XspecT utilizes the uniqueness of kmers and compares extracted kmers from the input-data to a reference database. Bloom Filter ensure a fast lookup in this process. For a final prediction the results are classified using a Support Vector Machine.

Local extensions of the reference database are supported.

The tool is available as a web-based application and a smaller command line interface.

To install Xspect, please download the lastest 64 bit Python version and install the package using pip:

pip install xspect

If you would like to train filters yourself, you need to install Jellyfish, which is used to count distinct k-meres in the assemblies. It can be installed using bioconda:

conda install -c bioconda jellyfish

On Apple Silicon, it is possible that this command installs an incorrect Jellyfish package. Please refer to the official Jellyfish project for installation guidance.

To download basic pre-trained filters, you can use the built-in command:

xspect download-filters

Additional species filters can be trained using:

xspect train you-ncbi-genus-name

Run the following command lines in a console, a browser window will open automatically after the application is fully loaded.

xspect web

Run xspect with the configuration you want to run it with as arguments.

xspect classify your-genus path/to/your/input-set

For further instructions on how to use the command line interface, execute:

xspect --help

XspecT is able to use either raw sequence-reads (FASTQ-format .fq/.fastq) or already assembled genomes (FASTA-format .fasta/.fna). Using sequence-reads saves up the assembly process but high-quality reads with a low error-rate are needed (e.g. Illumina-reads).

The amount of reads that will be used has to be set by the user when using sequence-reads. The minimum amount is 5000 reads for species classification and 500 reads for sub-type classification. The maximum number of reads is limited by the browser and is usually around ~8 million reads. Using more reads will lead to a increased runtime (xsec./1mio reads).