The rise in sequence data has brought both challenges to the way we exchange biological information and opportunities to discover new protein families, primarily through the investigation of uncultured metagenomic samples.The Distributed Annotation System, or DAS, provided a means for exchanging protein sequence data, but there were no open source, stand-alone DAS clients optimized for integrating and viewing these data. To address this need, we developed DASher. Complementary to visualizing DAS data with DASher, we also created and made available ten servers to offer real-time protein feature predictions via DAS. While DAS works well for genomic data, there was no such framework for exchanging orthology data in a consistent way. Consequently, we developed the first standards for orthology data exchange, SeqXML and OrthoXML. 64 reference proteomes are now available in SeqXML, and 14 orthology providers have agreed to offer their predictions in OrthoXML. Besides creating a uniform representation of common data types, these standards enable direct comparison and assessment of competing methods for the first time.A substantial percentage of newly sequenced genes are ORFans, which have no match to previously known sequences. Metagenomics samples uncover sequences from uncultivable and therefore previously unseen species, and ORFans constitute much of the metagenomics data that are completely uncharacterized. ORFans are by definition impervious to standard similarity-based methods, and the few existing metagenomics gene-finding methods performed poorly on short, error-prone next-generation sequence data. Therefore, we designed a new approach to predict protein-coding gene families from metagenomic data and applied it to 17 virally-enriched metagenomes derived from human patients. Of the 456 putative ORFan families we found in the nearly 1 billion nucleotides sequenced from these libraries, we identified 32 putative novel protein families with strong support.