AI designs new drugs based on protein structures

Table Of Content

• Machine learning for functional protein design
• Applications and examples of designed proteins
• Targeted genome-modification tools and their advanced applications in crop breeding
• Nature: “‘A landmark moment’: scientists use AI to design antibodies from scratch”
• How to create your protein label design
• Extended Data Fig. 4 RFdiffusion can condition on fold information to generate specific, thermostable folds.

We would like to thank the Rosetta community and members of the Kortemme lab for many contributions to computational design and insightful discussions. Whether you’re a student, researcher, partner, or supporter, there’s a place for you in the protein design revolution. Protein TherapeuticsComputer-generated molecules that reprogram cells, block infection, capture toxins, and more. Proteins can do remarkable things, from blocking infection to harnessing solar energy.

Machine learning for functional protein design

Small molecule-based methods are effective, but they are restricted by molecular diffusion through the plasma membrane and cell walls. In principle, light would be an ideal stimulus because illumination can be rapidly switched on or off, resulting in essentially instantaneous addition or removal of signal. Moreover, it can reach any part of the cell, a property that is not necessarily true for small molecules. Most strategies modify natural plant photoreceptors to create light sensitive proteins. One popular choice is the light oxygen voltage domain from phototropin (LOV2), which consists of a core flavin mononucleotide-binding domain followed by a C-terminal Jα helix.

Applications and examples of designed proteins

Advanced computational techniques, including novel machine learning algorithms, allow our scientists to model the behavior of proteins at the atomic level. This knowledge helps researchers generate novel proteins with optimized stability, binding affinity, or catalytic activity. Protein design is the process of creating new proteins with specific properties by manipulating the sequence of amino acids. It involves a combination of computational techniques, laboratory experiments, and interdisciplinary knowledge from fields such as biology, chemistry, and physics. To date, companies operating in the protein design space have largely focused on retooling existing proteins to perform new tasks or enhance specific properties, rather than true design from scratch. For example, scientists at Generate Biomedicines have drawn on existing knowledge about the SARS-CoV-2 spike protein and its interactions with the receptor protein ACE2 to design a synthetic protein that can consistently block viral entry across diverse variants.

Targeted genome-modification tools and their advanced applications in crop breeding

The interacting residues are combined into binding sites by Monte Carlo–simulated annealing (138) or built onto backbone scaffolds by an algorithm called Convergent Motifs for Binding Sites (25). The Convergent Motifs for Binding Sites method was applied to engineer de novo proteins that bind the drug apixaban with low and submicromolar affinity (Fig. 5A). The development of computational methods for de novo protein design in the last two decades has expanded the scope of designable protein structures and functions considerably.

For example, many viral glycoproteins are trimeric and symmetry matched arrangements of inhibitory domains can be extremely potent43,44,45,46. Conversely, symmetric presentation of viral epitopes in an arrangement that mimics the virus could induce new classes of neutralizing antibodies47,48. To explore this general direction, we sought to design trimeric multivalent binders to the SARS-CoV-2 spike protein. In previous work, flexible linkage of a binder to the ACE2 binding site (on the spike protein receptor binding domain) to a trimerization domain yielded a high-affinity inhibitor that had potent and broadly neutralizing antiviral activity in animal models43. Ideally, however, symmetric fusions to binders would be rigid, so as to reduce the entropic cost of binding while maintaining the avidity benefits from multivalency.

Illumination with blue light triggers formation of a covalent bond between the excited flavin and a cysteine residue in the core domain, which induces a conformational rearrangement that results in unfolding of the Jα helix. Renicke et al. fused a short, synthetic, destabilizing domain from murine ornithine decarboxylase (cODC1) to LOV2 to create a photosensitive degron.79 cODC1 is degraded through an ubiquitin independent mechanism, one of the requirements for which is exposure of a short unstructured region. Attaching cODC1 immediately after the Jα helix produced a protein that is only degraded when illuminated with blue light (Figure 18).

Underlying these successful applications are developments of computational design principles over the last decades. Many such principles have been learned from the wealth of existing architectures in the Protein Data Bank (PDB) (16). While many computational design applications modify existing proteins (12, 17, 18, 19, 20), it is becoming possible to design both structures and functions entirely de novo (1). It was recognized early that variations of helical architectures could be designed based on parametric equations (21). Helical bundle proteins have indeed proven to be very “designable” (22) and have consequently been adapted to many functions (13, 14, 15, 23, 24, 25, 26, 27). More recent developments have expanded the structural repertoire of de novo proteins to other fold classes (28, 29, 30, 31, 32).

Extended Data Fig. 4 RFdiffusion can condition on fold information to generate specific, thermostable folds.

"This means that when designing a drug molecule, we can be sure that it has as few side effects as possible," Atz says. Biological validation is an extremely important consideration for investors in this sector, says van Stekelenburg. “If you are doing de novo, the real gold standard is not which architecture are you using — it’s what percentage of your designed proteins had the end desired property,” she says.

Mega-scale experimental analysis of protein folding stability in biology and design - Nature.com

Mega-scale experimental analysis of protein folding stability in biology and design.

Posted: Wed, 19 Jul 2023 07:00:00 GMT [source]

Diffusion model expands RoseTTAFold’s power

The developers of the online game Foldit (70) crowd-sourced solutions for the challenge of de novo protein design (71). Online Foldit players were provided with a set of tools to generate, mutate, move, and score protein structures. Starting from a fully extended peptide chain, players were able to fold the chain into de novo structures and stabilize the structures by sequence optimization.

Drawing inspiration from these natural marvels, we seek to design equally useful molecules from scratch. Schematic illustration of a systematic, minimal approach to the design of the four-helix bundle protein. Protein design weaves together principles from biology, chemistry, and physics, allowing researchers to create novel molecules with remarkable precision and functionality. Antibodies are proteins that recognize and neutralize foreign substances like bacteria and viruses, protecting us from infections. On the other hand, enzymes, which are also proteins, speed up countless chemical reactions in our bodies, allowing us to digest food, synthesize DNA, and produce energy.

Next, the sequence from the target TM protein is “threaded” onto either one of the helices—in this example, the right helix with green side-chains (middle). Last, the sequence and side-chain configurations for the anti-TM peptide (represented by the left helix) is chosen via iteration over amino acids and rotamer re-packing (right). From Yin, H.; Slusky, J. S.; Berger, B. W.; Walters, R. S.; Vilaire, G; Litvinov, R.I.; Lear, J. D.; Caputo, G. A.; Bennett, J. S.; DeGrado, W. F. Science 2007, 315(5820), 1817–1822. Naturally occurring proteins with the same fold topology can have distinct functions because of fine-tuned differences in the precise geometries of structural elements (74, 75).

Thus, by supplying radioactively labeled ATP derivatives to cells expressing the mutant kinase, only substrates of that particular kinase would be labeled (Figure 17). The key is to engineer stimuli-responsiveness in a regime that is compatible with physiological temperature. Building on the pioneering work of Ho and DeGrado (J Am Chem Soc 1987, 109, 6751–6758) in the late 1980s, protein design approaches have revealed many fundamental features of protein structure and stability. We are now in the era that the early work presaged – the design of new proteins with practical applications and uses. Here we briefly survey some past milestones in protein design, in addition to highlighting recent progress and future aspirations.

Short fragments with desired secondary structures are then extracted from the PDB and assembled into a three-dimensional protein model (Fig. 2A). Top7 was the first protein designed by this method and has a fold topology not observed in nature (33). We reasoned that RFdiffusion might be able to address this challenge by directly generating binding proteins in the context of the target. For many therapeutic applications, for example, blocking a protein–protein interaction, it is desirable to bind to a particular site on a target protein. To enable this, we fine-tuned RFdiffusion on protein complex structures, providing a feature as input indicating a subset of the residues on the target chain (called ‘interface hotspots’) to which the diffused chain binds (Fig. 6a and Extended Data Fig. 8a,b).

Protein structures can be broken up into three-dimensional local pieces called tertiary structural motifs (TERMs) (133) (Fig. 4C). Half of the structures in the PDB can be described by only about 600 TERMs (37), indicating that the sequence preferences of each TERM could be used to calculate the fitness of a sequence for a given local structure. A strong correlation (133) was observed between the TERM-derived scores and protein structure model accuracies from the Critical Assessment of Structure Prediction.