What it does

BlockClust is an efficient approach to detect transcripts with similar processing patterns. We propose a novel way to encode expression profiles in compact discrete structures, which can then be processed using fast graph-kernel techniques. BlockClust allows both clustering and classification of small non-coding RNAs.

BlockClust runs in three operating modes:

1. Pre-processing - converts given mapped reads (BAM) into BED file of tags
2. Clustering and classification - of given input blockgroups (output of blockbuster tool) as explained in the original paper.
3. Post-processing - plots for overview of predicted clusters.

For a thorough analysis of your data, we suggest you to use complete blockclust workflow, which contains all three modes of operation.

Inputs

BlockClust input files are dependent on the mode of operation:

1. Pre-processing mode:

   • Binary Sequence Alignment Map (BAM) file

2. Clustering and classification:

   • A blockgroups file generated by blockbuster tool
   • Select reference genome

3. Post-processing:

   • Output of cmsearch, searched clusters generated by BlockClust against Rfam
   • BED file containing clusters generated by BlockClust
   • Pairwise similarities of blockgroups generated by BlockClust

Outputs

1. Pre-processing mode:

   • BED file of tags with expressions

2. Clustering and classification:

   • Hierarchical clustering plot of all input blockgroups by their similarity
   • Pairwise similarities of all input blockgroups
   • BED file containing predicted clusters
   • BED file containing prediction of blockgroups by pre-compiled SVM binary classification model.

3. Post-processing:

   • Plot of distribution of ncRNA families per predicted cluster (overview of cluster precissions). The annotation of ncRNA families are retrieved by searching cluster instances against Rfam database.
   • Hierarchical clustering made out of centroids of each BlockClust predicted cluster