from 13 014 to 16 889 genes. GENCODE builds on CCDS to validate additional transcripts and extend into UTR regions, building on the ENCODE pilot project (9,21–23) and incorporating additional computational and experimental input and validation (24). One new computational approach, which is being built on within GENCODE, is to use alignments across the many mammalian genomes now available to evaluate the conservation of putative coding sequences (25). Several hundred transcript predictions generated by the Ensembl gene build pipeline which were found to have low scores in this analysis have been identified as spurious and are now filtered out. The Ensembl/HAVANA collaboration includes further efforts to improve geneset consistency, such as tighter links with UniProt (26) and input into the Genome Reference Consortium (http://www.sanger.ac.uk/sequencing/grc/) to flag discrepancies between the human genome sequence and transcript evidence.
