Fig 1.
Structures of wild-type and mutant binding sites for known specificity altering mutations.
Close-up images of the substrate binding sites for the ten enzymes in our benchmark with known specificity altering mutations are shown in stick representation. The PDB IDs of the wild-type (green) and mutant (orange) structures are displayed in each panel.
Table 1.
Comparison of fixed backbone and coupled moves methods on predicting specificity altering mutations.
Fig 2.
Performance of computational protein design methods on predicting specificity altering mutations.
Percent of mutations predicted correctly for specificity altering mutations starting from A) the wild-type structure and B) the mutant structure. Results using fixed backbone design (red) and the coupled moves protocol (blue) are shown where protein–ligand interactions are up-weighted (ligand weight = 2.0) or not up-weighted (ligand weight = 1.0).
Fig 3.
Flowchart outlining the coupled moves method.
The protocol starts with an input structure of a protein–ligand interaction, and performs either coupled protein or ligand moves. Each protein move involves a backrub move coupled to side-chain repacking or design and each ligand move involves a rigid-body rotation and translation coupled to ligand repacking. A move is either accepted or rejected depending on the change in energy, and a total of N moves are performed, where N can be set by the user.
Fig 4.
Comparison of the predicted energetic effects of the specificity altering mutations between the fixed backbone and coupled moves methods.
Predicted energies (in Rosetta energy units) for each of the specificity altering mutations for A) one-body interactions and B) two-body interactions of the residue at the mutated position. Scatterplots show a comparison of energies from fixed backbone and coupled moves methods, where each dot denotes a mutation and y = x is shown as a dashed red line. Data points above the diagonal indicate larger (more unfavorable) predicted energies using fixed backbone design. The bottom scatterplots show close-ups of the plot area within the red boxes in the top scatterplots.
Fig 5.
Comparison of models for specificity altering mutations from fixed backbone and coupled moves methods, and crystal structures.
Each row displays an example specificity altering mutation from fixed backbone (magenta) or coupled moves (cyan) models, as well as the crystal structure (yellow) and the superimposition of all three (far right column). Red disks denote steric clashes and dashed black lines denote hydrogen-bonding interactions.
Table 2.
Comparison of fixed backbone and coupled moves methods on predicting co-factor binding site sequences.
Fig 6.
Performance of computational protein design methods on predicting ligand binding site sequences.
A) Boxplot of distributions of profile similarity values between natural and designed sequences for each of the 158 positions in 8 co-factor binding sites. Whiskers denote minimum and maximum, top and bottom of the box indicate 75th and 25th percentile, respectively, and the bold line shows the median. B) Scatterplot comparing profile similarity for each position in sequences designed with fixed backbone and coupled moves methods. y = x is shown as a dashed red line. Data points above the diagonal indicate improved predictions using the coupled moves method.
Fig 7.
Sequence logos for predicted and naturally occurring binding site sequences.
Two representative examples showing the largest (left) and the smallest (right) improvement of coupled moves (middle row) over fixed backbone design (bottom row) with respect to profile similarity with natural sequences (top row). The height of the letter representing each amino acid corresponds to its frequency and the height of each column is inversely proportional to the sequence variation at that position.
Fig 8.
Sequence profile similarity distributions for different subsets of positions and for variations of the coupled moves method.
A) Boxplots of profile similarity distributions for fixed backbone and coupled moves methods separated into three equal-sized groups based on sequence entropy in the natural sequences. B) Boxplots of profile similarity distributions for fixed backbone design and variations of the coupled moves method. Variants include using a Boltzmann distribution (“Boltz SC”) or a uniform distribution (“Uni SC”) to select mutations and side-chain conformations, and incorporating backbone flexibility (“Flex BB”) or using a fixed backbone (“Fix BB”).