As mentioned previously, we briefly held off from releasing the set of “high confidence” miRNAs for miRBase 21, because of a last-gasp bug. Those data are now available, tagged with the label “high confidence” on the entry pages, and for download on the FTP site. The total number of miRNAs labelled “high confidence” has increased by 168, to 1996. That increase is partly due to our incorporation of more deep sequencing datasets, and also because we’ve relaxed one criterion:
In miRBase 20, a high confidence sequence must have at least 10 reads that map to each of the two mature sequences (-5p and -3p). In miRBase 21, high confidence sequences must either (a) have at 10 reads mapping to each arm, as before, *or* (b) have at least 5 reads mapping to each arm *and* at least 100 reads mapping in total. The latter case helps us to catch some of the well-established, highly expressed miRNAs that have very high arm expression bias — that is, a large number of reads mapping to one arm, and a small number to the other.
A few sequences labelled as high confidence in miRBase 20 have disappeared in the miRBase 21 set, some because of the aforementioned bug.
Facilities remain in place for you to vote for whether or not you agree with our high confidence assertions — see individual entry pages, and sequencing read views.