Research

Background

Research in the Hayes group is focused on accurate computational methods for protein engineering and design. Proteins carry out a multitude of biological functions, and thus are a flexible platform to engineer new useful functions for many applications. Proteins are useful as biocatalysts that synthesize valuable chemicals with greater site and stereo selectivity under mild conditions, and can provide new or modified components for metabolic engineering. The specificity of protein interactions makes them promising candidates for therapeutics with fewer side effects. Proteins can also be engineered as capsids, structural scaffolds, or molecular motors for biotechnology applications.

Experimental protein engineering is a mature field including methods like directed evolution. Directed evolution iteratively optimizes a protein through cycles of variation and selection. However, it usually requires some low initial level of the new desired function along with a high throughput assay for that function. That initial level of function can be obtained by finding a promiscuous natural protein or computationally designing one.

Computational protein design is also a maturing field led by methods such as Rosetta. Rosetta has designed proteins with a myriad of structural, binding, catalytic, and other properties, which are all problems related to optimizing the free energy of a particular process. Due to approximations in the free energy estimation, success rates for individual candidate designs vary widely, but this is routinely overcome through extensive experimental screening. Consequently, more accurate computational methods are needed for difficult design targets with low success rates or tedious functional assays.

Methods Development

Projects in the Hayes group include several protein design applications as well as the methods development needed to realize these applications. The projects are supported by Hayes’ previous development of physics-based methods including multisite λ dynamics and growing interest in machine learning approaches.

Multisite λ dynamics is a member of the class of alchemical free energy methods, which estimate free energy from molecular dynamics simulations. These methods provide more accurate free energies than Rosetta, and are thus an appealing starting point for raising protein design success rates. Alchemical methods estimate the ΔΔG for a slow physical process like binding or folding upon a chemical change such as mutation. They do this by “alchemically” observing the chemical transition during simulation, rather than waiting to observe the slow physical transition. Multisite λ dynamics is uniquely well-suited for protein design because of its ability to perform mutations at many sites simultaneously during a simulation.

Multisite λ dynamics has been shown to have excellent accuracy in evaluating point mutations, and can also predict the effect of up to 15 simultaneous mutations with an unprecedented level of accuracy. These demonstrations are supported by the development of adaptive landscape flattening (ALF), which accelerates transitions and exploration through chemical space, and a basic lambda dynamics engine (BLaDE), which has decreased the computational cost of simulations by nearly an order of magnitude.

There are several ongoing methods development projects in the group. We are working to increase the number of perturbations possible at a single site beyond the current practical limit of about 9, to enable sampling all 20 amino acids at once. An osmostat is needed to allow fluctuations in the number of ions within a simulation to mitigate artifacts from charge changing mutations. Finally, we are looking to improve ALF to screen much larger numbers of ligands for computer aided drug design applications. (Alchemical methods have already found broad use in big pharma for computer-aided drug design.)

Antibodies

Antibodies are natural proteins selected by the immune system to bind antigens with high affinity and specificity. By harnessing natural antibody maturation processes, antibodies can be engineered for therapeutic and other purposes. Therefore, antibodies are a natural place to begin designing proteins for biomedical impact.

The first antibody design project in the group involves redesigning antibodies to target new antigens. As pathogenic antigens mutate, previously identified antibodies lose affinity for the new targets, but rescue mutations to the antibody should be able to restore affinity. Being able to identify such rescue mutations would enable rapid development of therapeutics during public health crises, for example by using SARS antibodies to target SARS-CoV-2, or SARS-CoV-2 antibodies to target the omicron variant. We are using multisite λ dynamics to identify the loss of affinity for mutating from SARS to SARS-CoV-2, and the mutations required to restore binding.

The second antibody project in the group involves thermostabilization of antibodies. Antibodies typically have shorter shelf lives than many small molecule therapeutics, but the shelf life can be improved by increasing the stability of antibodies. We are working to use multisite λ dynamics to more accurately identify stabilizing mutations to antibodies.

De Novo Design

One of the early tests of conventional protein design methods like Rosetta was de novo design of new protein sequences. Thus, designing new sequences with multisite λ dynamics will mark its maturity as a protein design method. The ability to design new sequences also ensures the ability of multisite λ dynamics to maintain protein stability while making much larger jumps in sequence space to optimize new functions.

Three helix bundles are one of the simplest topologies (or protein shapes) to design and are a natural starting point. Next we will explore more difficult topologies which have lower success rates with Rosetta.

Biocatalysis

Hayes is excited to be starting at UCI because there are already several experimental labs here working on protein engineering. Most of these labs are working on biocatalysis, and early collaborations will focus on these projects.