Figure 5.

Feature importances for various machine learning algorithms and featurizations. LASSO feature importances are coefficients: a positive coefficient indicates a positive correlation between a base and translation efficiency, a negative coefficient indicates a negative correlation. In RFR, feature importances are always positive and therefore not indicative of the directionality of the correlation. (A) Feature importances for algorithms using BPP featurization. (B) Feature importances for algorithms using one-hot encoding. Since only every third one-hot encoded base of the coding sequence varies, only every third base of the coding sequence was plotted. (C) Feature importances for algorithms using BPP + one-hot featurization. BCD = bicistronic design 5′ untranslated region; RBS = ribosome binding site; CDS = coding sequence; TER = terminator.