Figure 1. Correlation of predicted versus observed log(k_cat) (A) and log(K_m) (B) values derived from the proteochemometric models using substrate and protease descriptors, and square-terms of substrate descriptors. Predictive ability is assessed by cross-validation leaving out 1/7 of substrates at a time (gray diamonds) and by external validation (black circles). Shown is also model fit (small unfilled squares). Indicated by the oblique gray lines is the plot area where prediction errors do not exceed 0.5 logarithmic units.

Figure 2. Regression coefficients of proteochemometric models of log(k_cat) (A) and log(K_m) (B) data. In each panel, shown are regression coefficients for the four protease subtype descriptors (on the left) and descriptors of six properties for the four substrate residues, P1′–P4′. Solid black bars represent regression coefficients for original descriptors (Note that negative values of hydrophilicity descriptors are seen for hydrophobic amino acids whereas positive values are seen for hydrophilic amino acids; negative values of size descriptor characterize small, compact amino acids, while positive values are seen among large, bulky amino acids; for polarity descriptor, a negative value indicates that the amino acid is electrophilic, whereas a positive value indicates that it is electronegative. The charge descriptor reflects the charge of the amino acid. Positive values of the rigidity and flexibility descriptors indicate the presence of, respectively, a high fraction of rigid and flexible fragments in an amino acid, while negative values indicate the lack of such fragments.) and gray bars represent regression coefficients for their square-terms.

Kinetic constants obtained with DEN-1–4 proteases using novel peptide substrates modified within the P1′–P4′ sequence with the general structure Abz–RRRR↓XXXX–nY–NH₂

Results of proteochemometric modeling of log k_cat data using different combinations of substrate and protease descriptors

Results of proteochemometric modeling of log K_m data using different combinations of substrate and protease descriptors