Figure 6
Maximizing guides with positive signals and recommended feature ranges (Watson–Crick). Decision tree for maximizing the number of guides with positive signals. Percentage of positive guides for each set of conditions is highlighted in red. Using the rules output by the RuleFit classifier, conditional control statements were used to filter the initial 84 positive guides from the total set of 192. The features used are n, IQR, min, and PFS_1. Linearized, the decision rules for maximizing positive signals in our set of guides are 1) if IQR is ≥ 6.25, n is between 0 to 3 and pfs_1 is A, U, or C, 2) if IQR is < 6.25, n is between, 0 to 2, and 3) if n = 4, IQR is between 9 and 28 and min is between 1 and 3. These rules achieve 95% positive guides from an initial pool that is 44% positive, including guides with up to four mismatches.
