Week 5 HW：Protein Design Part-ii

Part A: SOD1 Binder Peptide Design

Part 1: Generate Binders with PepMLM

Generate four peptides of length 12 amino acids conditioned on the mutant SOD1 sequence.

Among the generated sequences, WRYPAAAAELKK (7.61) stands out the most. It has a lower perplexity than the other generated sequences, indicating that it may be more compatible with the pocket of the SOD1 A4V mutation site in terms of chemical environment and geometric conformation.

Part 2: Evaluate Binders with AlphaFold3

For each peptide, I submit the mutant SOD1 sequence followed by the peptide sequence as separate chains to model the protein-peptide complex. And I analysis these 4 peptide from different perspective inculding engagement of the β-barrel region,surface-bound, ipTM values ,peptide matches.

Part 3: Evaluate Properties of Generated Peptides in the PeptiVerse

1. Therapeutic Properties and Structural Observations: Comparative Analysis

By comparing the predicted data from PeptiVerse with the structural observations of AlphaFold3, we can draw the following conclusions:

Conformity between Binding Force and ipTM: Generally, short peptides with higher ipTM scores exhibit more stable interfacial interactions. In your prediction list, WRYPAAAAELKK showed the lowest perplexity score (7.61), which typically indicates stronger binding potential.

Solubility & Hemolysis:

Solubility: All candidate peptides (Binder 0-3) were predicted as Soluble (probability 1.000), which is a very desirable signal in therapeutic development.

Hemolysis: Most designs performed well and were predicted as Non-hemolytic. WRYPAAAAELKK showed an extremely low probability of hemolysis (0.005), indicating very high safety. In comparison, WLYYVVAVAWKK has a slightly higher hemolysis prediction probability (0.172), and although it is still classified as non-hemolytic, its safety profile is slightly inferior to other candidates.

The best performer in terms of attribute balance: WRYPAAAAELKK achieves the optimal balance between binding confidence, complete solubility, and extremely low risk of hemolysis.

2. Final Candidate Short Peptide Selection and Rationale

WRYPAAAAELKK

Rationale: This short peptide stood out in the PepMLM score, possessing the lowest Pseudo Perplexity (7.61), indicating the model’s strongest confidence in its binding to SOD1 (A4V). In therapeutic property prediction, it not only demonstrated perfect solubility prediction (1.000) but also exhibited excellent safety indicators, with an ultra-low hemolysis probability of 0.005, making it the most robust therapeutic candidate. Its acid-base balance (pI 9.38) and hydrophobicity index (GRAVY -0.72) also suggest good biochemical stability in the complex physiological environment within cells.

Part 4: Generate Optimized Peptides with moPPIt

Using the Multi-Objective Guided Discrete Flow Matching (MOG-DFM) framework, two highly optimized 12-mer peptide candidates were generated targeting the specific motif of the A4V mutant SOD1 sequence:

Candidate 1: CTQGMNVAYEWL

Hemolysis Score: 0.0564 (Lower means safer/less toxic)

Solubility Score: 0.9998 (Highly soluble)

Affinity Score: 5.2750

Motif Score: 0.3151

Candidate 2: VCEEWQLEFLCE

Hemolysis Score: 0.0324 (Excellent safety profile)

Solubility Score: 0.9998 (Highly soluble)

Affinity Score: 5.8080 (Slightly stronger predicted binding than Candidate 1)

Motif Score: 0.2732

2. Comparative Analysis: moPPIt vs. PepMLM Peptides

When comparing the moPPIt-generated peptides against my selected PepMLM peptide (WRYPAAAAELKK), the results demonstrate a major paradigm shift from probabilistic sampling to controlled, multi-objective design:

Spatial Specificity (Targeted vs. Blind Sampling): My PepMLM peptide was sampled purely based on statistical plausibility (Pseudo Perplexity: 7.61) across the entire SOD1 sequence, offering no control over the exact binding site. In contrast, moPPIt utilizes an explicit Motif guidance constraint. This forces the generative flow to target the specific pathological pocket—such as the A4V mutation site (residue 4) or the dimer interface—making moPPIt highly site-specific and biologically targeted.

Property Optimization (Embedded Flow vs. Screening Luck): While my PepMLM peptide yielded excellent baseline properties (Solubility: 1.000, Hemolysis: 0.005), this relies entirely on post-generation screening luck—if a sampled peptide fails, the user must blindly resample. Conversely, moPPIt integrates property optimization directly into its generative flow. It simultaneously balanced high affinity (5.27−5.80) and motif alignment while inherently securing near-perfect solubility (0.9998) and safe hemolysis profiles (0.03–0.05).