De novo protein design by inversion of the AlphaFold structure prediction network

Abstract

De novo protein design enhances our understanding of the principles that govern protein folding and interactions, and has the potential to revolutionize biotechnology through the engineering of novel protein functionalities. Despite recent progress in computational design strategies, de novo design of protein structures remains challenging, given the vast size of the sequence-structure space. AlphaFold2 (AF2), a state-of-the-art neural network architecture, achieved remarkable accuracy in predicting protein structures from amino acid sequences. This raises the question whether AF2 has learned the principles of protein folding sufficiently for de novo design. Here, we sought to answer this question by inverting the AF2 network, using the prediction weight set and a loss function to bias the generated sequences to adopt a target fold. Initial design trials resulted in de novo designs with an overrepresentation of hydrophobic residues on the protein surface compared to their natural protein family, requiring additional surface optimization. In silico validation of the designs showed protein structures with the correct fold, a hydrophilic surface and a densely packed hydrophobic core. In vitro validation showed that 7 out of 39 designs were folded and stable in solution with high melting temperatures. In summary, our design workflow solely based on AF2 does not seem to fully capture basic principles of de novo protein design, as observed in the protein surface's hydrophobic vs. hydrophilic patterning. However, with minimal post-design intervention, these pipelines generated viable sequences as assessed experimental characterization. Thus, such pipelines show the potential to contribute to solving outstanding challenges in de novo protein design.

Computational design of soluble functional analogues of integral membrane proteins

“…Wicky and coworkers 9 have demonstrated the efficiency of using ProteinMPNN on AF2-generated structures to enhance their expression and solubility, but it remained unclear whether it could be successfully applied to explore the sequence space of complex protein folds with intricate topological features including those only found in membrane environments (Fig 1a). To address this challenge, we integrated a previously developed AF2-based design approach (AF2 seq ) 8 with the ProteinMPNN framework (Fig 1b). In this pipeline we used AF2 seq to generate sequences that adopt a desired target fold.…”

“…Gradient descent. As previously described by Goverde et al 8 , the amino acid sequences were initialized based on the secondary structure of the target fold. The secondary structure assignments were then encoded in sequences using alanines for helix, valines for β-sheet and glycines for loop residues.…”

“…With the rise of deep learning-based methods, exploring novel sequence space has become increasingly feasible, allowing the discovery of proteins with stable topologies and new functions. Deep learning powered methods have been impactful in a number of tasks that include the generation of novel designable backbones [4][5][6][7][8] , oligomeric protein assemblies 9,10 , proteins with embedded functional motifs 11 , novel protein structural descriptors 12 , the sequence design problem [13][14][15] , and more recently the generation of a diverse range of protein topologies using diffusion models 5,10,[16][17][18][19] . It was also found that structure prediction networks can be inverted and used for protein design, resulting in the generation of plausible protein backbones 7,8,20 .…”

“…To this end, we developed a computational pipeline for robust de novo protein design based on the inversion of the AF2 network 8 coupled with sequence design using ProteinMPNN 15 (Fig. 1b).…”

Design of intrinsically disordered protein variants with diverse structural properties

“…In parallel with our continuously improving ability to predict structures of folded proteins, there has been substantial development in our ability to design sequences that fold into specific three- dimensional folded structures ( Pan and Kortemme, 2021 ; Woolfson, 2021 ; Goverde et al, 2023 ). Given the multitude of functions and properties of IDPs, there would be a great potential in design- ing IDPs with targeted properties.…”