“…The detailed methodology and a demonstration of its ability to identify mutants of four proteins with improved fitness have been reported elsewhere, although only one property was tested for each protein (Cadet et al, 2018a;Cadet et al, 2018b). Here, we applied innov'SAR to the problem of multiple parameter optimization, which has been tackled by other researchers, for example, to simultaneously optimize activity and stability (Giver, Gershenson, Freskgard, & Arnold, 1998), enantioselectivity and thermostability (Li, Zhang, Sun, Liu, & Reetz, 2016), thermostability, activity, and stereoselectivity (Acevedo-Rocha et al, 2018), activity and stereoselectivity (Hall, Sarkar, Lee, Munday, & Bell, 2016), or activity, thermostability, and pH range (Zheng et al, 2017). These earlier studies used classical experimental approaches such as iterative saturation mutagenesis (ISM) CASTing, random mutagenesis or recombination, screening, which did not include a machine learning component for fast and efficient screening and selection.…”