This repo contains the code for a nucleotide sequence clustering algorithm based on Hidden Markov Models. This approach effectively addresses the challenge of non-homologous segments within the alignment.
The easiest way to use this program is via the packaged jar file available for download from the release page to the right here.
Then assuming Java is installed you run the program by providing a single argument which is a multi-sequences alignment (MSA) file in fasta format e.g. java -jar hmmcluster.jar test/msa.fa
By default, the program considers the alignment verbatim, but with the -trim option the underlying statistical approach can be used to find the optimal start and end base positions of the cluster-able region. e.g. java -jar hmmcluster.jar test/msa.fa -trim
For full details see our paper
From a given sequence alignment, hmmcluster determines the optimal number of sequence clusters that best explain the diversity across the genomes. This approach groups together sequences based on the statistical likelihood that they come from the same underlying source.
The program outputs two types of data. The first prints how well n HMM clusters represents the present variability, where n is 1 to the number of sequences in the input. This is represented by a table containing log likeihood, number of parameters and AIC. The last printed line of this output is the optimal number of clusters. The second output below this table is the assignment of sequences to clusters for the optimal model as just described.
Richard Goldstein UCL & &