Machine-Learning-Guided Mutagenesis for Directed Evolution of Fluorescent Proteins

Fluorescent probes in autoimmune disease research: current status and future prospects

Autoimmune diseases (AD) present substantial challenges for early diagnosis and precise treatment due to their intricate pathogenesis and varied clinical manifestations. While existing diagnostic methods and treatment strategies have advanced, their sensitivity, specificity, and real-time applicability in clinical settings continue to exhibit significant limitations. In recent years, fluorescent probes have emerged as highly sensitive and specific biological imaging tools, demonstrating substantial potential in AD research.This review examines the response mechanisms and historical evolution of various types of fluorescent probes, systematically summarizing the latest research advancements in their application to autoimmune diseases. It highlights key applications in biomarker detection, dynamic monitoring of immune cell functions, and assessment of drug treatment efficacy. Furthermore, this article analyzes the technical challenges currently encountered in probe development and proposes potential directions for future research. With ongoing advancements in materials science, nanotechnology, and bioengineering, fluorescent probes are anticipated to achieve higher sensitivity and enhanced functional integration, thereby facilitating early detection, dynamic monitoring, and innovative treatment strategies for autoimmune diseases. Overall, fluorescent probes possess substantial scientific significance and application value in both research and clinical settings related to autoimmune diseases, signaling a new era of personalized and precision medicine.

Engineering highly active nuclease enzymes with machine learning and high-throughput screening

Optimizing enzymes to function in novel chemical environments is a central goal of synthetic biology, but optimization is often hindered by a rugged fitness landscape and costly experiments. In this work, we present TeleProt, a machine learning (ML) framework that blends evolutionary and experimental data to design diverse protein libraries, and employ it to improve the catalytic activity of a nuclease enzyme that degrades biofilms that accumulate on chronic wounds. After multiple rounds of high-throughput experiments, TeleProt found a significantly better top-performing enzyme than directed evolution (DE), had a better hit rate at finding diverse, high-activity variants, and was even able to design a high-performance initial library using no prior experimental data. We have released a dataset of 55,000 nuclease variants, one of the most extensive genotype-phenotype enzyme activity landscapes to date, to drive further progress in ML-guided design. A record of this paper's transparent peer review process is included in the supplemental information.

Principles and Design of Molecular Tools for Sensing and Perturbing Cell Surface Receptor Activity

Cell-surface receptors are vital for controlling numerous cellular processes with their dysregulation being linked to disease states. Therefore, it is necessary to develop tools to study receptors and the signaling pathways they control. This Review broadly describes molecular approaches that enable 1) the visualization of receptors to determine their localization and distribution; 2) sensing receptor activation with permanent readouts as well as readouts in real time; and 3) perturbing receptor activity and mimicking receptor-controlled processes to learn more about these processes. Together, these tools have provided valuable insight into fundamental receptor biology and helped to characterize therapeutics that target receptors.

Computational Design of Burkholderia cepacia Lipase Mutants that Show Enhanced Stereoselectivity in the Production of l-Menthol

l-Menthol, valued for its aroma, cooling properties, and biological activity, is often produced as a mixture with d-menthol, which negatively impacts taste and odor. To improve the purity of industrial l-menthol production, we engineered Burkholderia cepacia lipase (BCL) to improve its stereoselectivity. While wild-type BCL achieves only 98% enantiomeric excess, we used molecular dynamics simulations of BCL bound to menthol acetate, combined with computational rational design method (MSPER), to identify key mutation sites. We experimentally confirmed that five mutations that can effectively improve the enantiomeric excess. In particular, Q88A increased enantiomeric excess to 99.4% and productivity from 14.5% to 49.9%. Q88G exhibited the highest productivity at 54.0% with 99.3% optical purity. These BCL mutants offer a greener, more efficient approach to high-purity l-menthol production, contributing to more sustainable chemical processes.

Integrating protein language models and automatic biofoundry for enhanced protein evolution

Traditional protein engineering methods, such as directed evolution, while effective, are often slow and labor-intensive. Advances in machine learning and automated biofoundry present new opportunities for optimizing these processes. This study devises a protein language model-enabled automatic evolution platform, a closed-loop system for automated protein engineering within the Design-Build-Test-Learn cycle. The protein language model ESM-2 makes zero-shot prediction of 96 variants to initiate the cycle. The biofoundry constructs and evaluates these variants, and feeds the results back to a multi-layer perceptron to train a fitness predictor, which then makes prediction of second round of 96 variants with improved fitness. With the tRNA synthetase as a model enzyme, four-rounds of evolution carried out within 10 days lead to mutants with enzyme activity improved by up to 2.4-fold. Our system significantly enhances the speed and accuracy of protein evolution, driving faster advancements in protein engineering for industrial applications.

Biophysics-based protein language models for protein engineering

Protein language models trained on evolutionary data have emerged as powerful tools for predictive problems involving protein sequence, structure, and function. However, these models overlook decades of research into biophysical factors governing protein function. We propose Mutational Effect Transfer Learning (METL), a protein language model framework that unites advanced machine learning and biophysical modeling. Using the METL framework, we pretrain transformer-based neural networks on biophysical simulation data to capture fundamental relationships between protein sequence, structure, and energetics. We finetune METL on experimental sequence-function data to harness these biophysical signals and apply them when predicting protein properties like thermostability, catalytic activity, and fluorescence. METL excels in challenging protein engineering tasks like generalizing from small training sets and position extrapolation, although existing methods that train on evolutionary signals remain powerful for many types of experimental assays. We demonstrate METL's ability to design functional green fluorescent protein variants when trained on only 64 examples, showcasing the potential of biophysics-based protein language models for protein engineering.

Transitions in Immunoassay Leading to Next-Generation Lateral Flow Assays and Future Prospects

In the field of clinical testing, the traditional focus has been on the development of large-scale analysis equipment designed to process high volumes of samples with fully automatic and high-sensitivity measurements. However, there has been a growing demand in recent years for the development of analytical reagents tailored to point-of-care testing (POCT), which does not necessitate a specific location or specialized operator. This trend is epitomized using the lateral flow assay (LFA), which became a cornerstone during the 2019 pandemic due to its simplicity, speed of delivering results-within about 10 min from minimal sample concentrations-and user-friendly design. LFAs, with their paper-based construction, combine cost-effectiveness with ease of disposal, addressing both budgetary and environmental concerns comprehensively. Despite their compact size, LFAs encapsulate a wealth of technological ingenuity, embodying years of research and development. Current research is dedicated to further evolving LFA technology, paving the way for the next generation of diagnostic devices. These advancements aim to redefine accessibility, empower individuals, and enhance responsiveness to public health challenges. The future of LFAs, now unfolding, promises even greater integration into routine health management and emergency responses, underscoring their critical role in the evolution of decentralized and patient-centric healthcare solutions. In this review, the historical development of LFA and several of the latest LFA technologies using catalytic amplification, surface-enhanced Raman scattering, heat detection, electron chemical detections, magnetoresistance, and detection of reflected electrons detection are introduced to inspire readers for future research and development.

Machine learning-guided multi-site combinatorial mutagenesis enhances the thermostability of pectin lyase

Enhancing the thermostability of enzymes is crucial for industrial applications. Methods such as directed evolution are often limited by the huge sequence space and combinatorial explosion, making it difficult to obtain optimal mutants. In recent years, machine learning (ML)-guided protein engineering has become an attractive tool because of its ability to comprehensively explore the sequence space of enzymes and discover superior mutants. This study employed ML to perform combinatorial mutation design on the pectin lyase PMGL-Ba from Bacillus licheniformis, aiming to improve its thermostability. First, 18 single-point mutants with enhanced thermostability were identified through semi-rational design. Subsequently, the initial library containing a small number of low-order mutants was utilized to construct an ML model to explore the combinatorial sequence space (theoretically 196,608 mutants) of single-point mutants. The results showed that the ML-predicted second library was successfully enriched with highly thermostable combinatorial mutants. After one iteration of learning, the best-performing combinatorial mutant in the third library, P36, showed a 67-fold and 39-fold increase in half-life at 75 °C and 80 °C, respectively, as well as a 2.1-fold increase in activity. Structural analysis and molecular dynamics simulations provided insights into the improved performance of the engineered enzyme.

Transforming drug development with synthetic biology and AI

The COVID-19 pandemic has thrust RNA as a platform for drug development into the spotlight. However, identifying promising drug candidates is challenging. With advances in synthetic biology and artificial intelligence (AI) models, we can overcome this hurdle, transforming drug development and ushering in a new era in the pharmaceutical industry.

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.

Enhanced Sequence-Activity Mapping and Evolution of Artificial Metalloenzymes by Active Learning

Tailored enzymes are crucial for the transition to a sustainable bioeconomy. However, enzyme engineering is laborious and failure-prone due to its reliance on serendipity. The efficiency and success rates of engineering campaigns may be improved by applying machine learning to map the sequence-activity landscape based on small experimental data sets. Yet, it often proves challenging to reliably model large sequence spaces while keeping the experimental effort tractable. To address this challenge, we present an integrated pipeline combining large-scale screening with active machine learning, which we applied to engineer an artificial metalloenzyme (ArM) catalyzing a new-to-nature hydroamination reaction. Combining lab automation and next-generation sequencing, we acquired sequence-activity data for several thousand ArM variants. We then used Gaussian process regression to model the activity landscape and guide further screening rounds. Critical characteristics of our pipeline include the cost-effective generation of information-rich data sets, the integration of an explorative round to improve the model's performance, and the inclusion of experimental noise. Our approach led to an order-of-magnitude boost in the hit rate while making efficient use of experimental resources. Search strategies like this should find broad utility in enzyme engineering and accelerate the development of novel biocatalysts.

Application of Directed Evolution and Machine Learning to Enhance the Diastereoselectivity of Ketoreductase for Dihydrotetrabenazine Synthesis

Biocatalysis is an effective approach for producing chiral drug intermediates that are often difficult to synthesize using traditional chemical methods. A time-efficient strategy is required to accelerate the directed evolution process to achieve the desired enzyme function. In this research, we evaluated machine learning-assisted directed evolution as a potential approach for enzyme engineering, using a moderately diastereoselective ketoreductase library as a model system. Machine learning-assisted directed evolution and traditional directed evolution methods were compared for reducing (±)-tetrabenazine to dihydrotetrabenazine via kinetic resolution facilitated by BsSDR10, a short-chain dehydrogenase/reductase from Bacillus subtilis. Both methods successfully identified variants with significantly improved diastereoselectivity for each isomer of dihydrotetrabenazine. Furthermore, the preparation of (2S,3S,11bS)-dihydrotetrabenazine has been successfully scaled up, with an isolated yield of 40.7% and a diastereoselectivity of 91.3%.

Unsupervised evolution of protein and antibody complexes with a structure-informed language model

Large language models trained on sequence information alone can learn high-level principles of protein design. However, beyond sequence, the three-dimensional structures of proteins determine their specific function, activity, and evolvability. Here, we show that a general protein language model augmented with protein structure backbone coordinates can guide evolution for diverse proteins without the need to model individual functional tasks. We also demonstrate that ESM-IF1, which was only trained on single-chain structures, can be extended to engineer protein complexes. Using this approach, we screened about 30 variants of two therapeutic clinical antibodies used to treat severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection. We achieved up to 25-fold improvement in neutralization and 37-fold improvement in affinity against antibody-escaped viral variants of concern BQ.1.1 and XBB.1.5, respectively. These findings highlight the advantage of integrating structural information to identify efficient protein evolution trajectories without requiring any task-specific training data.

Fluorescent tools for the standardized work in Gram-negative bacteria

Standardized and thoroughly characterized genetic tools are a prerequisite for studying cellular processes to ensure the reusability and consistency of experimental results. The discovery of fluorescent proteins (FPs) represents a milestone in the development of genetic reporters for monitoring transcription or protein localization in vivo. FPs have revolutionized our understanding of cellular dynamics by enabling the real-time visualization and tracking of biological processes. Despite these advancements, challenges remain in the appropriate use of FPs, specifically regarding their proper application, protein turnover dynamics, and the undesired disruption of cellular functions. Here, we systematically compared a comprehensive set of 15 FPs and assessed their performance in vivo by focusing on key parameters, such as signal over background ratios and protein stability rates, using the Gram-negative model organism Salmonella enterica as a representative host. We evaluated four protein degradation tags in both plasmid- and genome-based systems and our findings highlight the necessity of introducing degradation tags to analyze time-sensitive cellular processes. We demonstrate that the gain of dynamics mediated by the addition of degradation tags impacts the cell-to-cell heterogeneity of plasmid-based but not genome-based reporters. Finally, we probe the applicability of FPs for protein localization studies in living cells using standard and super-resolution fluorescence microscopy. In summary, our study underscores the importance of careful FP selection and paves the way for the development of improved genetic reporters to enhance the reproducibility and reliability of fluorescence-based research in Gram-negative bacteria and beyond.

Machine Learning and Deep Learning in Synthetic Biology: Key Architectures, Applications, and Challenges

Machine learning (ML), particularly deep learning (DL), has made rapid and substantial progress in synthetic biology in recent years. Biotechnological applications of biosystems, including pathways, enzymes, and whole cells, are being probed frequently with time. The intricacy and interconnectedness of biosystems make it challenging to design them with the desired properties. ML and DL have a synergy with synthetic biology. Synthetic biology can be employed to produce large data sets for training models (for instance, by utilizing DNA synthesis), and ML/DL models can be employed to inform design (for example, by generating new parts or advising unrivaled experiments to perform). This potential has recently been brought to light by research at the intersection of engineering biology and ML/DL through achievements like the design of novel biological components, best experimental design, automated analysis of microscopy data, protein structure prediction, and biomolecular implementations of ANNs (Artificial Neural Networks). I have divided this review into three sections. In the first section, I describe predictive potential and basics of ML along with myriad applications in synthetic biology, especially in engineering cells, activity of proteins, and metabolic pathways. In the second section, I describe fundamental DL architectures and their applications in synthetic biology. Finally, I describe different challenges causing hurdles in the progress of ML/DL and synthetic biology along with their solutions.

Machine learning-guided engineering of genetically encoded fluorescent calcium indicators

Here we used machine learning to engineer genetically encoded fluorescent indicators, protein-based sensors critical for real-time monitoring of biological activity. We used machine learning to predict the outcomes of sensor mutagenesis by analyzing established libraries that link sensor sequences to functions. Using the GCaMP calcium indicator as a scaffold, we developed an ensemble of three regression models trained on experimentally derived GCaMP mutation libraries. The trained ensemble performed an in silico functional screen on 1,423 novel, uncharacterized GCaMP variants. As a result, we identified the ensemble-derived GCaMP (eGCaMP) variants, eGCaMP and eGCaMP+, which achieve both faster kinetics and larger ∆F/F0 responses upon stimulation than previously published fast variants. Furthermore, we identified a combinatorial mutation with extraordinary dynamic range, eGCaMP2+, which outperforms the tested sixth-, seventh- and eighth-generation GCaMPs. These findings demonstrate the value of machine learning as a tool to facilitate the efficient engineering of proteins for desired biophysical characteristics.

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.

Machine-learning-guided Directed Evolution for AAV Capsid Engineering

Target gene delivery is crucial to gene therapy. Adeno-associated virus (AAV) has emerged as a primary gene therapy vector due to its broad host range, long-term expression, and low pathogenicity. However, AAV vectors have some limitations, such as immunogenicity and insufficient targeting. Designing or modifying capsids is a potential method of improving the efficacy of gene delivery, but hindered by weak biological basis of AAV, complexity of the capsids, and limitations of current screening methods. Artificial intelligence (AI), especially machine learning (ML), has great potential to accelerate and improve the optimization of capsid properties as well as decrease their development time and manufacturing costs. This review introduces the traditional methods of designing AAV capsids and the general steps of building a sequence-function ML model, highlights the applications of ML in the development workflow, and summarizes its advantages and challenges.

Inverse folding of protein complexes with a structure-informed language model enables unsupervised antibody evolution

Large language models trained on sequence information alone are capable of learning high level principles of protein design. However, beyond sequence, the three-dimensional structures of proteins determine their specific function, activity, and evolvability. Here we show that a general protein language model augmented with protein structure backbone coordinates and trained on the inverse folding problem can guide evolution for diverse proteins without needing to explicitly model individual functional tasks. We demonstrate inverse folding to be an effective unsupervised, structure-based sequence optimization strategy that also generalizes to multimeric complexes by implicitly learning features of binding and amino acid epistasis. Using this approach, we screened ~30 variants of two therapeutic clinical antibodies used to treat SARS-CoV-2 infection and achieved up to 26-fold improvement in neutralization and 37-fold improvement in affinity against antibody-escaped viral variants-of-concern BQ.1.1 and XBB.1.5, respectively. In addition to substantial overall improvements in protein function, we find inverse folding performs with leading experimental success rates among other reported machine learning-guided directed evolution methods, without requiring any task-specific training data.

A computational toolbox for the assembly yield of complex and heterogeneous structures

The self-assembly of complex structures from a set of non-identical building blocks is a hallmark of soft matter and biological systems, including protein complexes, colloidal clusters, and DNA-based assemblies. Predicting the dependence of the equilibrium assembly yield on the concentrations and interaction energies of building blocks is highly challenging, owing to the difficulty of computing the entropic contributions to the free energy of the many structures that compete with the ground state configuration. While these calculations yield well known results for spherically symmetric building blocks, they do not hold when the building blocks have internal rotational degrees of freedom. Here we present an approach for solving this problem that works with arbitrary building blocks, including proteins with known structure and complex colloidal building blocks. Our algorithm combines classical statistical mechanics with recently developed computational tools for automatic differentiation. Automatic differentiation allows efficient evaluation of equilibrium averages over configurations that would otherwise be intractable. We demonstrate the validity of our framework by comparison to molecular dynamics simulations of simple examples, and apply it to calculate the yield curves for known protein complexes and for the assembly of colloidal shells.

Artificial intelligence-aided protein engineering: from topological data analysis to deep protein language models

Protein engineering is an emerging field in biotechnology that has the potential to revolutionize various areas, such as antibody design, drug discovery, food security, ecology, and more. However, the mutational space involved is too vast to be handled through experimental means alone. Leveraging accumulative protein databases, machine learning (ML) models, particularly those based on natural language processing (NLP), have considerably expedited protein engineering. Moreover, advances in topological data analysis (TDA) and artificial intelligence-based protein structure prediction, such as AlphaFold2, have made more powerful structure-based ML-assisted protein engineering strategies possible. This review aims to offer a comprehensive, systematic, and indispensable set of methodological components, including TDA and NLP, for protein engineering and to facilitate their future development.

Machine Learning Enables Prediction of Pyrrolysyl-tRNA Synthetase Substrate Specificity

Knowledge about the substrate scope for a given enzyme is informative for elucidating biochemical pathways and also for expanding applications of the enzyme. However, no general methods are available to accurately predict the substrate specificity of an enzyme. Pyrrolysyl-tRNA synthetase (PylRS) is a powerful tool for incorporating various noncanonical amino acids (NCAAs) into proteins, which enabled us to probe, image, rationally engineer, and evolve protein structure and function. However, the incorporation of a new NCAA typically requires the selection of large libraries of PylRS with randomized mutations at active sites, and this process requires multiple rounds of selection for each new substrate. Therefore, a single aminoacyl-tRNA synthetase with broad substrate promiscuity is ideal to facilitate widespread applications of the genetic NCAA incorporation technique. Herein, machine learning models were developed to predict the substrate specificity of PylRS to accept novel NCAAs that could be incorporated into proteins by three PylRS mutants. The models were built from a training set of 285 unique enzyme-substrate pairs of three PylRS mutants including IFRS, BtaRS, and MFRS against 95 NCAAs. The best BaggingTree (BT) model was then used for virtually screening a NCAAs library containing 1474 phenylalanine, tyrosine, tryptophan, and alanine analogues, and 156 NCAAs were predicted to be accepted by at least one of the three PylRS mutants. Then, 27 NCAAs including 24 positive and 3 negative substrates were experimentally tested for their activities, and 20 of the 24 positive substrates showed weak or strong activity and were accepted by at least one PylRS mutant, among which 11 NCAAs were never reported to be incorporated into proteins before. Three negative substrates did not show any activity. Experimental results suggested that the BT model provides a three-class classification accuracy of 0.69 and a binary classification accuracy of 0.86. This study expanded the substrate scope of three PylRS variants and provided a framework for developing machine learning models to predict substrate specificity of other PylRS variants.

Learning protein fitness landscapes with deep mutational scanning data from multiple sources

One of the key points of machine learning-assisted directed evolution (MLDE) is the accurate learning of the fitness landscape, a conceptual mapping from sequence variants to the desired function. Here, we describe a multi-protein training scheme that leverages the existing deep mutational scanning data from diverse proteins to aid in understanding the fitness landscape of a new protein. Proof-of-concept trials are designed to validate this training scheme in three aspects: random and positional extrapolation for single-variant effects, zero-shot fitness predictions for new proteins, and extrapolation for higher-order variant effects from single-variant effects. Moreover, our study identified previously overlooked strong baselines, and their unexpectedly good performance brings our attention to the pitfalls of MLDE. Overall, these results may improve our understanding of the association between different protein fitness profiles and shed light on developing better machine learning-assisted approaches to the directed evolution of proteins. A record of this paper's transparent peer review process is included in the supplemental information.

Recent research advances on non-linear phenomena in various biosystems

All biological phenomena can be classified as open, dissipative and non-linear. Moreover, the most typical phenomena are associated with non-linearity, dissipation and openness in biological systems. In this review article, four research topics on non-linear biosystems are described to show the examples from various biological systems. First, membrane dynamics of a lipid bilayer for the cell membrane is described. Since the cell membrane separates the inside of the cell from the outside, self-organizing systems that form spatial patterns on membranes often depend on non-linear dynamics. Second, various data banks based on recent genomics analysis supply the data including vast functional proteins from many organisms and their variable species. Since the proteins existing in nature are only a very small part of the space represented by amino acid sequence, success of mutagenesis-based molecular evolution approach crucially depends on preparing a library with high enrichment of functional proteins. Third, photosynthetic organisms depend on ambient light, the regular and irregular changes of which have a significant impact on photosynthetic processes. The light-driven process proceeds through many redox couples in the cyanobacteria constituting chain of redox reactions. Forth topics focuses on a vertebrate model, the zebrafish, which can help to understand, predict and control the chaos of complex biological systems. In particular, during early developmental stages, developmental differentiation occurs dynamically from a fertilized egg to divided and mature cells. These exciting fields of complexity, chaos, and non-linear science have experienced impressive growth in recent decades. Finally, future directions for non-liner biosystems are presented.

Role of distal sites in enzyme engineering

The limitations associated with natural enzyme catalysis have triggered the rise of the field of protein engineering. Traditional rational design was based on the analysis of protein structural information and catalytic mechanisms to identify key active sites or ligand binding sites to reshape the substrate pocket. The role and significance of functional sites in the active center have been studied extensively. With a deeper understanding of the structure-catalysis relationship map, the entire protein molecule can be filled with residues that play a substantial role in its structure and function. However, the catalytic mechanism underlying distal mutations remains unclear. The aim of this review was to highlight the criticality of the distal site in enzyme engineering based on the following three aspects: What can distal mutations exert on function from mutability landscape? How do distal sites influence enzyme function? How to predict and design distal mutations? This review provides insights into the catalytic mechanism of enzymes from the global interaction network, knowledge from sequence-structure-dynamics-function relationships, and strategies for distal mutation-based protein engineering.

Engineering the Substrate Specificity of Toluene Degrading Enzyme XylM Using Biosensor XylS and Machine Learning

Enzyme engineering using machine learning has been developed in recent years. However, to obtain a large amount of data on enzyme activities for training data, it is necessary to develop a high-throughput and accurate method for evaluating enzyme activities. Here, we examined whether a biosensor-based enzyme engineering method can be applied to machine learning. As a model experiment, we aimed to modify the substrate specificity of XylM, a rate-determining enzyme in a multistep oxidation reaction catalyzed by XylMABC in Pseudomonas putida. XylMABC naturally converts toluene and xylene to benzoic acid and toluic acid, respectively. We aimed to engineer XylM to improve its conversion efficiency to a non-native substrate, 2,6-xylenol. Wild-type XylMABC slightly converted 2,6-xylenol to 3-methylsalicylic acid, which is the ligand of the transcriptional regulator XylS in P. putida. By locating a fluorescent protein gene under the control of the Pm promoter to which XylS binds, a XylS-producing Escherichia coli strain showed higher fluorescence intensity in a 3-methylsalicylic acid concentration-dependent manner. We evaluated the 3-methylsalicylic acid productivity of XylM variants using the fluorescence intensity of the sensor strain as an indicator. The obtained data provided the training data for machine learning for the directed evolution of XylM. Two cycles of machine learning-assisted directed evolution resulted in the acquisition of XylM-D140E-V144K-F243L-N244S with 15 times higher productivity than wild-type XylM. These results demonstrate that an indirect enzyme activity evaluation method using biosensors is sufficiently quantitative and high-throughput to be used as training data for machine learning. The findings expand the versatility of machine learning in enzyme engineering.

Discrimination of psychrophilic enzymes using machine learning algorithms with amino acid composition descriptor

Introduction

Psychrophilic enzymes are a class of macromolecules with high catalytic activity at low temperatures. Cold-active enzymes possessing eco-friendly and cost-effective properties, are of huge potential application in detergent, textiles, environmental remediation, pharmaceutical as well as food industry. Compared with the time-consuming and labor-intensive experiments, computational modeling especially the machine learning (ML) algorithm is a high-throughput screening tool to identify psychrophilic enzymes efficiently.

Methods

In this study, the influence of 4 ML methods (support vector machines, K-nearest neighbor, random forest, and naïve Bayes), and three descriptors, i.e., amino acid composition (AAC), dipeptide combinations (DPC), and AAC + DPC on the model performance were systematically analyzed.

Results and discussion

Among the 4 ML methods, the support vector machine model based on the AAC descriptor using 5-fold cross-validation achieved the best prediction accuracy with 80.6%. The AAC outperformed than the DPC and AAC + DPC descriptors regardless of the ML methods used. In addition, amino acid frequencies between psychrophilic and non-psychrophilic proteins revealed that higher frequencies of Ala, Gly, Ser, and Thr, and lower frequencies of Glu, Lys, Arg, Ile,Val, and Leu could be related to the protein psychrophilicity. Further, ternary models were also developed that could classify psychrophilic, mesophilic, and thermophilic proteins effectively. The predictive accuracy of the ternary classification model using AAC descriptor via the support vector machine algorithm was 75.8%. These findings would enhance our insight into the cold-adaption mechanisms of psychrophilic proteins and aid in the design of engineered cold-active enzymes. Moreover, the proposed model could be used as a screening tool to identify novel cold-adapted proteins.

Transcription factor-based biosensors for screening and dynamic regulation

Advances in synthetic biology and genetic engineering are bringing into the spotlight a wide range of bio-based applications that demand better sensing and control of biological behaviours. Transcription factor (TF)-based biosensors are promising tools that can be used to detect several types of chemical compounds and elicit a response according to the desired application. However, the wider use of this type of device is still hindered by several challenges, which can be addressed by increasing the current metabolite-activated transcription factor knowledge base, developing better methods to identify new transcription factors, and improving the overall workflow for the design of novel biosensor circuits. These improvements are particularly important in the bioproduction field, where researchers need better biosensor-based approaches for screening production-strains and precise dynamic regulation strategies. In this work, we summarize what is currently known about transcription factor-based biosensors, discuss recent experimental and computational approaches targeted at their modification and improvement, and suggest possible future research directions based on two applications: bioproduction screening and dynamic regulation of genetic circuits.

Automating the High-Throughput Screening of Protein-Based Optical Indicators and Actuators

Over the last 25 years, protein engineers have developed an impressive collection of optical tools to interface with biological systems: indicators to eavesdrop on cellular activity and actuators to poke and prod native processes. To reach the performance level required for their downstream applications, protein-based tools are usually sculpted by iterative rounds of mutagenesis. In each round, libraries of variants are made and evaluated, and the most promising hits are then retrieved, sequenced, and further characterized. Early efforts to engineer protein-based optical tools were largely manual, suffering from low throughput, human error, and tedium. Here, we describe approaches to automating the screening of libraries generated as colonies on agar, multiwell plates, and pooled populations of single-cell variants. We also briefly discuss emerging approaches for screening, including cell-free systems and machine learning.

Improved Bst DNA Polymerase Variants Derived via a Machine Learning Approach

The DNA polymerase I from Geobacillus stearothermophilus (also known as Bst DNAP) is widely used in isothermal amplification reactions, where its strand displacement ability is prized. More robust versions of this enzyme should be enabled for diagnostic applications, especially for carrying out higher temperature reactions that might proceed more quickly. To this end, we appended a short fusion domain from the actin-binding protein villin that improved both stability and purification of the enzyme. In parallel, we have developed a machine learning algorithm that assesses the relative fit of individual amino acids to their chemical microenvironments at any position in a protein and applied this algorithm to predict sequence substitutions in Bst DNAP. The top predicted variants had greatly improved thermotolerance (heating prior to assay), and upon combination, the mutations showed additive thermostability, with denaturation temperatures up to 2.5 °C higher than the parental enzyme. The increased thermostability of the enzyme allowed faster loop-mediated isothermal amplification assays to be carried out at 73 °C, where both Bst DNAP and its improved commercial counterpart Bst 2.0 are inactivated. Overall, this is one of the first examples of the application of machine learning approaches to the thermostabilization of an enzyme.

Selection of target-binding proteins from the information of weakly enriched phage display libraries by deep sequencing and machine learning

Despite the advances in surface-display systems for directed evolution, variants with high affinity are not always enriched due to undesirable biases that increase target-unrelated variants during biopanning. Here, our goal was to design a library containing improved variants from the information of the "weakly enriched" library where functional variants were weakly enriched. Deep sequencing for the previous biopanning result, where no functional antibody mimetics were experimentally identified, revealed that weak enrichment was partly due to undesirable biases during phage infection and amplification steps. The clustering analysis of the deep sequencing data from appropriate steps revealed no distinct sequence patterns, but a Bayesian machine learning model trained with the selected deep sequencing data supplied nine clusters with distinct sequence patterns. Phage libraries were designed on the basis of the sequence patterns identified, and four improved variants with target-specific affinity (EC₅₀ = 80-277 nM) were identified by biopanning. The selection and use of deep sequencing data without undesirable bias enabled us to extract the information on prospective variants. In summary, the use of appropriate deep sequencing data and machine learning with the sequence data has the possibility of finding sequence space where functional variants are enriched.

Monte Carlo Thompson sampling-guided design for antibody engineering

Antibodies are one of the predominant treatment modalities for various diseases. To improve the characteristics of a lead antibody, such as antigen-binding affinity and stability, we conducted comprehensive substitutions and exhaustively explored their sequence space. However, it is practically unfeasible to evaluate all possible combinations of mutations owing to combinatorial explosion when multiple amino acid residues are incorporated. It was recently reported that a machine-learning guided protein engineering approach such as Thompson sampling (TS) has been used to efficiently explore sequence space in the framework of Bayesian optimization. For TS, over-exploration occurs when the initial data are biasedly distributed in the vicinity of the lead antibody. We handle a large-scale virtual library that includes numerous mutations. When the number of experiments is limited, this over-exploration causes a serious issue. Thus, we conducted Monte Carlo Thompson sampling (MTS) to balance the exploration-exploitation trade-off by defining the posterior distribution via the Monte Carlo method and compared its performance with TS in antibody engineering. Our results demonstrated that MTS largely outperforms TS in discovering desirable candidates at an earlier round when over-exploration occurs on TS. Thus, the MTS method is a powerful technique for efficiently discovering antibodies with desired characteristics when the number of rounds is limited.

Synthetic Biology Meets Machine Learning

This chapter outlines the myriad applications of machine learning (ML) in synthetic biology, specifically in engineering cell and protein activity, and metabolic pathways. Though by no means comprehensive, the chapter highlights several prominent computational tools applied in the field and their potential use cases. The examples detailed reinforce how ML algorithms can enhance synthetic biology research by providing data-driven insights into the behavior of living systems, even without detailed knowledge of their underlying mechanisms. By doing so, ML promises to increase the efficiency of research projects by modeling hypotheses in silico that can then be tested through experiments. While challenges related to training dataset generation and computational costs remain, ongoing improvements in ML tools are paving the way for smarter and more streamlined synthetic biology workflows that can be readily employed to address grand challenges across manufacturing, medicine, engineering, agriculture, and beyond.

Logistic Regression-Guided Identification of Cofactor Specificity-Contributing Residues in Enzyme with Sequence Datasets Partitioned by Catalytic Properties

Changing the substrate/cofactor specificity of an enzyme requires multiple mutations at spatially adjacent positions around the substrate pocket. However, this is challenging when solely based on crystal structure information because enzymes undergo dynamic conformational changes during the reaction process. Herein, we proposed a method for estimating the contribution of each amino acid residue to substrate specificity by deploying a phylogenetic analysis with logistic regression. Since this method can estimate the candidate amino acids for mutation by ranking, it is readable and can be used in protein engineering. We demonstrated our concept using redox cofactor conversion of the Escherichia coli malic enzyme as a model, which still lacks crystal structure elucidation. The use of logistic regression with amino acid sequences classified by cofactor specificity showed that the NADP⁺-dependent malic enzyme completely switched cofactor specificity to NAD⁺ dependence without the need for a practical screening step. The model showed that surrounding residues made a greater contribution to cofactor specificity than those in the interior of the substrate pocket. These residues might be difficult to identify from crystal structure observations. We show that a highly accurate and inferential machine learning model was obtained using amino acid sequences of structurally homologous and functionally distinct enzymes as input data.

A dataset comprised of binding interactions for 104,972 antibodies against a SARS-CoV-2 peptide

The dataset presented here contains quantitative binding scores of scFv-format antibodies against a SARS-CoV-2 target peptide collected via an AlphaSeq assay that can be used in the development and benchmarking of machine learning models. Starting from three seed sequences identified from a phage display campaign using a human naïve library, four sets of 29,900 antibodies were designed in silico by creating all k = 1 mutations and random k = 2 and k = 3 mutations throughout the complementary-determining regions (CDRs). Of the 119,600 designs, 104,972 were successfully built in to the AlphaSeq library and target binding was subsequently measured with 71,384 designs resulting in a predicted affinity value for at least one of the triplicate measurements. Data include antibodies with predicted affinity measurements ranging from 37 pM to 22 mM. To our knowledge, this dataset is the largest, publicly available dataset that contains antibody sequences, antigen sequence and quantitative measurements of binding scores and provides an opportunity to serve as a benchmark to evaluate antibody-specific representation models for machine learning.

Measurement(s)	Antibody Binding
Technology Type(s)	AlphaSeq
Factor Type(s)	Antibody sequence
Sample Characteristic - Organism	Homo sapiens

CLADE 2.0: Evolution-Driven Cluster Learning-Assisted Directed Evolution

Directed evolution, a revolutionary biotechnology in protein engineering, optimizes protein fitness by searching an astronomical mutational space via expensive experiments. The cluster learning-assisted directed evolution (CLADE) efficiently explores the mutational space via a combination of unsupervised hierarchical clustering and supervised learning. However, the initial-stage sampling in CLADE treats all clusters equally despite many clusters containing a large portion of non-functional mutations. Recent statistical and deep learning tools enable evolutionary density modeling to access protein fitness in an unsupervised manner. In this work, we construct an ensemble of multiple evolutionary scores to guide the initial sampling in CLADE. The resulting evolutionary score-enhanced CLADE, called CLADE 2.0, efficiently selects a training set within a small informative space using the evolution-driven clustering sampling. CLADE 2.0 is validated by using two benchmark libraries both having 160,000 sequences from four-site mutational combinations. Extensive computational experiments and comparisons with existing cutting-edge methods indicate that CLADE 2.0 is a new state-of-art tool for machine learning-assisted directed evolution.

Catalysis and Electron Transfer in De Novo Designed Metalloproteins

One of the hallmark advances in our understanding of metalloprotein function is showcased in our ability to design new, non-native, catalytically active protein scaffolds. This review highlights progress and milestone achievements in the field of de novo metalloprotein design focused on reports from the past decade with special emphasis on de novo designs couched within common subfields of bioinorganic study: heme binding proteins, monometal- and dimetal-containing catalytic sites, and metal-containing electron transfer sites. Within each subfield, we highlight several of what we have identified as significant and important contributions to either our understanding of that subfield or de novo metalloprotein design as a discipline. These reports are placed in context both historically and scientifically. General suggestions for future directions that we feel will be important to advance our understanding or accelerate discovery are discussed.

Computer-Aided Design of α-L-Rhamnosidase to Increase the Synthesis Efficiency of Icariside I

Icariside I, the glycosylation product of icaritin, is a novel effective anti-cancer agent with immunological anti-tumor activity. However, very limited natural icariside I content hinders its direct extraction from plants. Therefore, we employed a computer-aided protein design strategy to improve the catalytic efficiency and substrate specificity of the α-L-rhamnosidase from Thermotoga petrophila DSM 13995, to provide a highly-efficient preparation method. Several beneficial mutants were obtained by expanding the active cavity. The catalytic efficiencies of all mutants were improved 16-200-fold compared with the wild-type TpeRha. The double-point mutant DH was the best mutant and showed the highest catalytic efficiency (k cat /K M : 193.52 s-1 M-1) against icariin, which was a 209.76-fold increase compared with the wild-type TpeRha. Besides, the single-point mutant H570A showed higher substrate specificity than that of the wild-type TpeRha in hydrolysis of different substrates. This study provides enzyme design strategies and principles for the hydrolysis of rhamnosyl natural products.

Experimental Characterization of In Silico Red-Shift-Predicted iLOVL470T/Q489K and iLOVV392K/F410V/A426S Mutants

iLOV is a flavin mononucleotide-binding fluorescent protein used for in vivo cellular imaging similar to the green fluorescent protein. To expand the range of applications of iLOV, spectrally tuned red-shifted variants are desirable to reduce phototoxicity and allow for better tissue penetration. In this report, we experimentally tested two iLOV mutants, iLOV^L470T/Q489K and iLOV^{V392K/F410V/A426S}, which were previously computationally proposed by (KhrenovaJ. Phys. Chem. B2017, 121 ( (43), ), pp 10018-10025) to have red-shifted excitation and emission spectra. While iLOV^L470T/Q489K is about 20% brighter compared to the WT in vitro, it exhibits a blue shift in contrast to quantum mechanics/molecular mechanics (QM/MM) predictions. Additional optical characterization of an iLOV^V392K mutant revealed that V392 is essential for cofactor binding and, accordingly, variants with V392K mutation are unable to bind to FMN. iLOV^L470T/Q489K and iLOV^{V392K/F410V/A426S} are expressed at low levels and have no detectable fluorescence in living cells, preventing their utilization in imaging applications.

Energetic Basis and Design of Enzyme Function Demonstrated Using GFP, an Excited-State Enzyme

The past decades have witnessed an explosion of de novo protein designs with a remarkable range of scaffolds. It remains challenging, however, to design catalytic functions that are competitive with naturally occurring counterparts as well as biomimetic or nonbiological catalysts. Although directed evolution often offers efficient solutions, the fitness landscape remains opaque. Green fluorescent protein (GFP), which has revolutionized biological imaging and assays, is one of the most redesigned proteins. While not an enzyme in the conventional sense, GFPs feature competing excited-state decay pathways with the same steric and electrostatic origins as conventional ground-state catalysts, and they exert exquisite control over multiple reaction outcomes through the same principles. Thus, GFP is an "excited-state enzyme". Herein we show that rationally designed mutants and hybrids that contain environmental mutations and substituted chromophores provide the basis for a quantitative model and prediction that describes the influence of sterics and electrostatics on excited-state catalysis of GFPs. As both perturbations can selectively bias photoisomerization pathways, GFPs with fluorescence quantum yields (FQYs) and photoswitching characteristics tailored for specific applications could be predicted and then demonstrated. The underlying energetic landscape, readily accessible via spectroscopy for GFPs, offers an important missing link in the design of protein function that is generalizable to catalyst design.

In silico evolution of nucleic acid-binding proteins from a nonfunctional scaffold

Directed evolution emulates the process of natural selection to produce proteins with improved or altered functions. These approaches have proven to be very powerful but are technically challenging and particularly time and resource intensive. To bypass these limitations, we constructed a system to perform the entire process of directed evolution in silico. We employed iterative computational cycles of mutation and evaluation to predict mutations that confer high-affinity binding activities for DNA and RNA to an initial de novo designed protein with no inherent function. Beneficial mutations revealed modes of nucleic acid recognition not previously observed in natural proteins, highlighting the ability of computational directed evolution to access new molecular functions. Furthermore, the process by which new functions were obtained closely resembles natural evolution and can provide insights into the contributions of mutation rate, population size and selective pressure on functionalization of macromolecules in nature.

Algorithm-aided engineering of aliphatic halogenase WelO5* for the asymmetric late-stage functionalization of soraphens

Late-stage functionalization of natural products offers an elegant route to create novel entities in a relevant biological target space. In this context, enzymes capable of halogenating sp³ carbons with high stereo- and regiocontrol under benign conditions have attracted particular attention. Enabled by a combination of smart library design and machine learning, we engineer the iron/α-ketoglutarate dependent halogenase WelO5* for the late-stage functionalization of the complex and chemically difficult to derivatize macrolides soraphen A and C, potent anti-fungal agents. While the wild type enzyme WelO5* does not accept the macrolide substrates, our engineering strategy leads to active halogenase variants and improves upon their apparent k_cat and total turnover number by more than 90-fold and 300-fold, respectively. Notably, our machine-learning guided engineering approach is capable of predicting more active variants and allows us to switch the regio-selectivity of the halogenases facilitating the targeted analysis of the derivatized macrolides’ structure-function activity in biological assays.

The late-stage functionalization of unactivated carbon–hydrogen bonds is a difficult but important task, which has been met with promising but limited success through synthetic organic chemistry. Here the authors use machine learning to engineer WelO5* halogenase variants, which led to regioselective chlorination of inert C–H bonds on a representative polyketide that is a non-natural substrate for the enzyme.

Thermogenetics: Applications come of age

Temperature is a ubiquitous physical cue that is non-invasive, penetrative and easy to apply. In the growing field of thermogenetics, through beneficial repurposing of natural thermosensing mechanisms, synthetic biology is bringing new opportunities to design and build robust temperature-sensitive (TS) sensors which forms a thermogenetic toolbox of well characterised biological parts. Recent advancements in technological platforms available have expedited the discovery of novel or de novo thermosensors which are increasingly deployed in many practical temperature-dependent biomedical, industrial and biosafety applications. In all, the review aims to convey both the exhilarating recent technological developments underlying the advancement of thermosensors and the exciting opportunities the nascent thermogenetic field holds for biomedical and biotechnology applications.

Machine Learning-driven Protein Library Design: A Path Toward Smarter Libraries

Proteins are small yet valuable biomolecules that play a versatile role in therapeutics and diagnostics. The intricate sequence-structure-function paradigm in the realm of proteins opens the possibility for directly mapping amino acid sequence to function. However, the rugged nature of the protein fitness landscape and an astronomical number of possible mutations even for small proteins make navigating this system a daunting task. Moreover, the scarcity of functional proteins and the ease with which deleterious mutations are introduced, due to complex epistatic relationships, compound the existing challenges. This highlights the need for auxiliary tools in current techniques such as rational design and directed evolution. To that end, the state-of-the-art machine learning can offer time and cost efficiency in finding high fitness proteins, circumventing unnecessary wet-lab experiments. In the context of improving library design, machine learning provides valuable insights via its unique features such as high adaptation to complex systems, multi-tasking, and parallelism, and the ability to capture hidden trends in input data. Finally, both the advancements in computational resources and the rapidly increasing number of sequences in protein databases will allow more promising and detailed insights delivered from machine learning to protein library design. In this chapter, fundamental concepts and a method for machine learning-driven library design leveraging deep sequencing datasets will be discussed. We elaborate on (1) basic knowledge about machine learning algorithms, (2) the benefit of machine learning in library design, and (3) methodology for implementing machine learning in library design.

Learning Strategies in Protein Directed Evolution

Synthetic biology is a fast-evolving research field that combines biology and engineering principles to develop new biological systems for medical, pharmacological, and industrial applications. Synthetic biologists use iterative "design, build, test, and learn" cycles to efficiently engineer genetic systems that are reliable, reproducible, and predictable. Protein engineering by directed evolution can benefit from such a systematic engineering approach for various reasons. Learning can be carried out before starting, throughout or after finalizing a directed evolution project. Computational tools, bioinformatics, and scanning mutagenesis methods can be excellent starting points, while molecular dynamics simulations and other strategies can guide engineering efforts. Similarly, studying protein intermediates along evolutionary pathways offers fascinating insights into the molecular mechanisms shaped by evolution. The learning step of the cycle is not only crucial for proteins or enzymes that are not suitable for high-throughput screening or selection systems, but it is also valuable for any platform that can generate a large amount of data that can be aided by machine learning algorithms. The main challenge in protein engineering is to predict the effect of a single mutation on one functional parameter-to say nothing of several mutations on multiple parameters. This is largely due to nonadditive mutational interactions, known as epistatic effects-beneficial mutations present in a genetic background may not be beneficial in another genetic background. In this work, we provide an overview of experimental and computational strategies that can guide the user to learn protein function at different stages in a directed evolution project. We also discuss how epistatic effects can influence the success of directed evolution projects. Since machine learning is gaining momentum in protein engineering and the field is becoming more interdisciplinary thanks to collaboration between mathematicians, computational scientists, engineers, molecular biologists, and chemists, we provide a general workflow that familiarizes nonexperts with the basic concepts, dataset requirements, learning approaches, model capabilities and performance metrics of this intriguing area. Finally, we also provide some practical recommendations on how machine learning can harness epistatic effects for engineering proteins in an "outside-the-box" way.

SpeedyGenesXL: an Automated, High-Throughput Platform for the Preparation of Bespoke Ultralarge Variant Libraries for Directed Evolution

Directed evolution of proteins is a highly effective strategy for tailoring biocatalysts to a particular application, and is capable of engineering improvements such as k_cat, thermostability and organic solvent tolerance. It is recognized that large and systematic libraries are required to navigate a protein's vast and rugged sequence landscape effectively, yet their preparation is nontrivial and commercial libraries are extremely costly. To address this, we have developed SpeedyGenesXL, an automated, high-throughput platform for the production of wild-type genes, Boolean OR, combinatorial, or combinatorial-OR-type libraries based on the SpeedyGenes methodology. Together this offers a flexible platform for library synthesis, capable of generating many different bespoke, diverse libraries simultaneously.

Cluster learning-assisted directed evolution

Directed evolution, a strategy for protein engineering, optimizes protein properties (i.e., fitness) by expensive and time-consuming screening or selection of large mutational sequence space. Machine learning-assisted directed evolution (MLDE), which screens sequence properties in silico, can accelerate the optimization and reduce the experimental burden. This work introduces a MLDE framework, cluster learning-assisted directed evolution (CLADE), that combines hierarchical unsupervised clustering sampling and supervised learning to guide protein engineering. The clustering sampling selectively picks and screens variants in targeted subspaces, which guides the subsequent generation of diverse training sets. In the last stage, accurate predictions via supervised learning models improve final outcomes. By sequentially screening 480 sequences out of 160,000 in a four-site combinatorial library with five equal experimental batches, CLADE achieves the global maximal fitness hit rate up to 91.0% and 34.0% for GB1 and PhoQ datasets, respectively, improved from 18.6% and 7.2% obtained by random-sampling-based MLDE.

Neural networks to learn protein sequence-function relationships from deep mutational scanning data

The mapping from protein sequence to function is highly complex, making it challenging to predict how sequence changes will affect a protein's behavior and properties. We present a supervised deep learning framework to learn the sequence-function mapping from deep mutational scanning data and make predictions for new, uncharacterized sequence variants. We test multiple neural network architectures, including a graph convolutional network that incorporates protein structure, to explore how a network's internal representation affects its ability to learn the sequence-function mapping. Our supervised learning approach displays superior performance over physics-based and unsupervised prediction methods. We find that networks that capture nonlinear interactions and share parameters across sequence positions are important for learning the relationship between sequence and function. Further analysis of the trained models reveals the networks' ability to learn biologically meaningful information about protein structure and mechanism. Finally, we demonstrate the models' ability to navigate sequence space and design new proteins beyond the training set. We applied the protein G B1 domain (GB1) models to design a sequence that binds to immunoglobulin G with substantially higher affinity than wild-type GB1.