EvoPro: Rapid In Silico Evolution of High-Affinity Protein Binders

EvoPro is an automated protein design pipeline that uses iterative rounds of deep learning to "evolve" high-affinity binders without the need for extensive wet-lab screening. Developed by the Kuhlman Lab at the University of North Carolina, Chapel Hill, EvoPro couples the sequence optimization power of ProteinMPNN with the structural accuracy of AlphaFold2. By performing these cycles entirely in silico, researchers can rapidly identify therapeutic candidates—like autoinhibitory domains for cancer immunotherapy—with picomolar to nanomolar binding affinities.

Key Innovations: Iterative Deep Learning Evolution

EvoPro differentiates itself from "one-shot" design tools by using a cyclic genetic algorithm framework that mimics natural selection through deep learning models.

• Cyclic Design Loop: Iterates between sequence diversification (via random mutagenesis or ProteinMPNN) and structural scoring (via AlphaFold2).

• Multi-Objective Scoring: Evaluates candidates based on three critical structure-based metrics derived from AF2:

  • Placement Confidence: Uses Predicted Aligned Error (PAE) to assess the quality of the binding interface.

  • Fold Confidence: Uses pLDDT to ensure the designed miniprotein folds stably on its own.

  • Conformational Stability: Minimizes the structural change required for the binder to transition from an unbound to a bound state.

• No-Prior Versatility: Successfully generates high-affinity binders starting from de novo hyperstable scaffolds, even for targets with no known natural binding partners.

• Cross-Modality Potential: Recent updates enable multistate design and support for non-protein molecules through integration with AlphaFold 3.

Performance Benchmarks

In experimental validations targeting the critical cancer protein PD-L1, EvoPro generated binders that matched or exceeded the performance of traditional affinity maturation.

Scientific Breakthroughs in Smart Therapeutics

Autoinhibitory "Masked" Antagonists

EvoPro was used to design Autoinhibitory Domains (AiDs) for a PD-L1 antagonist. These domains act as a "mask" that prevents the drug from binding to its target in healthy tissue. When the drug reaches a tumor, tumor-enriched proteases cleave a linker, unmasking the drug only where it is needed—potentially reducing the toxic side effects of standard immunotherapy.

High-Precision Interface Engineering

Despite starting with diverse topologies (e.g., 2H3E, 2H4E), EvoPro converged on unique contacts that are structurally distinct from the natural PD-L1 interface. A key discovery included the consistent "hallucination" of a hydrophobic residue insertion into a specific pocket on the target, mimicking a natural binding principle with a purely synthetic geometry.

EvoPro on Tamarind Bio: Design Without Barriers

Tamarind Bio provides a managed, high-performance environment to run EvoPro’s compute-intensive evolutionary cycles without managing complex GPU clusters or software dependencies.

• No-Code Evolutionary Runs: Launch multi-cycle design trajectories through an intuitive interface.

• Managed AlphaFold Orchestration: Access high-speed structure prediction and sequence design (ProteinMPNN) out of the box.

How to Use EvoPro on Tamarind Bio

1. Access the Platform: Log in to tamarind.bio and select the EvoPro tool.

2. Define Your Target: Upload the PDB structure of the protein you wish to bind (e.g., PD-L1 or an enzyme active site).

3. Choose Starting Scaffolds: Provide your own scaffolds or select from a database of de novo hyperstable miniproteins.

4. Set Evolutionary Parameters: Define the number of iterations and the size of the sequence pool for optimization.

5. Run Optimization: The platform executes the cyclic loop, using ProteinMPNN for sequence "mutation" and AlphaFold2 for structural "selection".

6. Analyze & Filter: Use the interactive PAE/pLDDT reports and the Rosetta energy metrics to identify the top-ranked binders for experimental synthesis.

Source