[Paper Review] De novo design of high-affinity protein binders with AlphaProteo

Vinícius Zambaldi, David La+30 more|arXiv (Cornell University)|Sep 12, 2024

Monoclonal and Polyclonal Antibodies Research46 citations

TL;DR

AlphaProteo designs de novo protein binders using a two-component system (generative model and filter) to achieve sub-nanomolar to low-nanomolar affinities with high experimental success across diverse targets, often in a single round of medium-throughput screening without further optimization.

ABSTRACT

Computational design of protein-binding proteins is a fundamental capability with broad utility in biomedical research and biotechnology. Recent methods have made strides against some target proteins, but on-demand creation of high-affinity binders without multiple rounds of experimental testing remains an unsolved challenge. This technical report introduces AlphaProteo, a family of machine learning models for protein design, and details its performance on the de novo binder design problem. With AlphaProteo, we achieve 3- to 300-fold better binding affinities and higher experimental success rates than the best existing methods on seven target proteins. Our results suggest that AlphaProteo can generate binders "ready-to-use" for many research applications using only one round of medium-throughput screening and no further optimization.

Motivation & Objective

• Motivate computational design of high-affinity protein binders for diverse targets without iterative optimization.
• Introduce AlphaProteo as a generative-design + filtering framework.
• Demonstrate experimental validation across multiple structurally diverse targets.
• Show that designed binders are thermostable, expressible, and epitope-specific.

Proposed method

• Two-component AlphaProteo system: a generative model trained on PDB structure/sequence data plus a distillation set of AlphaFold predictions.
• A filter that scores designs to predict experimental success.
• Input target structure with optional hotspot residues to condition design.
• In silico generation of many binder candidates followed by automated filtering to a smaller experimental set.
• Experimental testing via yeast surface display to measure binding hits and affinities (K_D).
• Biochemical, biophysical, and structural validation (HTRF, BLI, CD, cryo-EM, X-ray crystallography).

Experimental results

Research questions

• RQ1 Can AlphaProteo generate de novo binders against structurally diverse targets with high experimental success?
• RQ2 What binding affinities (K_D) can be achieved without post-design optimization across multiple targets?
• RQ3 Do designed binders engage the intended target epitopes and demonstrate specificity?
• RQ4 Are the binders structurally diverse and thermally stable enough for practical use?
• RQ5 How well does the in silico performance generalize to additional, untested targets?

Key findings

• AlphaProteo achieved experimental success rates ranging from 9% to 88% across seven targets, outperforming the best existing methods for six of these targets.
• Best per-target K_D values were sub-1 nM for four targets, and sub-30 nM for seven targets, with the top binder at 82 pM.
• Across seven targets, AlphaProteo binders showed 3- to 300-fold better binding affinities than the best previous designed binders.
• Designed binders are structurally diverse (all-alpha, all-beta, and mixed folds) and bind specifically to intended epitopes, with competitive inhibition and interface-mutant analyses supporting the designed interactions.
• Cryo-EM and X-ray crystallography confirmed binder-target complex structures with high structural agreement to designs (RMSD ~0.8–3.1 Å for cryo-EM; 0.78 Å for VEGF-A complex).
• Functional validation demonstrated SARS-CoV-2 neutralization in live-virus assays and inhibition of VEGF signaling in human cells by designed binders.