More than 2,500 coding genes in the human reference gene set still have unsettled status.

In 2018 we analysed the three main repositories for the human proteome, Ensembl/GENCODE, RefSeq and UniProtKB. They disagreed on the coding status of one of every eight annotated coding genes. The analysis inspired bilateral collaborations between annotation groups. Here we have repeated our analysis with updated versions of the three reference coding gene sets. Superficially, little appears to have changed. Although there are slightly fewer genes predicted as coding overall, the three groups still disagree on the status of 2,606 annotated genes. However, a comparison without read-through genes and immunoglobulin fragments shows that the three reference sets have merged or reclassified more than 700 genes since the last analysis and that just 0.6% of Ensembl/GENCODE coding genes are not also annotated by the other two reference sets. We used eight features indicative of non-coding genes to examine the 21,873 coding genes annotated across the three reference sets. We found that more than 2,000 had one or more potential non-coding features. While some of these genes will be protein coding, we believe that most are likely to be non-coding genes or pseudogenes. Our results suggest that annotators still vastly overestimate the number of true coding genes.