HTGAA 2025 Week #5 Homework Protein Design Part II

Part B Option 1 - Phage Engineering Project: Computational Mutant Design and Evaluation

1. Generate Mutants

For this project, we used ProteinMPNN to generate point mutations aimed at improving phage capsid stability under high-temperature conditions. We focused on the coat protein's structural regions identified by AlphaFold2 predictions.

Additionally, AlphaFold-Multimer was employed to simulate phage tail protein binding to bacterial receptor DnaJ, in order to explore mutants that might enhance infection efficiency.

2. Evaluate Designs

Each designed mutant was evaluated using:

Predicted Stability: We examined pLDDT scores from AlphaFold2. High pLDDT (>85) suggests that the mutant remains well-folded.
Binding Prediction: AF2-Multimer results were inspected for changes in predicted binding interfaces with DnaJ.
Membrane Considerations: For tail fiber proteins with transmembrane domains, we checked whether hydrophobicity was maintained in membrane-spanning regions (using a Kyte-Doolittle plot).

Filtering Criteria:

Only mutants with no major structural clashes and high pLDDT were considered.
Mutations introducing charged or polar residues into membrane domains were avoided.

3. Five Selected Mutants

Mutation	Region	Rationale	Evaluation
W39R	Soluble domain	Introduces positive charge to enhance electrostatic binding to DnaJ.	pLDDT remains >90; binding interface gains new salt bridge.
L72M	Structural core	Methionine may stabilize hydrophobic core better at high temperatures.	pLDDT stable; core packing appears improved.
A150T	Surface loop	Adds polar group to enhance solubility without affecting folding.	pLDDT unchanged; surface exposure increases.
G201D	Tail fiber	Introduces negative charge to promote receptor docking.	AF2-Multimer shows new hydrogen bonds with DnaJ.
V250I	Transmembrane domain	Conservative change maintaining hydrophobicity, but slightly improving packing.	Membrane prediction stable; no new clashes.

4. Short Discussion

Limitations:

While AlphaFold2 and ProteinMPNN provide powerful predictive insights, there are important caveats. Some loop regions showed lower confidence in AF2-Multimer predictions, suggesting binding poses might be uncertain. Furthermore, stability predictions are based on static models and do not account for cellular environment stresses like protease activity or thermal denaturation.

Proposed Next Steps: