Combining machine learning with structure-based protein design to predict and engineer post-translational modifications of proteins.

Post-translational modifications (PTMs) of proteins play a vital role in their function and stability. These modifications influence protein folding, signaling, protein-protein interactions, enzyme activity, binding affinity, aggregation, degradation, and much more. To date, over 400 types of PTMs have been described, representing chemical diversity well beyond the genetically encoded amino acids. Such modifications pose a challenge to the successful design of proteins, but also represent a major opportunity to diversify the protein engineering toolbox. To this end, we first trained artificial neural networks (ANNs) to predict eighteen of the most abundant PTMs, including protein glycosylation, phosphorylation, methylation, and deamidation. In a second step, these models were implemented inside the computational protein modeling suite Rosetta, which allows flexible combination with existing protocols to model the modified sites and understand their impact on protein stability as well as function. Lastly, we developed a new design protocol that either maximizes or minimizes the predicted probability of a particular site being modified. We find that this combination of ANN prediction and structure-based design can enable the modification of existing, as well as the introduction of novel, PTMs. The potential applications of our work include, but are not limited to, glycan masking of epitopes, strengthening protein-protein interactions through phosphorylation, as well as protecting proteins from deamidation liabilities. These applications are especially important for the design of new protein therapeutics where PTMs can drastically change the therapeutic properties of a protein. Our work adds novel tools to Rosetta's protein engineering toolbox that allow for the rational design of PTMs.

XENet: Using a new graph convolution to accelerate the timeline for protein design on quantum computers.

Graph representations are traditionally used to represent protein structures in sequence design protocols in which the protein backbone conformation is known. This infrequently extends to machine learning projects: existing graph convolution algorithms have shortcomings when representing protein environments. One reason for this is the lack of emphasis on edge attributes during massage-passing operations. Another reason is the traditionally shallow nature of graph neural network architectures. Here we introduce an improved message-passing operation that is better equipped to model local kinematics problems such as protein design. Our approach, XENet, pays special attention to both incoming and outgoing edge attributes. We compare XENet against existing graph convolutions in an attempt to decrease rotamer sample counts in Rosetta's rotamer substitution protocol, used for protein side-chain optimization and sequence design. This use case is motivating because it both reduces the size of the search space for classical side-chain optimization algorithms, and allows larger protein design problems to be solved with quantum algorithms on near-term quantum computers with limited qubit counts. XENet outperformed competing models while also displaying a greater tolerance for deeper architectures. We found that XENet was able to decrease rotamer counts by 40% without loss in quality. This decreased the memory consumption for classical pre-computation of rotamer energies in our use case by more than a factor of 3, the qubit consumption for an existing sequence design quantum algorithm by 40%, and the size of the solution space by a factor of 165. Additionally, XENet displayed an ability to handle deeper architectures than competing convolutions.

Perturbing the energy landscape for improved packing during computational protein design.

The FastDesign protocol in the molecular modeling program Rosetta iterates between sequence optimization and structure refinement to stabilize de novo designed protein structures and complexes. FastDesign has been used previously to design novel protein folds and assemblies with important applications in research and medicine. To promote sampling of alternative conformations and sequences, FastDesign includes stages where the energy landscape is smoothened by reducing repulsive forces. Here, we discover that this process disfavors larger amino acids in the protein core because the protein compresses in the early stages of refinement. By testing alternative ramping strategies for the repulsive weight, we arrive at a scheme that produces lower energy designs with more native-like sequence composition in the protein core. We further validate the protocol by designing and experimentally characterizing over 4000 proteins and show that the new protocol produces higher stability proteins.

Stabilizing proteins, simplified: A Rosetta-based webtool for predicting favorable mutations.

Many proteins have low thermodynamic stability, which can lead to low expression yields and limit functionality in research, industrial and clinical settings. This article introduces two, web-based tools that use the high-resolution structure of a protein along with the Rosetta molecular modeling program to predict stabilizing mutations. The protocols were recently applied to three genetically and structurally distinct proteins and successfully predicted mutations that improved thermal stability and/or protein yield. In all three cases, combining the stabilizing mutations raised the protein unfolding temperatures by more than 20°C. The first protocol evaluates point mutations and can generate a site saturation mutagenesis heatmap. The second identifies mutation clusters around user-defined positions. Both applications only require a protein structure and are particularly valuable when a deep multiple sequence alignment is not available. These tools were created to simplify protein engineering and enable research that would otherwise be infeasible due to poor expression and stability of the native molecule.

From Protein Design to the Energy Landscape of a Cold Unfolding Protein.

Understanding protein folding is crucial for protein sciences. The conformational spaces and energy landscapes of cold (unfolded) protein states, as well as the associated transitions, are hardly explored. Furthermore, it is not known how structure relates to the cooperativity of cold transitions, if cold and heat unfolded states are thermodynamically similar, and if cold states play important roles for protein function. We created the cold unfolding 4-helix bundle DCUB1 with a de novo designed bipartite hydrophilic/hydrophobic core featuring a hydrogen bond network which extends across the bundle in order to study the relative importance of hydrophobic versus hydrophilic protein-water interactions for cold unfolding. Structural and thermodynamic characterization resulted in the discovery of a complex energy landscape for cold transitions, while the heat unfolded state is a random coil. Below ∼0 °C, the core of DCUB1 disintegrates in a largely cooperative manner, while a near-native helical content is retained. The resulting cold core-unfolded state is compact and features extensive internal dynamics. Below -5 °C, two additional cold transitions are seen, that is, (i) the formation of a water-mediated, compact, and highly dynamic dimer, and (ii) the onset of cold helix unfolding decoupled from cold core unfolding. Our results suggest that cold unfolding is initiated by the intrusion of water into the hydrophilic core network and that cooperativity can be tuned by varying the number of core hydrogen bond networks. Protein design has proven to be invaluable to explore the energy landscapes of cold states and to robustly test related theories.

Rapid Sampling of Hydrogen Bond Networks for Computational Protein Design.

Hydrogen bond networks play a critical role in determining the stability and specificity of biomolecular complexes, and the ability to design such networks is important for engineering novel structures, interactions, and enzymes. One key feature of hydrogen bond networks that makes them difficult to rationally engineer is that they are highly cooperative and are not energetically favorable until the hydrogen bonding potential has been satisfied for all buried polar groups in the network. Existing computational methods for protein design are ill-equipped for creating these highly cooperative networks because they rely on energy functions and sampling strategies that are focused on pairwise interactions. To enable the design of complex hydrogen bond networks, we have developed a new sampling protocol in the molecular modeling program Rosetta that explicitly searches for sets of amino acid mutations that can form self-contained hydrogen bond networks. For a given set of designable residues, the protocol often identifies many alternative sets of mutations/networks, and we show that it can readily be applied to large sets of residues at protein-protein interfaces or in the interior of proteins. The protocol builds on a recently developed method in Rosetta for designing hydrogen bond networks that has been experimentally validated for small symmetric systems but was not extensible to many larger protein structures and complexes. The sampling protocol we describe here not only recapitulates previously validated designs with performance improvements but also yields viable hydrogen bond networks for cases where the previous method fails, such as the design of large, asymmetric interfaces relevant to engineering protein-based therapeutics.