Fig. 4

Antibody features contributing to prediction of each function. Bar chart depicts the identity and extent of contribution of the top 30 biophysical measurements in models of each effector function. Feature contributions are reported as the percent increase in model mean squared error (MSE), which quantifies the degradation in model performance when a given feature is permuted. Feature contribution bars are color coded by feature type: features that characterize the magnitude of total IgG or IgA specific for each antigen are denoted as response magnitude measures (titer, black), and those that capture IgG subclass, FcγR, and lectin binding profiles are denoted as qualities of the response (quality, blue). Color bars at left indicate the Fv specificity and Fc characteristic of each feature