The quality clipping function clips the ends of readings when the average (over 31 bases) confidence value is lower than a user defined threshold. As with the difference clipping system the clips are only adjusted when the newly calculated clip points are more stringent than the originals.
After clipping Gap4 then identifies any holes (breaks in the contigs) that may have been created and fills them up again by extending the sequence(s) with the fewest number of expected errors.
An example output follows.
Hole from 32652 to 32725: extend #1378 and #1385 with 3.157324 expected errors
It has been our observations that when using confidence values propotional to log(error_rate) (such as those output by Phred which are -10*log(err_rate)) that it is sometimes better to not perform any quality clipping, especially when using a consensus algorithm which can make use of the confidence values. Note though that the difference clipping method does improve the consensus sequence. See section Difference Clipping.