Week 5 HW: Protein Design Part II

Part 1 — Generate Binders with PepMLM

PepMLM was used to generate four 12-amino-acid peptides conditioned on the A4V mutant SOD1 sequence. The generated peptides showed perplexity values ranging from approximately 5.8 to 7.0, suggesting moderate model confidence in their plausibility as binders. Lower perplexity indicates that the peptide sequence better fits the model’s learned representation of protein–peptide interactions. The known SOD1-binding peptide FLYRWLPSRRGG was included as a reference for downstream structural and functional comparison.

Part 2 — Evaluate Binders with AlphaFold3

AlphaFold3 modeling predicted moderate interaction confidence for the generated peptides, with ipTM values ranging from 0.55 to 0.68. Peptide P2 produced the highest ipTM score (0.68), slightly exceeding the known SOD1-binding peptide (0.65). Structural inspection suggested that several peptides localize near the N-terminal region where the A4V mutation occurs, while others bind the surface of the β-barrel or approach the dimer interface. These results indicate that PepMLM-generated peptides can achieve comparable structural binding confidence to the experimentally known binder.

Part 3 — Evaluate Peptides in PeptiVerse

PeptiVerse predictions revealed differences in therapeutic properties among the generated peptides. Peptide P2 showed the strongest predicted binding affinity, consistent with its high ipTM score from AlphaFold3. However, P4 demonstrated a more balanced profile, combining high predicted affinity with strong solubility and low hemolysis probability. Peptide P3 displayed lower solubility and a moderate hemolysis risk, which could limit therapeutic potential. Overall, peptides with higher structural confidence tended to show stronger predicted binding affinity, although favorable therapeutic properties were not always correlated with ipTM scores.

Part 4 — Optimized Peptides with moPPIt

Peptides generated with moPPIt differed from PepMLM peptides in that they were explicitly optimized for multiple objectives simultaneously, including binding affinity, motif targeting, solubility, and reduced hemolysis risk. As a result, moPPIt peptides tended to contain residue patterns consistent with known interaction motifs while maintaining physicochemical properties favorable for therapeutic development. Compared with the PepMLM peptides, which were generated through sequence-conditioned sampling, moPPIt designs appeared more targeted toward specific binding regions on SOD1, particularly near the A4V mutation site.

How to evaluate before clinical studies

Before advancing these peptides toward clinical studies, several additional evaluation steps would be necessary. First, structural validation using AlphaFold3 or molecular docking could confirm stable peptide–protein interfaces. Molecular dynamics simulations could then assess binding stability and conformational flexibility over time. In vitro experiments such as surface plasmon resonance or fluorescence binding assays would be required to measure actual binding affinity. Additionally, toxicity, protease stability, and aggregation assays would help determine whether the peptides are suitable therapeutic candidates.