The study of whole microbial communities through RNA-seq, or metatranscriptomics, offers a unique view of the relative levels of activity for different genes across a large number of species simultaneously. To make sense of these sequencing data, it is necessary to be able to assign both taxonomic and functional identities to each sequenced read. High-quality identifications are important not only for community profiling, but to also ensure that functional assignments of sequence reads are correctly attributed to their source taxa. Such assignments allow biochemical pathways to be appropriately allocated to discrete species, enabling the capture of cross-species interactions. Typically read annotation is performed by a single alignment-based search tool such as BLAST. However, due to the vast extent of bacterial diversity, these approaches tend to be highly error prone, particularly for taxonomic assignments. Here we introduce a novel program for generating taxonomic assignments, called Gist, which integrates information from a number of machine learning methods and the Burrows-Wheeler Aligner. Uniquely Gist establishes the most appropriate weightings of methods for individual genomes, facilitating high classification accuracy on next-generation sequencing reads. We validate our approach using a synthetic metatranscriptome generator based on Flux Simulator, termed Genepuddle. Further, unlike previous taxonomic classifiers, we demonstrate the capacity of composition-based techniques to accurately inform on taxonomic origin without resorting to longer scanning windows that mimic alignment-based methods. Gist is made freely available under the terms of the GNU General Public License at compsysbio.org/gist.