Robust deep learning–based protein sequence design using ProteinMPNN

Abstract

While deep learning has revolutionized protein structure prediction, almost all experimentally characterized de novo protein designs have been generated using physically based approaches such as Rosetta. Here we describe a deep learning–based protein sequence design method, ProteinMPNN, with outstanding performance in both in silico and experimental tests. On native protein backbones, ProteinMPNN has a sequence recovery of 52.4%, compared to 32.9% for Rosetta. The amino acid sequence at different positions can be coupled between single or multiple chains, enabling application to a wide range of current protein design challenges. We demonstrate the broad utility and high accuracy of ProteinMPNN using X-ray crystallography, cryoEM and functional studies by rescuing previously failed designs, made using Rosetta or AlphaFold, of protein monomers, cyclic homo-oligomers, tetrahedral nanoparticles, and target binding proteins.

Hallucinating symmetric protein assemblies

“…Adversarial samples have been generated by activation maximization in the context of image classification neural networks, which similarly leads to unrealistic outputs (30)(31)(32). To eliminate such over-fitting, we generated new sequences for the HAL backbones using the recently developed ProteinMPNN sequence design neural network (33). For each original backbone, 24 to 48 sequences were generated with ProteinMPNN, and assembly to the target oligomeric structure was validated with AF2 (these dozens of evaluations, compared with the hundreds performed during hallucination, make overfitting much less likely).…”

Between Protein Fold and Nucleophile Identity: Multiscale Modeling of the TEV Protease Enzyme–Substrate Complex

“…Once its ability to predict Michaelis complexes of proteases is proven, it may even be used to screen for protease decoys against a particular substrate peptide for further fine-tuning using methods of computational protein design. 82,83 AlphaFold also turned out to be right about the possibility of another Ser168 conformation. However, all five models produced the same result, thus giving no hint on the possibility of X-ray conformation and strengthening assumptions that AF is biased toward available structural data in PDB.…”

“…However, its overall decent ability to perform protein–peptide docking combined with data presented herein makes it possible to speculate that AlphaFold-multimer might be a general solution to such a task. Once its ability to predict Michaelis complexes of proteases is proven, it may even be used to screen for protease decoys against a particular substrate peptide for further fine-tuning using methods of computational protein design. , …”

Scientists are using AI to dream up revolutionary new proteins

“…Around the same time, another researcher in the lab, machine-learning scientist Justas Dauparas, was developing a deep-learning tool to address what is known as the inverse folding problem -determining a protein sequence that corresponds to a given protein's overall shape 3 . The network, called ProteinMPNN, can act as a 'spellcheck' for designer proteins created using AlphaFold and other tools, says Ovchinnikov, by tweaking sequences while maintaining the molecules' overall shape.…”

“…The vaccine is based on a spherical protein 'nanoparticle' that was created by researchers nearly a decade ago, through a labour-intensive trial-anderror-process 1 . Now, thanks to gargantuan advances in artificial intelligence (AI), a team led by David Baker, a biochemist at the University of Washington in Seattle, reports in Science 2,3 that it can design such molecules in seconds instead of months.…”