Representing Short Read Clusters As Feature Vectors- Which Properties To Include?

Question

[Click downvote]

I want to train an SVM to recognize piRNA-clusters as opposed to clusters of miRNA, tRNA, mRNA.

I have found a few thousand clusters which are almost certainly piRNA clusters. I want to represent those clusters as feature vectors, to enable an SVM to find other likely candidates for piRNA clusters. However, the only features I can think of including in such a vector is: the length of the cluster (nb. of basepairs) and the number of reads in the clusters.

Are there other properties I should consider adding?

Answer 1

[Pavel Senin]

maybe binucleotide, trinucleotide frequencies? (k-mers frequencies) and their ordering?

[Click downvote]

Will have to look those up. Thanks!

Answer 2

[JC]

Time ago I wrote some complexity and composition function in Perl, check: https://github.com/caballero/SeqComplex