In DNA sequencing, many genes are actually predicted by algorithms but never functionally confirmed. Results from two studies "suggest there's a considerable rate of error creeping into experiments that rely on the gene sequences that are predicted by current algorithms," says a post at Ars Technica. In one published in August in BMC Bioinformatics scientists used a software called MisPred, which scanned protein domains to find unlikely matchups and therefore questionable genes. In another study in PLoS, researchers isolated all the proteins from Arabidopsis cells and mapped the fragments back to the genome. They found that more than 18,000 were from sequences outside known genes, and they identified nearly 800 new genes and corrected for errors in another 700 predicted genes.