“…This scaling, however, assumes that control and experimental datasets have the same complexity, which is not always true, especially in experimental systems with significant bottlenecks. As noted above, several methods have been used to computationally address this problem 24-26, 31 by using the difference in library complexity between TIS datasets to estimate the severity of technical and experimental bottlenecks; the more complex library is then computationally passed through a bottleneck of similar size to model which insertion mutants are likely to be lost entirely by chance during the experiment. Loss resulting from bottlenecks can be simulated by proportionally removing reads equal to the overall difference in diversity between datasets 31 , proportionally removing reads based on the difference in unique reads in a set of known neutral genes 25, 26, 30 or using the difference in overall complexity to simulate reads using a multinomial distribution 24 .…”