Bases having low relative coverage are of particular interest, provided that the low coverage is not an accident of sample size. For example, at 20-fold mean coverage, some bases whose 'true' relative coverage is 1 (corresponding to an expectation of 20 overlapping reads), will occasionally have measured relative coverage of 0.5 (corresponding to an observation of 10 overlapping reads), as that measurement is only off by (20-10)/20≈2.2 standard deviations (based on a Poisson model). Thus, deep sequencing is required to accurately identify bases having low relative coverage.
