A machine learning pipeline to improve De Bruijn graph metatranscriptomic assemblies

Motivation: With the growing significance of metatranscriptomic assemblies, the need to improve their quality and maintain their controllable size has become essential. That would help in boosting all applications based on metatranscriptomic assembly. In this paper, we propose a pipeline that filters de novo assemblies while preserving or improving their quality. Original assemblies are based on De Bruijn graphs and were created by Oases. Auxiliary scripts that help reporting statistics about all kinds of metatranscriptomic assemblies are integrated with the pipeline as well. Results: Experimental results show that the pipeline helped improving the accuracy of the assemblies with up to 6+% in addition to filtering 5000+ transcripts from 6 original assemblies each made up of 21000+ transcripts. The high precision of filtered assemblies and the reasonable running time of the pipeline makes it a potential postprocessing step of different de novo assemblies. Availability: All pipeline scripts are publicly available at https://sourceforge.net/projects/metatranspipeline/files/ Contact: [email protected]