Week 5: Protein Design II

Part A: SOD1 A4V Therapeutic Peptide Design

1. Project Overview & Pharmacological Target

This research targets the A4V mutation (Alanine-to-Valine at residue 4) in human Superoxide Dismutase 1 (SOD1). In Pharmaceutical Sciences, this is a critical target for Familial ALS. The mutation destabilizes the N-terminal “zipper” of the protein, leading to the exposure of hydrophobic residues and subsequent toxic aggregation. Our goal is to design a peptide binder that cap-stabilizes this region.

2. Candidate Generation (PepMLM)

We utilized PepMLM-650M to generate binder candidates via Masked Language Modeling. By conditioning the model on the SOD1 A4V pocket, we identified sequences with high statistical likelihood of interaction.

Peptide IDSequencePerplexityipTM (AF3)
Pep-3 (Lead)RHVPVAVAAQWWX8.950.30
Pep-1WWVYVVAGVLEWX20.680.31
PepMLM Results PepMLM Results
Warning

Toxicological De-risking: Peptide 1 showed high structural affinity but was excluded after PeptiVerse screening flagged it for high hemolytic activity, making it unsuitable for systemic pharmaceutical application.

​

Peptide 3 localizes effectively near the N-terminal mutation site to provide scaffolding stability. AF3 Peptide 3 AF3 Peptide 3

PeptiVerse screening dashboard showing the exclusion of toxic candidates. PeptiVerse Pep1 PeptiVerse Pep1

Access Data: PepMLM Sequences (CSV) | moPPIt Optimization (CSV)

Part C: MS2 Lysis Protein (L-Protein) Mutants

1. Objective: Stability & Folding in Phage Therapy

The Lysis Protein of the MS2 bacteriophage is a single-pass transmembrane protein. Improving its stability and auto-folding is essential for developing robust phage-based alternatives to traditional antibiotics, specifically targeting cell wall synthesis inhibition.

2. The “Correlation Gap”: PLM Predictions vs. Lab Reality

A major focus of this computational track was comparing ESM-2 (Protein Language Model) scores against experimental lysis data.

  • The Discrepancy: Analysis of protein_mutations_scores.csv identified C29R as the top-scoring mutation (+2.40). However, the experimental record in L-Protein_Mutants_-_Sheet1.csv confirmed Lysis = 0.
  • Insight: This highlights the limitations of PLMs in predicting functional outcomes versus simple structural stability.

3. Structural Validation: K50L Mutant (ColabFold)

We performed high-resolution structural modeling for K50L (the highest ESM score in the protein at +2.56) using AlphaFold2-Multimer.

​

Visualizing the K50L mutation’s impact on transmembrane helix orientation. Red regions indicate Very Low (<50) pLDDT, suggesting disordered or flexible regions. AlphaFold Structure AlphaFold Structure

Analysis of pLDDT and PAE matrices. The confidence scores confirm a stable fold despite the mutation. Confidence Plots Confidence Plots

Multiple Sequence Alignment (MSA) coverage for the L-protein variant. Sequence Coverage Sequence Coverage

4. Final Mutation Selection Logic

The following 5 mutations were selected for their balance of computational potential and experimental proof:

  1. S9Q (Soluble): High ESM score (+2.01); likely improves N-terminal solubility.
  2. R30Q (Soluble): Confirmed Lysis = 1; maintains functional cell wall degradation.
  3. K50L (TM): Theoretical stability peak (+2.56) in the transmembrane domain.
  4. A45P (TM): Validated Lysis = 1; potential helix-breaker that aids the lysis mechanism.
  5. K23E (Free): Highly robust candidate with three separate experimental confirmations of Lysis = 1.

5. Final Conclusion

This workflow demonstrates that while models like ESM-2 and AlphaFold are powerful for scanning protein landscapes, experimental benchmarking remains the gold standard for identifying functional variants in phage-based antimicrobials.

Access Data: L-Protein Mutations (CSV) | ESM Scores (CSV) | Homology Report (TXT)