Fig. 5

The experimentally informed models (ExpCM) correctly identify the three β-lactamase sites in Fig. 1 that contribute to drug resistance. Figure 1 showed five sites in β-lactamase, three of which (238, 240, and 244) experience substitutions that contribute to drug resistance. However, a d N/d S analysis (GY94) fails to identify any of these sites as under diversifying selection (d N/d S>1) at a FDR of 0.05 for testing all sites (dotted blue lines). In contrast, ExpCM correctly determine that the three resistance sites are under diversifying (238 and 244) or differential (238 and 240) selection, and that the two non-resistance sites (201 and 242) are evolving as expected. ExpCM outperform the d N/d S method because they implement a null model that accounts for the site-specific amino-acid preferences shown in Fig. 1; for instance, this null model is not surprised that site 242 remains fixed at the highly preferred amino-acid R, but does find it noteworthy that site 240 substitutes to K multiple times even though that is not a particularly preferred amino acid