Structure-based protein design with deep learning

Protein subcellular localization prediction tools

Protein subcellular localization prediction is of great significance in bioinformatics and biological research. Most of the proteins do not have experimentally determined localization information, computational prediction methods and tools have been acting as an active research area for more than two decades now. Knowledge of the subcellular location of a protein provides valuable information about its functionalities, the functioning of the cell, and other possible interactions with proteins. Fast, reliable, and accurate predictors provides platforms to harness the abundance of sequence data to predict subcellular locations accordingly. During the last decade, there has been a considerable amount of research effort aimed at developing subcellular localization predictors. This paper reviews recent subcellular localization prediction tools in the Eukaryotic, Prokaryotic, and Virus-based categories followed by a detailed analysis. Each predictor is discussed based on its main features, strengths, weaknesses, algorithms used, prediction techniques, and analysis. This review is supported by prediction tools taxonomies that highlight their rele- vant area and examples for uncomplicated categorization and ease of understandability. These taxonomies help users find suitable tools according to their needs. Furthermore, recent research gaps and challenges are discussed to cover areas that need the utmost attention. This survey provides an in-depth analysis of the most recent prediction tools to facilitate readers and can be considered a quick guide for researchers to identify and explore the recent literature advancements.

Cytochrome P450 Enzyme Design by Constraining the Catalytic Pocket in a Diffusion Model

Although cytochrome P450 enzymes are the most versatile biocatalysts in nature, there is insufficient comprehension of the molecular mechanism underlying their functional innovation process. Here, by combining ancestral sequence reconstruction, reverse mutation assay, and progressive forward accumulation, we identified 5 founder residues in the catalytic pocket of flavone 6-hydroxylase (F6H) and proposed a "3-point fixation" model to elucidate the functional innovation mechanisms of P450s in nature. According to this design principle of catalytic pocket, we further developed a de novo diffusion model (P450Diffusion) to generate artificial P450s. Ultimately, among the 17 non-natural P450s we generated, 10 designs exhibited significant F6H activity and 6 exhibited a 1.3- to 3.5-fold increase in catalytic capacity compared to the natural CYP706X1. This work not only explores the design principle of catalytic pockets of P450s, but also provides an insight into the artificial design of P450 enzymes with desired functions.

The State-of-the-Art Overview to Application of Deep Learning in Accurate Protein Design and Structure Prediction

In recent years, there has been a notable increase in the scientific community's interest in rational protein design. The prospect of designing an amino acid sequence that can reliably fold into a desired three-dimensional structure and exhibit the intended function is captivating. However, a major challenge in this endeavor lies in accurately predicting the resulting protein structure. The exponential growth of protein databases has fueled the advancement of the field, while newly developed algorithms have pushed the boundaries of what was previously achievable in structure prediction. In particular, using deep learning methods instead of brute force approaches has emerged as a faster and more accurate strategy. These deep-learning techniques leverage the vast amount of data available in protein databases to extract meaningful patterns and predict protein structures with improved precision. In this article, we explore the recent developments in the field of protein structure prediction. We delve into the newly developed methods that leverage deep learning approaches, highlighting their significance and potential for advancing our understanding of protein design.

Rationally seeded computational protein design of ɑ-helical barrels

Computational protein design is advancing rapidly. Here we describe efficient routes starting from validated parallel and antiparallel peptide assemblies to design two families of α-helical barrel proteins with central channels that bind small molecules. Computational designs are seeded by the sequences and structures of defined de novo oligomeric barrel-forming peptides, and adjacent helices are connected by loop building. For targets with antiparallel helices, short loops are sufficient. However, targets with parallel helices require longer connectors; namely, an outer layer of helix-turn-helix-turn-helix motifs that are packed onto the barrels. Throughout these computational pipelines, residues that define open states of the barrels are maintained. This minimizes sequence sampling, accelerating the design process. For each of six targets, just two to six synthetic genes are made for expression in Escherichia coli. On average, 70% of these genes express to give soluble monomeric proteins that are fully characterized, including high-resolution structures for most targets that match the design models with high accuracy.

Fast, accurate ranking of engineered proteins by target-binding propensity using structure modeling

Deep-learning-based methods for protein structure prediction have achieved unprecedented accuracy, yet their utility in the engineering of protein-based binders remains constrained due to a gap between the ability to predict the structures of candidate proteins and the ability toprioritize proteins by their potential to bind to a target. To bridge this gap, we introduce Automated Pairwise Peptide-Receptor Analysis for Screening Engineered proteins (APPRAISE), a method for predicting the target-binding propensity of engineered proteins. After generating structural models of engineered proteins competing for binding to a target using an established structure prediction tool such as AlphaFold-Multimer or ESMFold, APPRAISE performs a rapid (under 1 CPU second per model) scoring analysis that takes into account biophysical and geometrical constraints. As proof-of-concept cases, we demonstrate that APPRAISE can accurately classify receptor-dependent vs. receptor-independent adeno-associated viral vectors and diverse classes of engineered proteins such as miniproteins targeting the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) spike, nanobodies targeting a G-protein-coupled receptor, and peptides that specifically bind to transferrin receptor or programmed death-ligand 1 (PD-L1). APPRAISE is accessible through a web-based notebook interface using Google Colaboratory (https://tiny.cc/APPRAISE). With its accuracy, interpretability, and generalizability, APPRAISE promises to expand the utility of protein structure prediction and accelerate protein engineering for biomedical applications.

Enhancing Protein Solubility via Glycosylation: From Chemical Synthesis to Machine Learning Predictions

Glycosylation is a valuable tool for modulating protein solubility; however, the lack of reliable research strategies has impeded efficient progress in understanding and applying this modification. This study aimed to bridge this gap by investigating the solubility of a model glycoprotein molecule, the carbohydrate-binding module (CBM), through a two-stage process. In the first stage, an approach involving chemical synthesis, comparative analysis, and molecular dynamics simulations of a library of glycoforms was employed to elucidate the effect of different glycosylation patterns on solubility and the key factors responsible for the effect. In the second stage, a predictive mathematical formula, innovatively harnessing machine learning algorithms, was derived to relate solubility to the identified key factors and accurately predict the solubility of the newly designed glycoforms. Demonstrating feasibility and effectiveness, this two-stage approach offers a valuable strategy for advancing glycosylation research, especially for the discovery of glycoforms with increased solubility.

The structural landscape of the immunoglobulin fold by large-scale de novo design

De novo designing immunoglobulin-like frameworks that allow for functional loop diversification shows great potential for crafting antibody-like scaffolds with fully customizable structures and functions. In this work, we combined de novo parametric design with deep-learning methods for protein structure prediction and design to explore the structural landscape of 7-stranded immunoglobulin domains. After screening folding of nearly 4 million designs, we have assembled a structurally diverse library of ~50,000 immunoglobulin domains with high-confidence AlphaFold2 predictions and structures diverging from naturally occurring ones. The designed dataset enabled us to identify structural requirements for the correct folding of immunoglobulin domains, shed light on β-sheet-β-sheet rotational preferences and how these are linked to functional properties. Our approach eliminates the need for preset loop conformations and opens the route to large-scale de novo design of immunoglobulin-like frameworks.

SPDesign: protein sequence designer based on structural sequence profile using ultrafast shape recognition

Protein sequence design can provide valuable insights into biopharmaceuticals and disease treatments. Currently, most protein sequence design methods based on deep learning focus on network architecture optimization, while ignoring protein-specific physicochemical features. Inspired by the successful application of structure templates and pre-trained models in the protein structure prediction, we explored whether the representation of structural sequence profile can be used for protein sequence design. In this work, we propose SPDesign, a method for protein sequence design based on structural sequence profile using ultrafast shape recognition. Given an input backbone structure, SPDesign utilizes ultrafast shape recognition vectors to accelerate the search for similar protein structures in our in-house PAcluster80 structure database and then extracts the sequence profile through structure alignment. Combined with structural pre-trained knowledge and geometric features, they are further fed into an enhanced graph neural network for sequence prediction. The results show that SPDesign significantly outperforms the state-of-the-art methods, such as ProteinMPNN, Pifold and LM-Design, leading to 21.89%, 15.54% and 11.4% accuracy gains in sequence recovery rate on CATH 4.2 benchmark, respectively. Encouraging results also have been achieved on orphan and de novo (designed) benchmarks with few homologous sequences. Furthermore, analysis conducted by the PDBench tool suggests that SPDesign performs well in subdivided structures. More interestingly, we found that SPDesign can well reconstruct the sequences of some proteins that have similar structures but different sequences. Finally, the structural modeling verification experiment indicates that the sequences designed by SPDesign can fold into the native structures more accurately.

Opportunities and Challenges for Machine Learning-Assisted Enzyme Engineering

Enzymes can be engineered at the level of their amino acid sequences to optimize key properties such as expression, stability, substrate range, and catalytic efficiency-or even to unlock new catalytic activities not found in nature. Because the search space of possible proteins is vast, enzyme engineering usually involves discovering an enzyme starting point that has some level of the desired activity followed by directed evolution to improve its "fitness" for a desired application. Recently, machine learning (ML) has emerged as a powerful tool to complement this empirical process. ML models can contribute to (1) starting point discovery by functional annotation of known protein sequences or generating novel protein sequences with desired functions and (2) navigating protein fitness landscapes for fitness optimization by learning mappings between protein sequences and their associated fitness values. In this Outlook, we explain how ML complements enzyme engineering and discuss its future potential to unlock improved engineering outcomes.

Leveraging Artificial Intelligence to Expedite Antibody Design and Enhance Antibody-Antigen Interactions

This perspective sheds light on the transformative impact of recent computational advancements in the field of protein therapeutics, with a particular focus on the design and development of antibodies. Cutting-edge computational methods have revolutionized our understanding of protein-protein interactions (PPIs), enhancing the efficacy of protein therapeutics in preclinical and clinical settings. Central to these advancements is the application of machine learning and deep learning, which offers unprecedented insights into the intricate mechanisms of PPIs and facilitates precise control over protein functions. Despite these advancements, the complex structural nuances of antibodies pose ongoing challenges in their design and optimization. Our review provides a comprehensive exploration of the latest deep learning approaches, including language models and diffusion techniques, and their role in surmounting these challenges. We also present a critical analysis of these methods, offering insights to drive further progress in this rapidly evolving field. The paper includes practical recommendations for the application of these computational techniques, supplemented with independent benchmark studies. These studies focus on key performance metrics such as accuracy and the ease of program execution, providing a valuable resource for researchers engaged in antibody design and development. Through this detailed perspective, we aim to contribute to the advancement of antibody design, equipping researchers with the tools and knowledge to navigate the complexities of this field.

Is Novelty Predictable?

Machine learning-based design has gained traction in the sciences, most notably in the design of small molecules, materials, and proteins, with societal applications ranging from drug development and plastic degradation to carbon sequestration. When designing objects to achieve novel property values with machine learning, one faces a fundamental challenge: how to push past the frontier of current knowledge, distilled from the training data into the model, in a manner that rationally controls the risk of failure. If one trusts learned models too much in extrapolation, one is likely to design rubbish. In contrast, if one does not extrapolate, one cannot find novelty. Herein, we ponder how one might strike a useful balance between these two extremes. We focus in particular on designing proteins with novel property values, although much of our discussion is relevant to machine learning-based design more broadly.

De novo protein design-From new structures to programmable functions

Methods from artificial intelligence (AI) trained on large datasets of sequences and structures can now "write" proteins with new shapes and molecular functions de novo, without starting from proteins found in nature. In this Perspective, I will discuss the state of the field of de novo protein design at the juncture of physics-based modeling approaches and AI. New protein folds and higher-order assemblies can be designed with considerable experimental success rates, and difficult problems requiring tunable control over protein conformations and precise shape complementarity for molecular recognition are coming into reach. Emerging approaches incorporate engineering principles-tunability, controllability, and modularity-into the design process from the beginning. Exciting frontiers lie in deconstructing cellular functions with de novo proteins and, conversely, constructing synthetic cellular signaling from the ground up. As methods improve, many more challenges are unsolved.

Unlocking novel therapies: cyclic peptide design for amyloidogenic targets through synergies of experiments, simulations, and machine learning

Existing therapies for neurodegenerative diseases like Parkinson's and Alzheimer's address only their symptoms and do not prevent disease onset. Common therapeutic agents, such as small molecules and antibodies struggle with insufficient selectivity, stability and bioavailability, leading to poor performance in clinical trials. Peptide-based therapeutics are emerging as promising candidates, with successful applications for cardiovascular diseases and cancers due to their high bioavailability, good efficacy and specificity. In particular, cyclic peptides have a long in vivo stability, while maintaining a robust antibody-like binding affinity. However, the de novo design of cyclic peptides is challenging due to the lack of long-lived druggable pockets of the target polypeptide, absence of exhaustive conformational distributions of the target and/or the binder, unknown binding site, methodological limitations, associated constraints (failed trials, time, money) and the vast combinatorial sequence space. Hence, efficient alignment and cooperation between disciplines, and synergies between experiments and simulations complemented by popular techniques like machine-learning can significantly speed up the therapeutic cyclic-peptide development for neurodegenerative diseases. We review the latest advancements in cyclic peptide design against amyloidogenic targets from a computational perspective in light of recent advancements and potential of machine learning to optimize the design process. We discuss the difficulties encountered when designing novel peptide-based inhibitors and we propose new strategies incorporating experiments, simulations and machine learning to design cyclic peptides to inhibit the toxic propagation of amyloidogenic polypeptides. Importantly, these strategies extend beyond the mere design of cyclic peptides and serve as template for the de novo generation of (bio)materials with programmable properties.

PLM-GAN: A Large-Scale Protein Loop Modeling Using pix2pix GAN

Revealing the tertiary structure of proteins holds huge significance as it unveils their vital properties and functions. These intricate three-dimensional configurations comprise diverse interactions including ionic, hydrophobic, and disulfide forces. In certain instances, these structures exhibit missing regions, necessitating the reconstruction of specific segments, thereby resulting in challenges in protein design, which encompasses loop modeling, circular permutation, and interface prediction. To address this problem, we present two pioneering models: pix2pix generative adversarial network (GAN) and PLM-GAN. The pix2pix GAN model is adept at generating and inpainting distance matrices of protein structures, whereas the PLM-GAN model incorporates residual blocks into the U-Net network of the GAN, building upon the foundation of the pix2pix GAN model. To bolster the models' performance, we introduce a novel loss function named the "missing to real regions loss" (LMTR) within the GAN framework. Additionally, we introduce a distinctive approach of pairing two different distance matrices: one representing the native protein structure and the other representing the same structure with a missing region that undergoes changes in each successive epoch. Moreover, we extend the reconstruction of missing regions, encompassing up to 30 amino acids and increase the protein length by 128 amino acids. The evaluation of our pix2pix GAN and PLM-GAN models on a random selection of natural proteins (4ZCB, 3FJB, and 2REZ) demonstrated promising experimental results. Our models constitute significant contributions to addressing intricate challenges in protein structure design. These contributions hold immense potential to propel advancements in protein-protein interactions, drug design, and further innovations in protein engineering. Data, code, trained models, examples, and measurements are available on https://github.com/mena01/PLM-GAN-A-Large-Scale-Protein-Loop-Modeling-Using-pix2pix-GAN_.

Accelerating the Discovery and Design of Antimicrobial Peptides with Artificial Intelligence

Peptides modulate many processes of human physiology targeting ion channels, protein receptors, or enzymes. They represent valuable starting points for the development of new biologics against communicable and non-communicable disorders. However, turning native peptide ligands into druggable materials requires high selectivity and efficacy, predictable metabolism, and good safety profiles. Machine learning models have gradually emerged as cost-effective and time-saving solutions to predict and generate new proteins with optimal properties. In this chapter, we will discuss the evolution and applications of predictive modeling and generative modeling to discover and design safe and effective antimicrobial peptides. We will also present their current limitations and suggest future research directions, applicable to peptide drug design campaigns.

Intelligent Protein Design and Molecular Characterization Techniques: A Comprehensive Review

In recent years, the widespread application of artificial intelligence algorithms in protein structure, function prediction, and de novo protein design has significantly accelerated the process of intelligent protein design and led to many noteworthy achievements. This advancement in protein intelligent design holds great potential to accelerate the development of new drugs, enhance the efficiency of biocatalysts, and even create entirely new biomaterials. Protein characterization is the key to the performance of intelligent protein design. However, there is no consensus on the most suitable characterization method for intelligent protein design tasks. This review describes the methods, characteristics, and representative applications of traditional descriptors, sequence-based and structure-based protein characterization. It discusses their advantages, disadvantages, and scope of application. It is hoped that this could help researchers to better understand the limitations and application scenarios of these methods, and provide valuable references for choosing appropriate protein characterization techniques for related research in the field, so as to better carry out protein research.

Unsupervisedly Prompting AlphaFold2 for Accurate Few-Shot Protein Structure Prediction

Data-driven predictive methods that can efficiently and accurately transform protein sequences into biologically active structures are highly valuable for scientific research and medical development. Determining an accurate folding landscape using coevolutionary information is fundamental to the success of modern protein structure prediction methods. As the state of the art, AlphaFold2 has dramatically raised the accuracy without performing explicit coevolutionary analysis. Nevertheless, its performance still shows strong dependence on available sequence homologues. Based on the interrogation on the cause of such dependence, we presented EvoGen, a meta generative model, to remedy the underperformance of AlphaFold2 for poor MSA targets. By prompting the model with calibrated or virtually generated homologue sequences, EvoGen helps AlphaFold2 fold accurately in the low-data regime and even achieve encouraging performance with single-sequence predictions. Being able to make accurate predictions with few-shot MSA not only generalizes AlphaFold2 better for orphan sequences but also democratizes its use for high-throughput applications. Besides, EvoGen combined with AlphaFold2 yields a probabilistic structure generation method that could explore alternative conformations of protein sequences, and the task-aware differentiable algorithm for sequence generation will benefit other related tasks including protein design.

Combined prediction and design reveals the target recognition mechanism of an intrinsically disordered protein interaction domain

An increasing number of protein interaction domains and their targets are being found to be intrinsically disordered proteins (IDPs). The corresponding target recognition mechanisms are mostly elusive because of challenges in performing detailed structural analysis of highly dynamic IDP-IDP complexes. Here, we show that by combining recently developed computational approaches with experiments, the structure of the complex between the intrinsically disordered C-terminal domain (CTD) of protein 4.1G and its target IDP region in NuMA can be dissected at high resolution. First, we carry out systematic mutational scanning using dihydrofolate reductase-based protein complementarity analysis to identify essential interaction regions and key residues. The results are found to be highly consistent with an α/β-type complex structure predicted by AlphaFold2 (AF2). We then design mutants based on the predicted structure using a deep learning protein sequence design method. The solved crystal structure of one mutant presents the same core structure as predicted by AF2. Further computational prediction and experimental assessment indicate that the well-defined core structure is conserved across complexes of 4.1G CTD with other potential targets. Thus, we reveal that an intrinsically disordered protein interaction domain uses an α/β-type structure module formed through synergistic folding to recognize broad IDP targets. Moreover, we show that computational prediction and experiment can be jointly applied to segregate true IDP regions from the core structural domains of IDP-IDP complexes and to uncover the structure-dependent mechanisms of some otherwise elusive IDP-IDP interactions.

AlphaFold2 and Deep Learning for Elucidating Enzyme Conformational Flexibility and Its Application for Design

The recent success of AlphaFold2 (AF2) and other deep learning (DL) tools in accurately predicting the folded three-dimensional (3D) structure of proteins and enzymes has revolutionized the structural biology and protein design fields. The 3D structure indeed reveals key information on the arrangement of the catalytic machinery of enzymes and which structural elements gate the active site pocket. However, comprehending enzymatic activity requires a detailed knowledge of the chemical steps involved along the catalytic cycle and the exploration of the multiple thermally accessible conformations that enzymes adopt when in solution. In this Perspective, some of the recent studies showing the potential of AF2 in elucidating the conformational landscape of enzymes are provided. Selected examples of the key developments of AF2-based and DL methods for protein design are discussed, as well as a few enzyme design cases. These studies show the potential of AF2 and DL for allowing the routine computational design of efficient enzymes.

Accelerating therapeutic protein design with computational approaches toward the clinical stage

Therapeutic protein, represented by antibodies, is of increasing interest in human medicine. However, clinical translation of therapeutic protein is still largely hindered by different aspects of developability, including affinity and selectivity, stability and aggregation prevention, solubility and viscosity reduction, and deimmunization. Conventional optimization of the developability with widely used methods, like display technologies and library screening approaches, is a time and cost-intensive endeavor, and the efficiency in finding suitable solutions is still not enough to meet clinical needs. In recent years, the accelerated advancement of computational methodologies has ushered in a transformative era in the field of therapeutic protein design. Owing to their remarkable capabilities in feature extraction and modeling, the integration of cutting-edge computational strategies with conventional techniques presents a promising avenue to accelerate the progression of therapeutic protein design and optimization toward clinical implementation. Here, we compared the differences between therapeutic protein and small molecules in developability and provided an overview of the computational approaches applicable to the design or optimization of therapeutic protein in several developability issues.

Machine learning for evolutionary-based and physics-inspired protein design: Current and future synergies

Computational protein design facilitates the discovery of novel proteins with prescribed structure and functionality. Exciting designs were recently reported using novel data-driven methodologies that can be roughly divided into two categories: evolutionary-based and physics-inspired approaches. The former infer characteristic sequence features shared by sets of evolutionary-related proteins, such as conserved or coevolving positions, and recombine them to generate candidates with similar structure and function. The latter approaches estimate key biochemical properties, such as structure free energy, conformational entropy, or binding affinities using machine learning surrogates, and optimize them to yield improved designs. Here, we review recent progress along both tracks, discuss their strengths and weaknesses, and highlight opportunities for synergistic approaches.

Biomimetic Nanotechnology Vol. 3

Biomimetic nanotechnology pertains to the fundamental elements of living systems and the translation of their properties into human applications. The underlying functionalities of biological materials, structures and processes are primarily rooted in the nanoscale domain, serving as a source of inspiration for materials science, medicine, physics, sensor technologies, smart materials science and other interdisciplinary fields. The Biomimetics Special Issues Biomimetic Nanotechnology Vols. 1–3 feature a collection of research and review articles contributed by experts in the field, delving into significant realms of biomimetic nanotechnology. This publication, Vol. 3, comprises four research articles and one review article, which offer valuable insights and inspiration for innovative approaches inspired by Nature’s living systems. The spectrum of the articles is wide and deep and ranges from genetics, traditional medicine, origami, fungi and quartz to green synthesis of nanoparticles.

De novo design and Rosetta-based assessment of high-affinity antibody variable regions (Fv) against the SARS-CoV-2 spike receptor binding domain (RBD)

The continued emergence of new SARS-CoV-2 variants has accentuated the growing need for fast and reliable methods for the design of potentially neutralizing antibodies (Abs) to counter immune evasion by the virus. Here, we report on the de novo computational design of high-affinity Ab variable regions (Fv) through the recombination of VDJ genes targeting the most solvent-exposed hACE2-binding residues of the SARS-CoV-2 spike receptor binding domain (RBD) protein using the software tool OptMAVEn-2.0. Subsequently, we carried out computational affinity maturation of the designed variable regions through amino acid substitutions for improved binding with the target epitope. Immunogenicity of designs was restricted by preferring designs that match sequences from a 9-mer library of "human Abs" based on a human string content score. We generated 106 different antibody designs and reported in detail on the top five that trade-off the greatest computational binding affinity for the RBD with human string content scores. We further describe computational evaluation of the top five designs produced by OptMAVEn-2.0 using a Rosetta-based approach. We used Rosetta SnugDock for local docking of the designs to evaluate their potential to bind the spike RBD and performed "forward folding" with DeepAb to assess their potential to fold into the designed structures. Ultimately, our results identified one designed Ab variable region, P1.D1, as a particularly promising candidate for experimental testing. This effort puts forth a computational workflow for the de novo design and evaluation of Abs that can quickly be adapted to target spike epitopes of emerging SARS-CoV-2 variants or other antigenic targets.

Neural network-derived Potts models for structure-based protein design using backbone atomic coordinates and tertiary motifs

Designing novel proteins to perform desired functions, such as binding or catalysis, is a major goal in synthetic biology. A variety of computational approaches can aid in this task. An energy-based framework rooted in the sequence-structure statistics of tertiary motifs (TERMs) can be used for sequence design on predefined backbones. Neural network models that use backbone coordinate-derived features provide another way to design new proteins. In this work, we combine the two methods to make neural structure-based models more suitable for protein design. Specifically, we supplement backbone-coordinate features with TERM-derived data, as inputs, and we generate energy functions as outputs. We present two architectures that generate Potts models over the sequence space: TERMinator, which uses both TERM-based and coordinate-based information, and COORDinator, which uses only coordinate-based information. Using these two models, we demonstrate that TERMs can be utilized to improve native sequence recovery performance of neural models. Furthermore, we demonstrate that sequences designed by TERMinator are predicted to fold to their target structures by AlphaFold. Finally, we show that both TERMinator and COORDinator learn notions of energetics, and these methods can be fine-tuned on experimental data to improve predictions. Our results suggest that using TERM-based and coordinate-based features together may be beneficial for protein design and that structure-based neural models that produce Potts energy tables have utility for flexible applications in protein science.

Masked inverse folding with sequence transfer for protein representation learning

Self-supervised pretraining on protein sequences has led to state-of-the art performance on protein function and fitness prediction. However, sequence-only methods ignore the rich information contained in experimental and predicted protein structures. Meanwhile, inverse folding methods reconstruct a protein's amino-acid sequence given its structure, but do not take advantage of sequences that do not have known structures. In this study, we train a masked inverse folding protein masked language model parameterized as a structured graph neural network. During pretraining, this model learns to reconstruct corrupted sequences conditioned on the backbone structure. We then show that using the outputs from a pretrained sequence-only protein masked language model as input to the inverse folding model further improves pretraining perplexity. We evaluate both of these models on downstream protein engineering tasks and analyze the effect of using information from experimental or predicted structures on performance.

Novel Split Intein-Mediated Enzymatic Channeling Accelerates the Multimeric Bioconversion Pathway of Ginsenoside

Cascade reaction systems, such as protein fusion and synthetic protein scaffold systems, can both spatially control the metabolic flux and boost the productivity of multistep enzymatic reactions. Despite many efforts to generate fusion proteins, this task remains challenging due to the limited expression of complex enzymes. Therefore, we developed a novel fusion system that bypasses the limited expression of complex enzymes via a post-translational linkage. Here, we report a split intein-mediated cascade system wherein orthogonal split inteins serve as adapters for protein ligation. A genetically programmable, self-assembled, and traceless split intein was utilized to generate a biocatalytic cascade to produce the ginsenoside compound K (CK) with various pharmacological activities, including anticarcinogenic, anti-inflammatory, and antidiabetic effects. We used two types of split inteins, consensus atypical (Cat) and Rma DnaB, to form a covalent scaffold with the three enzymes involved in the CK conversion pathway. The multienzymatic complex with a size greater than 240 kDa was successfully assembled in a soluble form and exhibited specific activity toward ginsenoside conversion. Furthermore, our split intein cascade system significantly increased the CK conversion rate and reduced the production time by more than 2-fold. Our multienzymatic cascade system that uses split inteins can be utilized as a platform for regulating multimeric bioconversion pathways and boosting the production of various high-value substances.

Hallucinating symmetric protein assemblies

Deep learning generative approaches provide an opportunity to broadly explore protein structure space beyond the sequences and structures of natural proteins. Here, we use deep network hallucination to generate a wide range of symmetric protein homo-oligomers given only a specification of the number of protomers and the protomer length. Crystal structures of seven designs are very similar to the computational models (median root mean square deviation: 0.6 angstroms), as are three cryo-electron microscopy structures of giant 10-nanometer rings with up to 1550 residues and C33 symmetry; all differ considerably from previously solved structures. Our results highlight the rich diversity of new protein structures that can be generated using deep learning and pave the way for the design of increasingly complex components for nanomachines and biomaterials.

Scaffolding protein functional sites using deep learning

The binding and catalytic functions of proteins are generally mediated by a small number of functional residues held in place by the overall protein structure. Here, we describe deep learning approaches for scaffolding such functional sites without needing to prespecify the fold or secondary structure of the scaffold. The first approach, "constrained hallucination," optimizes sequences such that their predicted structures contain the desired functional site. The second approach, "inpainting," starts from the functional site and fills in additional sequence and structure to create a viable protein scaffold in a single forward pass through a specifically trained RoseTTAFold network. We use these two methods to design candidate immunogens, receptor traps, metalloproteins, enzymes, and protein-binding proteins and validate the designs using a combination of in silico and experimental tests.

Rotamer-free protein sequence design based on deep learning and self-consistency

Several previously proposed deep learning methods to design amino acid sequences that autonomously fold into a given protein backbone yielded promising results in computational tests but did not outperform conventional energy function-based methods in wet experiments. Here we present the ABACUS-R method, which uses an encoder-decoder network trained using a multitask learning strategy to predict the sidechain type of a central residue from its three-dimensional local environment, which includes, besides other features, the types but not the conformations of the surrounding sidechains. This eliminates the need to reconstruct and optimize sidechain structures, and drastically simplifies the sequence design process. Thus iteratively applying the encoder-decoder to different central residues is able to produce self-consistent overall sequences for a target backbone. Results of wet experiments, including five structures solved by X-ray crystallography, show that ABACUS-R outperforms state-of-the-art energy function-based methods in success rate and design precision.

Ig-VAE: Generative modeling of protein structure by direct 3D coordinate generation

While deep learning models have seen increasing applications in protein science, few have been implemented for protein backbone generation—an important task in structure-based problems such as active site and interface design. We present a new approach to building class-specific backbones, using a variational auto-encoder to directly generate the 3D coordinates of immunoglobulins. Our model is torsion- and distance-aware, learns a high-resolution embedding of the dataset, and generates novel, high-quality structures compatible with existing design tools. We show that the Ig-VAE can be used with Rosetta to create a computational model of a SARS-CoV2-RBD binder via latent space sampling. We further demonstrate that the model’s generative prior is a powerful tool for guiding computational protein design, motivating a new paradigm under which backbone design is solved as constrained optimization problem in the latent space of a generative model.

Many essential biochemical processes are governed by protein-protein interactions (PPIs), and our ability to make binding proteins that modulate PPIs is crucial to the creation of therapeutics and the study of cell-signaling. One critical aspect of PPI design is to capture protein conformational flexibility. Deep generative models are a class of mathematical models that are able to synthesize novel data from a finite set of training examples. Here, we make advances in computational protein design methodology by developing a deep generative model that creates protein backbones adopting the immunoglobulin fold, which is found in natural binding proteins such as antibodies. While generative models have been powerful in tasks such as image generation, using them to create proteins has remained a challenge. We solve this problem with a new model that allows for the direct generation of novel 3D molecules and show that they are of high chemical accuracy. Generated structures work well with existing protein design methods such as Rosetta, providing access to a large collection of novel immunoglobulin structures. Finally, we present a new protein design framework, called “generative design,” that shows how deep generative models such as ours can be applied to virtually any protein design problem.

Progress and challenges for the machine learning-based design of fit-for-purpose monoclonal antibodies

Although the therapeutic efficacy and commercial success of monoclonal antibodies (mAbs) are tremendous, the design and discovery of new candidates remain a time and cost-intensive endeavor. In this regard, progress in the generation of data describing antigen binding and developability, computational methodology, and artificial intelligence may pave the way for a new era of in silico on-demand immunotherapeutics design and discovery. Here, we argue that the main necessary machine learning (ML) components for an in silico mAb sequence generator are: understanding of the rules of mAb-antigen binding, capacity to modularly combine mAb design parameters, and algorithms for unconstrained parameter-driven in silico mAb sequence synthesis. We review the current progress toward the realization of these necessary components and discuss the challenges that must be overcome to allow the on-demand ML-based discovery and design of fit-for-purpose mAb therapeutic candidates.

Deep learning approaches for conformational flexibility and switching properties in protein design

Following the hugely successful application of deep learning methods to protein structure prediction, an increasing number of design methods seek to leverage generative models to design proteins with improved functionality over native proteins or novel structure and function. The inherent flexibility of proteins, from side-chain motion to larger conformational reshuffling, poses a challenge to design methods, where the ideal approach must consider both the spatial and temporal evolution of proteins in the context of their functional capacity. In this review, we highlight existing methods for protein design before discussing how methods at the forefront of deep learning-based design accommodate flexibility and where the field could evolve in the future.