the analysis more useful because when I grouped them based on SCOP folds or CATH architectures, most of the groups contain only one structure. Therefore, I was willing to try a program that would employ some clustering algorithm for my data set. I have already used DALI but it did not seem to be useful. Can anyone tell me if there are such programs (other than DALI)? Any suggestions?

Thanks!!

Answer 1

[Thaman]

For statistical analysis what is your biological question/hypothesis/quest regarding the available structures? Do you want to figure out which set of structures fall under defined group (SCOP fold or CATH architecture)? Cluster is a technique to find groups of similar items within a data set. If you already know that abc proteins is in SCOP group and xyz in CATH and if no differentiation in structure then I dont think there is need of clustering.

You can give try with Chipster 2.4 Microarray analysis tool. Initial step is normalization either with Chipster or you can import normalized data. After the normalization you can generate phenodata and define group (1,2). Then you can use either Clustering tools or Visualization-> dendogram for clustering your dataset.

[Pals]

Yes, as there are many structures which lack CATH/SCOP classification, I want to include them into one of the available groups based on some criteria or alternative ways of classification. Of course I have compared the unassigned sets with available folds and group them manually. I guess Chipster does not work for structures. I want programs that consider protein domains.

[Thaman]

In particular chipster doesn't work for structures that's right. But, my point was regarding clustering and chipster does that in one click if dataset from any source has been already preprocessed. Right now I am using Chipster for Proteomics dataset even though its entirely meant for Microarray and NGS.

Answer 2

[Alastair Kerr]

As Thaman mentioned, it depends on your question. If your structures are independent of the CATH/SCOP groups they still be be linked by functional regions such as those found in Interpro. Factor based clustering algorithms would then work with these if there are too many to manually sort.

[Pals]

Of course I have classified the structures based on CATH/SCOP but as you can see there might be some structures which can have very small difference and still put into different groups. In addition nearly 30% of the structures are not classified with these (CATH/SCOP) programs not only because they are novel structures but also might be because they partially represent two or more or none of the available folds. In other words, my wish is to make a DALI search among the set I have (90 structures).