Optimizing antibody affinity and stability by the automated design of the variable light-heavy chain interfaces

FlowDesign: Improved design of antibody CDRs through flow matching and better prior distributions

Designing antibodies with desired binding specificity and affinity is essential for pharmaceutical research. While diffusion-based models have advanced the co-design of the complementarity-determining region (CDR) sequences and structures, challenges remain, including non-informative priors, incompatibility with discrete amino acid types, and impractical computational costs in large-scale sampling. To address these, we propose FlowDesign, a sequence-structure co-design approach via flow matching, offering (1) flexible prior selection, (2) direct matching of discrete distributions, and (3) enhanced efficiency for large-scale sampling. By leveraging various priors, data-driven structural models proved the most informative. FlowDesign outperformed baselines in amino acid recovery (AAR), root-mean-square deviation (RMSD), and Rosetta energy. We also applied FlowDesign to design antibodies targeting the HIV-1 receptor CD4. FlowDesign yielded antibodies with improved binding affinity and neutralizing potency compared with the antibody ibalizumab across multiple HIV mutants, validated by biolayer interferometry (BLI) and pseudovirus neutralization. This highlights FlowDesign's potential in antibody and protein design. A record of this paper's transparent peer review process is included in the supplemental information.

Evolutionary paths that link orthogonal pairs of binding proteins

Some protein-binding pairs exhibit extreme specificities that functionally insulate them from homologs. Such pairs evolve mostly by accumulating single-point mutations, and mutants are selected if they exhibit sufficient affinity. Until now, finding a fully functional single-mutation path connecting orthogonal pairs could only be achieved by full enumeration of intermediates and was restricted to pairs that were mutationally close. We present a computational framework for discovering single-mutation paths with low molecular strain and apply it to two orthogonal bacterial endonuclease-immunity pairs separated by 17 interfacial mutations. By including mutations that bridge identities that could not be exchanged by single-nucleotide mutations, we discovered a strain-free 19-mutation path that was fully functional in vivo. The change in binding preference occurred remarkably abruptly, resulting from only one radical mutation in each partner. Furthermore, each of the specificity-switch mutations increased fitness, demonstrating that functional divergence could be driven by positive Darwinian selection.

Revolutionizing oncology: the role of Artificial Intelligence (AI) as an antibody design, and optimization tools

Antibodies play a crucial role in defending the human body against diseases, including life-threatening conditions like cancer. They mediate immune responses against foreign antigens and, in some cases, self-antigens. Over time, antibody-based technologies have evolved from monoclonal antibodies (mAbs) to chimeric antigen receptor T cells (CAR-T cells), significantly impacting biotechnology, diagnostics, and therapeutics. Although these advancements have enhanced therapeutic interventions, the integration of artificial intelligence (AI) is revolutionizing antibody design and optimization. This review explores recent AI advancements, including large language models (LLMs), diffusion models, and generative AI-based applications, which have transformed antibody discovery by accelerating de novo generation, enhancing immune response precision, and optimizing therapeutic efficacy. Through advanced data analysis, AI enables the prediction and design of antibody sequences, 3D structures, complementarity-determining regions (CDRs), paratopes, epitopes, and antigen-antibody interactions. These AI-powered innovations address longstanding challenges in antibody development, significantly improving speed, specificity, and accuracy in therapeutic design. By integrating computational advancements with biomedical applications, AI is driving next-generation cancer therapies, transforming precision medicine, and enhancing patient outcomes.

AntiFold: improved structure-based antibody design using inverse folding

Summary

The design and optimization of antibodies requires an intricate balance across multiple properties. Protein inverse folding models, capable of generating diverse sequences folding into the same structure, are promising tools for maintaining structural integrity during antibody design. Here, we present AntiFold, an antibody-specific inverse folding model, fine-tuned from ESM-IF1 on solved and predicted antibody structures. AntiFold outperforms existing inverse folding tools on sequence recovery across complementarity-determining regions, with designed sequences showing high structural similarity to their solved counterpart. It additionally achieves stronger correlations when predicting antibody-antigen binding affinity in a zero-shot manner. AntiFold assigns low probabilities to mutations that disrupt antigen binding, synergizing with protein language model residue probabilities, and demonstrates promise for guiding antibody optimization while retaining structure-related properties.

Availability and implementation

AntiFold is freely available under the BSD 3-Clause as a web server (https://opig.stats.ox.ac.uk/webapps/antifold/) and pip-installable package (https://github.com/oxpig/AntiFold).

Learning the language of protein-protein interactions

Protein Language Models (PLMs) trained on large databases of protein sequences have proven effective in modeling protein biology across a wide range of applications. However, while PLMs excel at capturing individual protein properties, they face challenges in natively representing protein-protein interactions (PPIs), which are crucial to understanding cellular processes and disease mechanisms. Here, we introduce MINT, a PLM specifically designed to model sets of interacting proteins in a contextual and scalable manner. Using unsupervised training on a large curated PPI dataset derived from the STRING database, MINT outperforms existing PLMs in diverse tasks relating to protein-protein interactions, including binding affinity prediction and estimation of mutational effects. Beyond these core capabilities, it excels at modeling interactions in complex protein assemblies and surpasses specialized models in antibody-antigen modeling and T cell receptor-epitope binding prediction. MINT's predictions of mutational impacts on oncogenic PPIs align with experimental studies, and it provides reliable estimates for the potential for cross-neutralization of antibodies against SARS-CoV-2 variants of concern. These findings position MINT as a powerful tool for elucidating complex protein interactions, with significant implications for biomedical research and therapeutic discovery.

htFuncLib: Designing Libraries of Active-site Multipoint Mutants for Protein Optimization

Protein function relies on accurate and densely packed constellations of amino acids within the active site. The high density in the active site optimizes activity but reduces tolerance to mutations, thereby frustrating efforts to engineer or design new or dramatically improved activity. Introducing new activities may therefore require simultaneous multipoint mutations. Still, in a phenomenon known as epistasis, the outcome of combinations of mutations can differ significantly-and even reverse-the impact of the individual mutations, limiting predictability. To address these challenges we previously developed FuncLib, a method for the computational design of multipoint mutants in active sites. We recently extended FuncLib to enable the design of large combinatorial mutation libraries for high-throughput screening in a method called htFuncLib that generates compatible sets of mutations likely to yield functional multipoint mutants. htFuncLib enables scalable library design and experimental screening of hundreds and up to millions of active-site variants. This approach has generated thousands of active enzymes and fluorescent proteins with diverse functional properties. We have updated the FuncLib web server (https://FuncLib.weizmann.ac.il/) to support htFuncLib and introduced an electronic notebook (https://github.com/Fleishman-Lab/htFuncLib-web-server) for customizable library design, making those tools easily accessible for protein engineering and design. The new FuncLib web server enables reliable and scalable design of function for low-, medium- and high-throughput experiments through a single computational platform. We envision that this server will accelerate the optimization and discovery of function in enzymes, antibodies, and other proteins.

Underappreciated layers of antibody-mediated immune synapse architecture and dynamics

The biologic activities of antibody drugs are dictated by structure-function relationships-emerging from the kind, composition, and degree of interactions with a target antigen and with soluble and cellular antibody receptors of the innate immune system. These activities are canonically understood to be both modular: antigen recognition is driven by the heterodimeric antigen-binding fragment, and innate immune recruitment by the homodimeric constant/crystallizable fragment. The model that treats these domains with a high degree of independence has served the field well but is not without limitations. Here, we consider how new insights, particularly from structural studies, complicate the model of neat biophysical separation between these domains and shape our understanding of antibody effector functions. The emerging model endeavors to explain the phenotypic impact of both antibody intrinsic characteristics and extrinsic features-fitting them within a spatiotemporal paradigm that better accounts for observed antibody activities. In this review, we will use insights from recent models of classical complement complexes and T cell immune synapse formation to explore how structural differences in antibody-mediated immune synapses may relate to their functional diversity.

Large scale paired antibody language models

Antibodies are proteins produced by the immune system that can identify and neutralise a wide variety of antigens with high specificity and affinity, and constitute the most successful class of biotherapeutics. With the advent of next-generation sequencing, billions of antibody sequences have been collected in recent years, though their application in the design of better therapeutics has been constrained by the sheer volume and complexity of the data. To address this challenge, we present IgBert and IgT5, the best performing antibody-specific language models developed to date which can consistently handle both paired and unpaired variable region sequences as input. These models are trained comprehensively using the more than two billion unpaired sequences and two million paired sequences of light and heavy chains present in the Observed Antibody Space dataset. We show that our models outperform existing antibody and protein language models on a diverse range of design and regression tasks relevant to antibody engineering. This advancement marks a significant leap forward in leveraging machine learning, large scale data sets and high-performance computing for enhancing antibody design for therapeutic development.

AbMelt: Learning antibody thermostability from molecular dynamics

Antibody thermostability is challenging to predict from sequence and/or structure. This difficulty is likely due to the absence of direct entropic information. Herein, we present AbMelt where we model the inherent flexibility of homologous antibody structures using molecular dynamics simulations at three temperatures and learn the relevant descriptors to predict the temperatures of aggregation (Tagg), melt onset (Tm,on), and melt (Tm). We observed that the radius of gyration deviation of the complementarity determining regions at 400 K is the highest Pearson correlated descriptor with aggregation temperature (rp = -0.68 ± 0.23) and the deviation of internal molecular contacts at 350 K is the highest correlated descriptor with both Tm,on (rp = -0.74 ± 0.04) as well as Tm (rp = -0.69 ± 0.03). Moreover, after descriptor selection and machine learning regression, we predict on a held-out test set containing both internal and public data and achieve robust performance for all endpoints compared with baseline models (Tagg R2 = 0.57 ± 0.11, Tm,on R2 = 0.56 ± 0.01, and Tm R2 = 0.60 ± 0.06). In addition, the robustness of the AbMelt molecular dynamics methodology is demonstrated by only training on <5% of the data and outperforming more traditional machine learning models trained on the entire data set of more than 500 internal antibodies. Users can predict thermostability measurements for antibody variable fragments by collecting descriptors and using AbMelt, which has been made available.

Addressing epistasis in the design of protein function

Mutations in protein active sites can dramatically improve function. The active site, however, is densely packed and extremely sensitive to mutations. Therefore, some mutations may only be tolerated in combination with others in a phenomenon known as epistasis. Epistasis reduces the likelihood of obtaining improved functional variants and dramatically slows natural and lab evolutionary processes. Research has shed light on the molecular origins of epistasis and its role in shaping evolutionary trajectories and outcomes. In addition, sequence- and AI-based strategies that infer epistatic relationships from mutational patterns in natural or experimental evolution data have been used to design functional protein variants. In recent years, combinations of such approaches and atomistic design calculations have successfully predicted highly functional combinatorial mutations in active sites. These were used to design thousands of functional active-site variants, demonstrating that, while our understanding of epistasis remains incomplete, some of the determinants that are critical for accurate design are now sufficiently understood. We conclude that the space of active-site variants that has been explored by evolution may be expanded dramatically to enhance natural activities or discover new ones. Furthermore, design opens the way to systematically exploring sequence and structure space and mutational impacts on function, deepening our understanding and control over protein activity.

Robust Prediction of Relative Binding Energies for Protein-Protein Complex Mutations Using Free Energy Perturbation Calculations

Computational free energy-based methods have the potential to significantly improve throughput and decrease costs of protein design efforts. Such methods must reach a high level of reliability, accuracy, and automation to be effectively deployed in practical industrial settings in a way that impacts protein design projects. Here, we present a benchmark study for the calculation of relative changes in protein-protein binding affinity for single point mutations across a variety of systems from the literature, using free energy perturbation (FEP+) calculations. We describe a method for robust treatment of alternate protonation states for titratable amino acids, which yields improved correlation with and reduced error compared to experimental binding free energies. Following careful analysis of the largest outlier cases in our dataset, we assess limitations of the default FEP+ protocols and introduce an automated script which identifies probable outlier cases that may require additional scrutiny and calculates an empirical correction for a subset of charge-related outliers. Through a series of three additional case study systems, we discuss how Protein FEP+ can be applied to real-world protein design projects, and suggest areas of further study.

An in silico investigation on the binding site preference of PD-1 and PD-L1 for designing antibodies for targeted cancer therapy

Cancer control and treatment remain a significant challenge in cancer therapy and recently immune checkpoints has considered as a novel treatment strategy to develop anti-cancer drugs. Many cancer types use the immune checkpoints and its ligand, PD-1/PD-L1 pathway, to evade detection and destruction by the immune system, which is associated with altered effector function of PD-1 and PD-L1 overexpression on cancer cells to deactivate T cells. In recent years, mAbs have been employed to block immune checkpoints, therefore normalization of the anti-tumor response has enabled the scientists to develop novel biopharmaceuticals. In vivo affinity maturation of antibodies in targeted therapy has sometimes failed, and current experimental methods cannot accommodate the accurate structural details of protein-protein interactions. Therefore, determining favorable binding sites on the protein surface for modulator design of these interactions is a major challenge. In this study, we used the in silico methods to identify favorable binding sites on the PD-1 and PD-L1 and to optimize mAb variants on a large scale. At first, all the binding areas on PD-1 and PD-L1 have been identified. Then, using the RosettaDesign protocol, thousands of antibodies have been generated for 11 different regions on PD-1 and PD-L1 and then the designs with higher stability, affinity, and shape complementarity were selected. Next, molecular dynamics simulations and MM-PBSA analysis were employed to understand the dynamic, structural features of the complexes and measure the binding affinity of the final designs. Our results suggest that binding sites 1, 3 and 6 on PD-1 and binding sites 9 and 11 on PD-L1 can be regarded as the most appropriate sites for the inhibition of PD-1-PD-L1 interaction by the designed antibodies. This study provides comprehensive information regarding the potential binding epitopes on PD-1 which could be considered as hotspots for designing potential biopharmaceuticals. We also showed that mutations in the CDRs regions will rearrange the interaction pattern between the designed antibodies and targets (PD-1 and PD-L1) with improved affinity to effectively inhibit protein-protein interaction and block the immune checkpoint.

Robust prediction of relative binding energies for protein-protein complex mutations using free energy perturbation calculations

Computational free energy-based methods have the potential to significantly improve throughput and decrease costs of protein design efforts. Such methods must reach a high level of reliability, accuracy, and automation to be effectively deployed in practical industrial settings in a way that impacts protein design projects. Here, we present a benchmark study for the calculation of relative changes in protein-protein binding affinity for single point mutations across a variety of systems from the literature, using free energy perturbation (FEP+) calculations. We describe a method for robust treatment of alternate protonation states for titratable amino acids, which yields improved correlation with and reduced error compared to experimental binding free energies. Following careful analysis of the largest outlier cases in our dataset, we assess limitations of the default FEP+ protocols and introduce an automated script which identifies probable outlier cases that may require additional scrutiny and calculates an empirical correction for a subset of charge-related outliers. Through a series of three additional case study systems, we discuss how protein FEP+ can be applied to real-world protein design projects, and suggest areas of further study.

Generation of novel anti-apoE monoclonal antibodies that selectively recognize apoE isoforms

Apolipoprotein E (apoE) is a regulator of lipid metabolism, cholesterol transport, and the clearance and aggregation of amyloid β in the brain. The three human apoE isoforms apoE2, apoE3, and apoE4 only differ in one or two residues. Nevertheless, the functions highly depend on the isoform types and lipidated states. Here, we generated novel anti-apoE monoclonal antibodies (mAbs) and obtained an apoE4-selective mAb whose epitope is within residues 110-117. ELISA and bio-layer interferometry measurements demonstrated that the dissociation constants of mAbs are within the nanomolar range. Using the generated antibodies, we successfully constructed sandwich ELISA systems, which can detect all apoE isoforms or selectively detect apoE4. These results suggest the usability of the generated anti-apoE mAbs for selective detection of apoE isoforms.

Computational optimization of antibody humanness and stability by systematic energy-based ranking

Conventional methods for humanizing animal-derived antibodies involve grafting their complementarity-determining regions onto homologous human framework regions. However, this process can substantially lower antibody stability and antigen-binding affinity, and requires iterative mutational fine-tuning to recover the original antibody properties. Here we report a computational method for the systematic grafting of animal complementarity-determining regions onto thousands of human frameworks. The method, which we named CUMAb (for computational human antibody design; available at http://CUMAb.weizmann.ac.il ), starts from an experimental or model antibody structure and uses Rosetta atomistic simulations to select designs by energy and structural integrity. CUMAb-designed humanized versions of five antibodies exhibited similar affinities to those of the parental animal antibodies, with some designs showing marked improvement in stability. We also show that (1) non-homologous frameworks are often preferred to highest-homology frameworks, and (2) several CUMAb designs that differ by dozens of mutations and that use different human frameworks are functionally equivalent.

Toward enhancement of antibody thermostability and affinity by computational design in the absence of antigen

Over the past two decades, therapeutic antibodies have emerged as a rapidly expanding domain within the field of biologics. In silico tools that can streamline the process of antibody discovery and optimization are critical to support a pipeline that is growing more numerous and complex every year. High-quality structural information remains critical for the antibody optimization process, but antibody-antigen complex structures are often unavailable and in silico antibody docking methods are still unreliable. In this study, DeepAb, a deep learning model for predicting antibody Fv structure directly from sequence, was used in conjunction with single-point experimental deep mutational scanning (DMS) enrichment data to design 200 potentially optimized variants of an anti-hen egg lysozyme (HEL) antibody. We sought to determine whether DeepAb-designed variants containing combinations of beneficial mutations from the DMS exhibit enhanced thermostability and whether this optimization affected their developability profile. The 200 variants were produced through a robust high-throughput method and tested for thermal and colloidal stability (Tonset, Tm, Tagg), affinity (KD) relative to the parental antibody, and for developability parameters (nonspecific binding, aggregation propensity, self-association). Of the designed clones, 91% and 94% exhibited increased thermal and colloidal stability and affinity, respectively. Of these, 10% showed a significantly increased affinity for HEL (5- to 21-fold increase) and thermostability (>2.5C increase in Tm1), with most clones retaining the favorable developability profile of the parental antibody. Additional in silico tests suggest that these methods would enrich for binding affinity even without first collecting experimental DMS measurements. These data open the possibility of in silico antibody optimization without the need to predict the antibody-antigen interface, which is notoriously difficult in the absence of crystal structures.

Matrixed CDR grafting: A neoclassical framework for antibody humanization and developability

Discovery and optimization of a biotherapeutic monoclonal antibody requires a careful balance of target engagement and physicochemical developability properties. To take full advantage of the sequence diversity provided by different antibody discovery platforms, a rapid and reliable process for humanization of antibodies from nonhuman sources is required. Canonically, maximizing homology of the human variable region (V-region) to the original germline was believed to result in preservation of binding, often without much consideration for inherent molecular properties. We expand on this approach by grafting the complementary determining regions (CDRs) of a mouse anti-LAG3 antibody into an extensive matrix of human variable heavy chain (VH) and variable light chain (VL) framework regions with substantially broader sequence homology to assess the impact on complementary determining region-framework compatibility through progressive evaluation of expression, affinity, biophysical developability, and function. Specific VH and VL framework sequences were associated with major expression and purification phenotypes. Greater VL sequence conservation was correlated with retained or improved affinity. Analysis of grafts that bound the target demonstrated that initial developability criteria were significantly impacted by VH, but not VL. In contrast, cell binding and functional characteristics were significantly impacted by VL, but not VH. Principal component analysis of all factors identified multiple grafts that exhibited more favorable antibody properties, notably with nonoptimal sequence conservation. Overall, this study demonstrates that modern throughput systems enable a more thorough, customizable, and systematic analysis of graft-framework combinations, resulting in humanized antibodies with improved global properties that may progress through development more quickly and with a greater probability of success.

ProGen2: Exploring the boundaries of protein language models

Attention-based models trained on protein sequences have demonstrated incredible success at classification and generation tasks relevant for artificial-intelligence-driven protein design. However, we lack a sufficient understanding of how very large-scale models and data play a role in effective protein model development. We introduce a suite of protein language models, named ProGen2, that are scaled up to 6.4B parameters and trained on different sequence datasets drawn from over a billion proteins from genomic, metagenomic, and immune repertoire databases. ProGen2 models show state-of-the-art performance in capturing the distribution of observed evolutionary sequences, generating novel viable sequences, and predicting protein fitness without additional fine-tuning. As large model sizes and raw numbers of protein sequences continue to become more widely accessible, our results suggest that a growing emphasis needs to be placed on the data distribution provided to a protein sequence model. Our models and code are open sourced for widespread adoption in protein engineering. A record of this paper's Transparent Peer Review process is included in the supplemental information.

Position-Specific Enrichment Ratio Matrix scores predict antibody variant properties from deep sequencing data

MOTIVATION

Deep sequencing of antibody and related protein libraries after phage or yeast-surface display sorting is widely used to identify variants with increased affinity, specificity, and/or improvements in key biophysical properties. Conventional approaches for identifying optimal variants typically use the frequencies of observation in enriched libraries or the corresponding enrichment ratios. However, these approaches disregard the vast majority of deep sequencing data and often fail to identify the best variants in the libraries.

RESULTS

Here, we present a method, Position-Specific Enrichment Ratio Matrix (PSERM) scoring, that uses entire deep sequencing datasets from pre- and post-selections to score each observed protein variant. The PSERM scores are the sum of the site-specific enrichment ratios observed at each mutated position. We find that PSERM scores are much more reproducible and correlate more strongly with experimentally measured properties than frequencies or enrichment ratios, including for multiple antibody properties (affinity and non-specific binding) for a clinical-stage antibody (emibetuzumab). We expect that this method will be broadly applicable to diverse protein engineering campaigns.

AVAILABILITY AND IMPLEMENTATION

All deep sequencing datasets and code to perform the analyses presented within are available via https://github.com/Tessier-Lab-UMich/PSERM_paper.

Affinity maturation of antibody fragments: A review encompassing the development from random approaches to computational rational optimization

Routinely screened antibody fragments usually require further in vitro maturation to achieve the desired biophysical properties. Blind in vitro strategies can produce improved ligands by introducing random mutations into the original sequences and selecting the resulting clones under more and more stringent conditions. Rational approaches exploit an alternative perspective that aims first at identifying the specific residues potentially involved in the control of biophysical mechanisms, such as affinity or stability, and then to evaluate what mutations could improve those characteristics. The understanding of the antigen-antibody interactions is instrumental to develop this process the reliability of which, consequently, strongly depends on the quality and completeness of the structural information. Recently, methods based on deep learning approaches critically improved the speed and accuracy of model building and are promising tools for accelerating the docking step. Here, we review the features of the available bioinformatic instruments and analyze the reports illustrating the result obtained with their application to optimize antibody fragments, and nanobodies in particular. Finally, the emerging trends and open questions are summarized.

Exploring rigid-backbone protein docking in biologics discovery: a test using the DARPin scaffold

Accurate protein-protein docking remains challenging, especially for artificial biologics not coevolved naturally against their protein targets, like antibodies and other engineered scaffolds. We previously developed ProPOSE, an exhaustive docker with full atomistic details, which delivers cutting-edge performance by allowing side-chain rearrangements upon docking. However, extensive protein backbone flexibility limits its practical applicability as indicated by unbound docking tests. To explore the usefulness of ProPOSE on systems with limited backbone flexibility, here we tested the engineered scaffold DARPin, which is characterized by its relatively rigid protein backbone. A prospective screening campaign was undertaken, in which sequence-diversified DARPins were docked and ranked against a directed epitope on the target protein BCL-W. In this proof-of-concept study, only a relatively small set of 2,213 diverse DARPin interfaces were selected for docking from the huge theoretical library from mutating 18 amino-acid positions. A computational selection protocol was then applied for enrichment of binders based on normalized computed binding scores and frequency of binding modes against the predefined epitope. The top-ranked 18 designed DARPin interfaces were selected for experimental validation. Three designs exhibited binding affinities to BCL-W in the nanomolar range comparable to control interfaces adopted from known DARPin binders. This result is encouraging for future screening and engineering campaigns of DARPins and possibly other similarly rigid scaffolds against targeted protein epitopes. Method limitations are discussed and directions for future refinements are proposed.

Investigating Uncertainties in Single-Molecule Localization Microscopy Using Experimentally Informed Monte Carlo Simulation

Single-molecule localization microscopy (SMLM) enables the visualization of cellular nanostructures in vitro with sub-20 nm resolution. While substructures can generally be imaged with SMLM, the structural understanding of the images remains elusive. To better understand the link between SMLM images and the underlying structure, we developed a Monte Carlo (MC) simulation based on experimental imaging parameters and geometric information to generate synthetic SMLM images. We chose the nuclear pore complex (NPC), a nanosized channel on the nuclear membrane which gates nucleo-cytoplasmic transport of biomolecules, as a test geometry for testing our MC model. Using the MC model to simulate SMLM images, we first optimized our clustering algorithm to separate >106 molecular localizations of fluorescently labeled NPC proteins into hundreds of individual NPCs in each cell. We then illustrated using our MC model to generate cellular substructures with different angles of labeling to inform our structural understanding through the SMLM images obtained.

Computational Methods in Immunology and Vaccinology: Design and Development of Antibodies and Immunogens

The design of new biomolecules able to harness immune mechanisms for the treatment of diseases is a prime challenge for computational and simulative approaches. For instance, in recent years, antibodies have emerged as an important class of therapeutics against a spectrum of pathologies. In cancer, immune-inspired approaches are witnessing a surge thanks to a better understanding of tumor-associated antigens and the mechanisms of their engagement or evasion from the human immune system. Here, we provide a summary of the main state-of-the-art computational approaches that are used to design antibodies and antigens, and in parallel, we review key methodologies for epitope identification for both B- and T-cell mediated responses. A special focus is devoted to the description of structure- and physics-based models, privileged over purely sequence-based approaches. We discuss the implications of novel methods in engineering biomolecules with tailored immunological properties for possible therapeutic uses. Finally, we highlight the extraordinary challenges and opportunities presented by the possible integration of structure- and physics-based methods with emerging Artificial Intelligence technologies for the prediction and design of novel antigens, epitopes, and antibodies.

Defining and Studying B Cell Receptor and TCR Interactions

BCRs (Abs) and TCRs (or adaptive immune receptors [AIRs]) are the means by which the adaptive immune system recognizes foreign and self-antigens, playing an integral part in host defense, as well as the emergence of autoimmunity. Importantly, the interaction between AIRs and their cognate Ags defies a simple key-in-lock paradigm and is instead a complex many-to-many mapping between an individual's massively diverse AIR repertoire, and a similarly diverse antigenic space. Understanding how adaptive immunity balances specificity with epitopic coverage is a key challenge for the field, and terms such as broad specificity, cross-reactivity, and polyreactivity remain ill-defined and are used inconsistently. In this Immunology Notes and Resources article, a group of experimental, structural, and computational immunologists define commonly used terms associated with AIR binding, describe methodologies to study these binding modes, as well as highlight the implications of these different binding modes for therapeutic design.

KA-Search, a method for rapid and exhaustive sequence identity search of known antibodies

Antibodies with similar amino acid sequences, especially across their complementarity-determining regions, often share properties. Finding that an antibody of interest has a similar sequence to naturally expressed antibodies in healthy or diseased repertoires is a powerful approach for the prediction of antibody properties, such as immunogenicity or antigen specificity. However, as the number of available antibody sequences is now in the billions and continuing to grow, repertoire mining for similar sequences has become increasingly computationally expensive. Existing approaches are limited by either being low-throughput, non-exhaustive, not antibody specific, or only searching against entire chain sequences. Therefore, there is a need for a specialized tool, optimized for a rapid and exhaustive search of any antibody region against all known antibodies, to better utilize the full breadth of available repertoire sequences. We introduce Known Antibody Search (KA-Search), a tool that allows for the rapid search of billions of antibody variable domains by amino acid sequence identity across either the variable domain, the complementarity-determining regions, or a user defined antibody region. We show KA-Search in operation on the [Formula: see text]2.4 billion antibody sequences available in the OAS database. KA-Search can be used to find the most similar sequences from OAS within 30 minutes and a representative subset of 10 million sequences in less than 9 seconds. We give examples of how KA-Search can be used to obtain new insights about an antibody of interest. KA-Search is freely available at https://github.com/oxpig/kasearch .

Position-Specific Enrichment Ratio Matrix scores predict antibody variant properties from deep sequencing data

Motivation

Deep sequencing of antibody and related protein libraries after phage or yeast-surface display sorting is widely used to identify variants with increased affinity, specificity and/or improvements in key biophysical properties. Conventional approaches for identifying optimal variants typically use the frequencies of observation in enriched libraries or the corresponding enrichment ratios. However, these approaches disregard the vast majority of deep sequencing data and often fail to identify the best variants in the libraries.

Results

Here, we present a method, Position-Specific Enrichment Ratio Matrix (PSERM) scoring, that uses entire deep sequencing datasets from pre- and post-selections to score each observed protein variant. The PSERM scores are the sum of the site-specific enrichment ratios observed at each mutated position. We find that PSERM scores are much more reproducible and correlate more strongly with experimentally measured properties than frequencies or enrichment ratios, including for multiple antibody properties (affinity and non-specific binding) for a clinical-stage antibody (emibetuzumab). We expect that this method will be broadly applicable to diverse protein engineering campaigns.

Availability

All deep sequencing datasets and code to do the analyses presented within are available via GitHub.

Contact

Peter Tessier, ptessier@umich.edu.

Supplementary information

Supplementary data are available at Bioinformatics online.

Engineering of conserved residues near antibody heavy chain complementary determining region 3 (HCDR3) improves both affinity and stability

Affinity and stability are crucial parameters in antibody development and engineering approaches. Although improvement in both metrics is desirable, trade-offs are almost unavoidable. Heavy chain complementarity determining region 3 (HCDR3) is the best-known region for antibody affinity but its impact on stability is often neglected. Here, we present a mutagenesis study of conserved residues near HCDR3 to elicit the role of this region in the affinity-stability trade-off. These key residues are positioned around the conserved salt bridge between VH-K94 and VH-D101 which is crucial for HCDR3 integrity. We show that the additional salt bridge at the stem of HCDR3 (VH-K94:VH-D101:VH-D102) has an extensive impact on this loop's conformation, therefore simultaneous improvement in both affinity and stability. We find that the disruption of π-π stacking near HCDR3 (VH-Y100E:VL-Y49) at the VH-VL interface cause an irrecoverable loss in stability even if it improves the affinity. Molecular simulations of putative rescue mutants exhibit complex and often non-additive effects. We confirm that our experimental measurements agree with the molecular dynamic simulations providing detailed insights for the spatial orientation of HCDR3. VH-V102 right next to HCDR3 salt bridge might be an ideal candidate to overcome affinity-stability trade-off.

Allosteric regulation of the 20S proteasome by the Catalytic Core Regulators (CCRs) family

Controlled degradation of proteins is necessary for ensuring their abundance and sustaining a healthy and accurately functioning proteome. One of the degradation routes involves the uncapped 20S proteasome, which cleaves proteins with a partially unfolded region, including those that are damaged or contain intrinsically disordered regions. This degradation route is tightly controlled by a recently discovered family of proteins named Catalytic Core Regulators (CCRs). Here, we show that CCRs function through an allosteric mechanism, coupling the physical binding of the PSMB4 β-subunit with attenuation of the complex's three proteolytic activities. In addition, by dissecting the structural properties that are required for CCR-like function, we could recapitulate this activity using a designed protein that is half the size of natural CCRs. These data uncover an allosteric path that does not involve the proteasome's enzymatic subunits but rather propagates through the non-catalytic subunit PSMB4. This way of 20S proteasome-specific attenuation opens avenues for decoupling the 20S and 26S proteasome degradation pathways as well as for developing selective 20S proteasome inhibitors.

An optimized thermodynamics integration protocol for identifying beneficial mutations in antibody design

Accurate identification of beneficial mutations is central to antibody design. Many knowledge-based (KB) computational approaches have been developed to predict beneficial mutations, but their accuracy leaves room for improvement. Thermodynamic integration (TI) is an alchemical free energy algorithm that offers an alternative technique for identifying beneficial mutations, but its performance has not been evaluated. In this study, we developed an efficient TI protocol with high accuracy for predicting binding free energy changes of antibody mutations. The improved TI method outperforms KB methods at identifying both beneficial and deleterious mutations. We observed that KB methods have higher accuracies in predicting deleterious mutations than beneficial mutations. A pipeline using KB methods to efficiently exclude deleterious mutations and TI to accurately identify beneficial mutations was developed for high-throughput mutation scanning. The pipeline was applied to optimize the binding affinity of a broadly sarbecovirus neutralizing antibody 10-40 against the circulating severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) omicron variant. Three identified beneficial mutations show strong synergy and improve both binding affinity and neutralization potency of antibody 10-40. Molecular dynamics simulation revealed that the three mutations improve the binding affinity of antibody 10-40 through the stabilization of an altered binding mode with increased polar and hydrophobic interactions. Above all, this study presents an accurate and efficient TI-based approach for optimizing antibodies and other biomolecules.

Structural mechanism of Fab domain dissociation as a measure of interface stability

Therapeutic antibodies should not only recognize antigens specifically, but also need to be free from developability issues, such as poor stability. Thus, the mechanistic understanding and characterization of stability are critical determinants for rational antibody design. In this study, we use molecular dynamics simulations to investigate the melting process of 16 antigen binding fragments (Fabs). We describe the Fab dissociation mechanisms, showing a separation in the VH-VL and in the CH1-CL domains. We found that the depths of the minima in the free energy curve, corresponding to the bound states, correlate with the experimentally determined melting temperatures. Additionally, we provide a detailed structural description of the dissociation mechanism and identify key interactions in the CDR loops and in the CH1-CL interface that contribute to stabilization. The dissociation of the VH-VL or CH1-CL domains can be represented by conformational changes in the bend angles between the domains. Our findings elucidate the melting process of antigen binding fragments and highlight critical residues in both the variable and constant domains, which are also strongly germline dependent. Thus, our proposed mechanisms have broad implications in the development and design of new and more stable antigen binding fragments.

Engineering the kinetic stability of a β-trefoil protein by tuning its topological complexity

Kinetic stability, defined as the rate of protein unfolding, is central to determining the functional lifetime of proteins, both in nature and in wide-ranging medical and biotechnological applications. Further, high kinetic stability is generally correlated with high resistance against chemical and thermal denaturation, as well as proteolytic degradation. Despite its significance, specific mechanisms governing kinetic stability remain largely unknown, and few studies address the rational design of kinetic stability. Here, we describe a method for designing protein kinetic stability that uses protein long-range order, absolute contact order, and simulated free energy barriers of unfolding to quantitatively analyze and predict unfolding kinetics. We analyze two β-trefoil proteins: hisactophilin, a quasi-three-fold symmetric natural protein with moderate stability, and ThreeFoil, a designed three-fold symmetric protein with extremely high kinetic stability. The quantitative analysis identifies marked differences in long-range interactions across the protein hydrophobic cores that partially account for the differences in kinetic stability. Swapping the core interactions of ThreeFoil into hisactophilin increases kinetic stability with close agreement between predicted and experimentally measured unfolding rates. These results demonstrate the predictive power of readily applied measures of protein topology for altering kinetic stability and recommend core engineering as a tractable target for rationally designing kinetic stability that may be widely applicable.

Homology Modeling of Antibody Variable Regions: Methods and Applications

Adaptive immunity specifically protects us from antigenic challenges. Antibodies are key effector proteins of adaptive immunity, and they are remarkable in their ability to recognize a virtually limitless number of antigens. Fragment variable (F_V), the antigen-binding region of antibodies, can be split into two main components, namely, framework and complementarity determining regions. The framework (FR) consists of light-chain framework (FR_L) and heavy-chain framework (FR_H). Similarly, the complementarity determining regions (CDRs) comprises of light-chain CDRs 1-3 (CDRs L1-3) and heavy-chain CDRs 1-3 (CDRs H1-3). While FRs are relatively constant in sequence and structure across diverse antibodies, sequence variation in CDRs leading to differential conformations of CDR loops accounts for the distinct antigenic specificities of diverse antibodies. The conserved structural features in FRs and conformity of CDRs to a limited set of standard conformations allow for the accurate prediction of F_V models using homology modeling techniques. Antibody structure prediction from its amino acid sequence has numerous important applications including prediction of antibody-antigen interaction interfaces and redesign of therapeutically and biotechnologically useful antibodies with improved affinity. This chapter summarizes the current practices employed in the successful homology modeling of antibody variable regions and the potential applications of the generated homology models.

Single-Cell B-Cell Sequencing to Generate Natively Paired scFab Yeast Surface Display Libraries

The immune cell profiling capabilities of single-cell RNA sequencing (scRNA-seq) are powerful tools that can be applied to the design of theranostic monoclonal antibodies (mAbs). Using scRNA-seq to determine natively paired B-cell receptor (BCR) sequences of immunized mice as a starting point for design, this method outlines a simplified workflow to express single-chain antibody fragments (scFabs) on the surface of yeast for high-throughput characterization and further refinement with directed evolution experiments. While not extensively detailed in this chapter, this method easily accommodates the implementation of a growing body of in silico tools that improve affinity and stability among a range of other developability criteria (e.g., solubility and immunogenicity).

Toward generalizable prediction of antibody thermostability using machine learning on sequence and structure features

Over the last three decades, the appeal for monoclonal antibodies (mAbs) as therapeutics has been steadily increasing as evident with FDA's recent landmark approval of the 100th mAb. Unlike mAbs that bind to single targets, multispecific biologics (msAbs) have garnered particular interest owing to the advantage of engaging distinct targets. One important modular component of msAbs is the single-chain variable fragment (scFv). Despite the exquisite specificity and affinity of these scFv modules, their relatively poor thermostability often hampers their development as a potential therapeutic drug. In recent years, engineering antibody sequences to enhance their stability by mutations has gained considerable momentum. As experimental methods for antibody engineering are time-intensive, laborious and expensive, computational methods serve as a fast and inexpensive alternative to conventional routes. In this work, we show two machine learning approaches - one with pre-trained language models (PTLM) capturing functional effects of sequence variation, and second, a supervised convolutional neural network (CNN) trained with Rosetta energetic features - to better classify thermostable scFv variants from sequence. Both of these models are trained over temperature-specific data (TS50 measurements) derived from multiple libraries of scFv sequences. On out-of-distribution (refers to the fact that the out-of-distribution sequnes are blind to the algorithm) sequences, we show that a sufficiently simple CNN model performs better than general pre-trained language models trained on diverse protein sequences (average Spearman correlation coefficient, ρ , of 0.4 as opposed to 0.15). On the other hand, an antibody-specific language model performs comparatively better than the CNN model on the same task (ρ = 0.52). Further, we demonstrate that for an independent mAb with available thermal melting temperatures for 20 experimentally characterized thermostable mutations, these models trained on TS50 data could identify 18 residue positions and 5 identical amino-acid mutations showing remarkable generalizability. Our results suggest that such models can be broadly applicable for improving the biological characteristics of antibodies. Further, transferring such models for alternative physicochemical properties of scFvs can have potential applications in optimizing large-scale production and delivery of mAbs or bsAbs.

Yeast Display Enables Identification of Covalent Single-Domain Antibodies against Botulinum Neurotoxin Light Chain A

While covalent drug discovery is reemerging as an important route to small-molecule therapeutic leads, strategies for the discovery and engineering of protein-based irreversible binding agents remain limited. Here, we describe the use of yeast display in combination with noncanonical amino acids (ncAAs) to identify irreversible variants of single-domain antibodies (sdAbs), also called VHHs and nanobodies, targeting botulinum neurotoxin light chain A (LC/A). Starting from a series of previously described, structurally characterized sdAbs, we evaluated the properties of antibodies substituted with reactive ncAAs capable of forming covalent bonds with nearby groups after UV irradiation (when using 4-azido-l-phenylalanine) or spontaneously (when using O-(2-bromoethyl)-l-tyrosine). Systematic evaluations in yeast display format of more than 40 ncAA-substituted variants revealed numerous clones that retain binding function while gaining either UV-mediated or spontaneous crosslinking capabilities. Solution-based analyses indicate that ncAA-substituted clones exhibit site-dependent target specificity and crosslinking capabilities uniquely conferred by ncAAs. Interestingly, not all ncAA substitution sites resulted in crosslinking events, and our data showed no apparent correlation between detected crosslinking levels and distances between sdAbs and LC/A residues. Our findings highlight the power of yeast display in combination with genetic code expansion in the discovery of binding agents that covalently engage their targets. This platform streamlines the discovery and characterization of antibodies with therapeutically relevant properties that cannot be accessed in the conventional genetic code.

Prediction of polyreactive and nonspecific single-chain fragment variables through structural biochemical features and protein language-based descriptors

BACKGROUND

Monoclonal antibodies (mAbs) have been used as therapeutic agents, which must overcome many developability issues after the discovery from in vitro display libraries. Especially, polyreactive mAbs can strongly bind to a specific target and weakly bind to off-target proteins, which leads to poor antibody pharmacokinetics in clinical development. Although early assessment of polyreactive mAbs is important in the early discovery stage, experimental assessments are usually time-consuming and expensive. Therefore, computational approaches for predicting the polyreactivity of single-chain fragment variables (scFvs) in the early discovery stage would be promising for reducing experimental efforts.

RESULTS

Here, we made prediction models for the polyreactivity of scFvs with the known polyreactive antibody features and natural language model descriptors. We predicted 19,426 protein structures of scFvs with trRosetta to calculate the polyreactive antibody features and investigated the classifying performance of each factor for polyreactivity. In the known polyreactive features, the net charge of the CDR2 loop, the tryptophan and glycine residues in CDR-H3, and the lengths of the CDR1 and CDR2 loops, importantly contributed to the performance of the models. Additionally, the hydrodynamic features, such as partial specific volume, gyration radius, and isoelectric points of CDR loops and scFvs, were newly added to improve model performance. Finally, we made the prediction model with a robust performance ([Formula: see text]) with an ensemble learning of the top 3 best models.

CONCLUSION

The prediction models for polyreactivity would help assess polyreactive scFvs in the early discovery stage and our approaches would be promising to develop machine learning models with quantitative data from high throughput assays for antibody screening.

Expression of mammalian proteins for diagnostics and therapeutics: a review

BACKGROUND

Antibodies have proven to be remarkably successful for biomedical applications. They play important roles in epidemiology and medicine from diagnostics of diseases to therapeutics, treating diseases from incessant chronic diseases such as rheumatology to pandemic outbreaks. With no end in sight for the demand for antibody products, optimizations and new techniques must be expanded to accommodate this.

METHODS AND RESULTS

This review discusses optimizations and techniques for antibody production through choice of discovery platforms, expression systems, cell culture mediums, and other strategies to increase expression yield. Each system has its own merits and demerits, and the strategy chosen is critical in addressing various biological aspects.

CONCLUSIONS

There is still insufficient evidence to validate the efficacy of some of these techniques, and further research is needed to consolidate these industrial production systems. There is no doubt that more strategies, systems, and pipelines will contribute to enhance biopharmaceutical production.

Hallucinating structure-conditioned antibody libraries for target-specific binders

Antibodies are widely developed and used as therapeutics to treat cancer, infectious disease, and inflammation. During development, initial leads routinely undergo additional engineering to increase their target affinity. Experimental methods for affinity maturation are expensive, laborious, and time-consuming and rarely allow the efficient exploration of the relevant design space. Deep learning (DL) models are transforming the field of protein engineering and design. While several DL-based protein design methods have shown promise, the antibody design problem is distinct, and specialized models for antibody design are desirable. Inspired by hallucination frameworks that leverage accurate structure prediction DL models, we propose the FvHallucinator for designing antibody sequences, especially the CDR loops, conditioned on an antibody structure. Such a strategy generates targeted CDR libraries that retain the conformation of the binder and thereby the mode of binding to the epitope on the antigen. On a benchmark set of 60 antibodies, FvHallucinator generates sequences resembling natural CDRs and recapitulates perplexity of canonical CDR clusters. Furthermore, the FvHallucinator designs amino acid substitutions at the VH-VL interface that are enriched in human antibody repertoires and therapeutic antibodies. We propose a pipeline that screens FvHallucinator designs to obtain a library enriched in binders for an antigen of interest. We apply this pipeline to the CDR H3 of the Trastuzumab-HER2 complex to generate in silico designs predicted to improve upon the binding affinity and interfacial properties of the original antibody. Thus, the FvHallucinator pipeline enables generation of inexpensive, diverse, and targeted antibody libraries enriched in binders for antibody affinity maturation.

Advances in antibody discovery from human BCR repertoires

Antibodies make up an important and growing class of compounds used for the diagnosis or treatment of disease. While traditional antibody discovery utilized immunization of animals to generate lead compounds, technological innovations have made it possible to search for antibodies targeting a given antigen within the repertoires of B cells in humans. Here we group these innovations into four broad categories: cell sorting allows the collection of cells enriched in specificity to one or more antigens; BCR sequencing can be performed on bulk mRNA, genomic DNA or on paired (heavy-light) mRNA; BCR repertoire analysis generally involves clustering BCRs into specificity groups or more in-depth modeling of antibody-antigen interactions, such as antibody-specific epitope predictions; validation of antibody-antigen interactions requires expression of antibodies, followed by antigen binding assays or epitope mapping. Together with innovations in Deep learning these technologies will contribute to the future discovery of diagnostic and therapeutic antibodies directly from humans.

Assessing and enhancing foldability in designed proteins

Recent advances in protein-design methodology have led to a dramatic increase in reliability and scale. With these advances, dozens and even thousands of designed proteins are automatically generated and screened. Nevertheless, the success rate, particularly in design of functional proteins, is low and fundamental goals such as reliable de novo design of efficient enzymes remain beyond reach. Experimental analyses have consistently indicated that a major reason for design failure is inaccuracy and misfolding relative to the design conception. To address this challenge, we describe complementary methods to diagnose and ameliorate suboptimal regions in designed proteins: first, we develop a Rosetta atomistic computational mutation scanning approach to detect energetically suboptimal positions in designs (available on a web server https://pSUFER.weizmann.ac.il); second, we demonstrate that AlphaFold2 ab initio structure prediction flags regions that may misfold in designed enzymes and binders; and third, we focus FuncLib design calculations on suboptimal positions in a previously designed low-efficiency enzyme, improving its catalytic efficiency by 330-fold. Furthermore, applied to a de novo designed protein that exhibited limited stability, the same approach markedly improved stability and expressibility. Thus, foldability analysis and enhancement may dramatically increase the success rate in design of functional proteins.

Co-optimization of therapeutic antibody affinity and specificity using machine learning models that generalize to novel mutational space

Therapeutic antibody development requires selection and engineering of molecules with high affinity and other drug-like biophysical properties. Co-optimization of multiple antibody properties remains a difficult and time-consuming process that impedes drug development. Here we evaluate the use of machine learning to simplify antibody co-optimization for a clinical-stage antibody (emibetuzumab) that displays high levels of both on-target (antigen) and off-target (non-specific) binding. We mutate sites in the antibody complementarity-determining regions, sort the antibody libraries for high and low levels of affinity and non-specific binding, and deep sequence the enriched libraries. Interestingly, machine learning models trained on datasets with binary labels enable predictions of continuous metrics that are strongly correlated with antibody affinity and non-specific binding. These models illustrate strong tradeoffs between these two properties, as increases in affinity along the co-optimal (Pareto) frontier require progressive reductions in specificity. Notably, models trained with deep learning features enable prediction of novel antibody mutations that co-optimize affinity and specificity beyond what is possible for the original antibody library. These findings demonstrate the power of machine learning models to greatly expand the exploration of novel antibody sequence space and accelerate the development of highly potent, drug-like antibodies.

Charged Residue Implantation Improves the Affinity of a Cross-Reactive Dengue Virus Antibody

Dengue virus (DENV) has four serotypes that complicate vaccine development. Envelope protein domain III (EDIII) of DENV is a promising target for therapeutic antibody development. One EDIII-specific antibody, dubbed 1A1D-2, cross-reacts with DENV 1, 2, and 3 but not 4. To improve the affinity of 1A1D-2, in this study, we analyzed the previously solved structure of 1A1D-2-DENV2 EDIII complex. Mutations were designed, including A54E and Y105R in the heavy chain, with charges complementary to the epitope. Molecular dynamics simulation was then used to validate the formation of predicted salt bridges. Interestingly, a surface plasmon resonance experiment showed that both mutations increased affinities of 1A1D-2 toward EDIII of DENV1, 2, and 3 regardless of their sequence variation. Results also revealed that A54E improved affinities through both a faster association and slower dissociation, whereas Y105R improved affinities through a slower dissociation. Further simulation suggested that the same mutants interacted with different residues in different serotypes. Remarkably, combination of the two mutations additively improved 1A1D-2 affinity by 8, 36, and 13-fold toward DENV1, 2, and 3, respectively. In summary, this study demonstrated the utility of tweaking antibody-antigen charge complementarity for affinity maturation and emphasized the complexity of improving antibody affinity toward multiple antigens.

What Have We Learned from Design of Function in Large Proteins?

The overarching goal of computational protein design is to gain complete control over protein structure and function. The majority of sophisticated binders and enzymes, however, are large and exhibit diverse and complex folds that defy atomistic design calculations. Encouragingly, recent strategies that combine evolutionary constraints from natural homologs with atomistic calculations have significantly improved design accuracy. In these approaches, evolutionary constraints mitigate the risk from misfolding and aggregation, focusing atomistic design calculations on a small but highly enriched sequence subspace. Such methods have dramatically optimized diverse proteins, including vaccine immunogens, enzymes for sustainable chemistry, and proteins with therapeutic potential. The new generation of deep learning-based ab initio structure predictors can be combined with these methods to extend the scope of protein design, in principle, to any natural protein of known sequence. We envision that protein engineering will come to rely on completely computational methods to efficiently discover and optimize biomolecular activities.

Affinity maturation of TCR-like antibodies using phage display guided by structural modeling

TCR-like antibodies represent a unique type of engineered antibodies with specificity toward pHLA, a ligand normally restricted to the sensitive recognition by T cells. Here, we report a phage display-based sequential development path of such antibodies. The strategy goes from initial lead identification through in silico informed CDR engineering in combination with framework engineering for affinity and thermostability optimization, respectively. The strategy allowed the identification of HLA-DQ2.5 gluten peptide-specific TCR-like antibodies with low picomolar affinity. Our method outlines an efficient and general method for development of this promising class of antibodies, which should facilitate their utility including translation to human therapy.

Stabilization of the SARS-CoV-2 receptor binding domain by protein core redesign and deep mutational scanning

Stabilizing antigenic proteins as vaccine immunogens or diagnostic reagents is a stringent case of protein engineering and design as the exterior surface must maintain recognition by receptor(s) and antigen-specific antibodies at multiple distinct epitopes. This is a challenge, as stability enhancing mutations must be focused on the protein core, whereas successful computational stabilization algorithms typically select mutations at solvent-facing positions. In this study, we report the stabilization of SARS-CoV-2 Wuhan Hu-1 Spike receptor binding domain using a combination of deep mutational scanning and computational design, including the FuncLib algorithm. Our most successful design encodes I358F, Y365W, T430I, and I513L receptor binding domain mutations, maintains recognition by the receptor ACE2 and a panel of different anti-receptor binding domain monoclonal antibodies, is between 1 and 2°C more thermally stable than the original receptor binding domain using a thermal shift assay, and is less proteolytically sensitive to chymotrypsin and thermolysin than the original receptor binding domain. Our approach could be applied to the computational stabilization of a wide range of proteins without requiring detailed knowledge of active sites or binding epitopes. We envision that this strategy may be particularly powerful for cases when there are multiple or unknown binding sites.

Antibody structure prediction using interpretable deep learning

Therapeutic antibodies make up a rapidly growing segment of the biologics market. However, rational design of antibodies is hindered by reliance on experimental methods for determining antibody structures. Here, we present DeepAb, a deep learning method for predicting accurate antibody F_V structures from sequence. We evaluate DeepAb on a set of structurally diverse, therapeutically relevant antibodies and find that our method consistently outperforms the leading alternatives. Previous deep learning methods have operated as “black boxes” and offered few insights into their predictions. By introducing a directly interpretable attention mechanism, we show our network attends to physically important residue pairs (e.g., proximal aromatics and key hydrogen bonding interactions). Finally, we present a novel mutant scoring metric derived from network confidence and show that for a particular antibody, all eight of the top-ranked mutations improve binding affinity. This model will be useful for a broad range of antibody prediction and design tasks.

•

DeepAb, a deep learning method for antibody structure, is presented

•

Structures from DeepAb are more accurate than alternatives

•

Outputs of DeepAb provide interpretable insights into structure predictions

•

DeepAb predictions should facilitate design of novel antibody therapeutics

Accurate structure models are critical for understanding the properties of potential therapeutic antibodies. Conventional methods for protein structure determination require significant investments of time and resources and may fail. Although greatly improved, methods for general protein structure prediction still cannot consistently provide the accuracy necessary to understand or design antibodies. We present a deep learning method for antibody structure prediction and demonstrate improvement over alternatives on diverse, therapeutically relevant benchmarks. In addition to its improved accuracy, our method reveals interpretable outputs about specific amino acids and residue interactions that should facilitate design of novel therapeutic antibodies.

Assessment of Therapeutic Antibody Developability by Combinations of In Vitro and In Silico Methods

Although antibodies have become the fastest-growing class of therapeutics on the market, it is still challenging to develop them for therapeutic applications, which often require these molecules to withstand stresses that are not present in vivo. We define developability as the likelihood of an antibody candidate with suitable functionality to be developed into a manufacturable, stable, safe, and effective drug that can be formulated to high concentrations while retaining a long shelf life. The implementation of reliable developability assessments from the early stages of antibody discovery enables flagging and deselection of potentially problematic candidates, while focussing available resources on the development of the most promising ones. Currently, however, thorough developability assessment requires multiple in vitro assays, which makes it labor intensive and time consuming to implement at early stages. Furthermore, accurate in vitro analysis at the early stage is compromised by the high number of potential candidates that are often prepared at low quantities and purity. Recent improvements in the performance of computational predictors of developability potential are beginning to change this scenario. Many computational methods only require the knowledge of the amino acid sequences and can be used to identify possible developability issues or to rank available candidates according to a range of biophysical properties. Here, we describe how the implementation of in silico tools into antibody discovery pipelines is increasingly offering time- and cost-effective alternatives to in vitro experimental screening, thus streamlining the drug development process. We discuss in particular the biophysical and biochemical properties that underpin developability potential and their trade-offs, review various in vitro assays to measure such properties or parameters that are predictive of developability, and give an overview of the growing number of in silico tools available to predict properties important for antibody development, including the CamSol method developed in our laboratory.

Paving the way towards continuous biosensing by implementing affinity-based nanoswitches on state-dependent readout platforms

Combining affinity-based nanoswitches with state-dependent readout platforms allows for continuous biosensing and acquisition of real-time information about biochemical processes occurring in the environment of interest.