.. _home-page-mopconsensus: ******************* MOP_CONSENSUS ******************* .. autosummary:: :toctree: generated This pipeline takes as input the output from MOP_MOD with all the four worklows. It outputs the consensus of the diferent predictions running the tool `Nanoconsensus `__ in parallel on each transcript for each comparison. Input Parameters ====================== The input parameters are stored in yaml files like the one represented here: .. literalinclude:: ../mop_consensus/params.yaml :language: yaml How to run the pipeline ============================= Before launching the pipeline,user should: 1. Decide which containers to use - either docker or singularity **[-with-docker / -with-singularity]**. 2. Fill in both **params.config** and **tools_opt.tsv** files. To launch the pipeline, please use the following command: .. code-block:: console nextflow run mop_consensus.nf -params-file params.yaml -with-singularity > log.txt You can run the pipeline in the background adding the nextflow parameter **-bg**: .. code-block:: console nextflow run mop_consensus.nf -params-file params.yaml -with-singularity -bg > log.txt You can change the parameters either by changing **params.config** file or by feeding the parameters via command line: .. code-block:: console nextflow run mop_consensus.nf -params-file params.yaml -with-singularity -bg --output test2 > log.txt You can specify a different working directory with temporary files: .. code-block:: console nextflow run mop_consensus.nf -params-file params.yaml -with-singularity -bg -w /path/working_directory > log.txt Results ==================== Here an example of a result: .. image:: ../img/nanocons.png :width: 800