<i>Ab initio</i> gene prediction reduces data volume at a small cost to accuracy.

<p>Reads of various length (70–500 bp) were simulated with 1% error rate from mock community 160319967-stool1. Reads were translated naïvely in 6 frames (6FT) or were translated using a metagenomic gene prediction tool (Prodigal, FragGeneScan, or MetaGeneMark). Predicted ORFs were searched and classified into SFams using RAPsearch2. All other parameters were tuned to minimize relative abundance error. (A) Amino acid sequence length (relative to naïve 6FT) resulting from read translation. (B) Minimum relative abundance estimation error for mock community 160319967-stool1 corresponding to different translation methods.