Fig. 2.

Comparison between predicted and actual carboxylation turnover rate (Kcat: s^-1), Michaelis-Menten constant for CO₂ at ambient O₂ (Kc^21%O2: µM) and specificity for CO₂ over O₂ (Sc/o: mol mol^-1) at 25 °C. The performance was determined using leave-one-out cross-validation with the learned encoding scheme (Rives et al., 2021) (green) and classical encoding scheme (orange). The better performance of the learned encodings with an additive non-linear kernel justified the adoption of this method over classical for the final machine learning tasks.