Large Sequence Space Research Articles

Rapid affinity optimization of an anti-TREM2 clinical lead antibody by cross-lineage immune repertoire mining

We describe a process for rapid antibody affinity optimization by repertoire mining to identify clones across B cell clonal lineages based on convergent immune responses where antigen-specific clones with the same heavy (VH) and light chain germline segment pairs, or parallel lineages, bind a single epitope on the antigen. We use this convergence framework to mine unique and distinct VH lineages from rat anti-triggering receptor on myeloid cells 2 (TREM2) antibody repertoire datasets with high diversity in the third complementarity-determining loop region (CDR H3) to further affinity-optimize a high-affinity agonistic anti-TREM2 antibody while retaining critical functional properties. Structural analyses confirm a nearly identical binding mode of anti-TREM2 variants with subtle but significant structural differences in the binding interface. Parallel lineage repertoire mining is uniquely tailored to rationally explore the large CDR H3 sequence space in antibody repertoires and can be easily and generally applied to antibodies discovered in vivo.

Massively parallel analysis of single-molecule dynamics on next-generation sequencing chips.

Single-molecule techniques are ideally poised to characterize complex dynamics but are typically limited to investigating a small number of different samples. However, a large sequence or chemical space often needs to be explored to derive a comprehensive understanding of complex biological processes. Here we describe multiplexed single-molecule characterization at the library scale (MUSCLE), a method that combines single-molecule fluorescence microscopy with next-generation sequencing to enable highly multiplexed observations of complex dynamics. We comprehensively profiled the sequence dependence of DNA hairpin properties and Cas9-induced target DNA unwinding-rewinding dynamics. The ability to explore a large sequence space for Cas9 allowed us to identify a number of target sequences with unexpected behaviors. We envision that MUSCLE will enable the mechanistic exploration of many fundamental biological processes.

Predicting monoclonal antibody binding sequences from a sparse sampling of all possible sequences

Previous work has shown that binding of target proteins to a sparse, unbiased sample of all possible peptide sequences is sufficient to train a machine learning model that can then predict, with statistically high accuracy, target binding to any possible peptide sequence of similar length. Here, highly sequence-specific molecular recognition is explored by measuring binding of 8 monoclonal antibodies (mAbs) with specific linear cognate epitopes to an array containing 121,715 near-random sequences about 10 residues in length. Network models trained on resulting sequence-binding values are used to predict the binding of each mAb to its cognate sequence and to an in silico generated one million random sequences. The model always ranks the binding of the cognate sequence in the top 100 sequences, and for 6 of the 8 mAbs, the cognate sequence ranks in the top ten. Practically, this approach has potential utility in selecting highly specific mAbs for therapeutics or diagnostics. More fundamentally, this demonstrates that very sparse random sampling of a large amino acid sequence spaces is sufficient to generate comprehensive models predictive of highly specific molecular recognition.

An Atom-Precise Understanding of DNA-Stabilized Silver Nanoclusters.

ConspectusDNA-stabilized silver nanoclusters (AgN-DNAs) are sequence-encoded fluorophores. Like other noble metal nanoclusters, the optical properties of AgN-DNAs are dictated by their atomically precise sizes and shapes. What makes AgN-DNAs unique is that nanocluster size and shape are controlled by nucleobase sequence of the templating DNA oligomer. By choice of DNA sequence, it is possible to synthesize a wide range of AgN-DNAs with diverse emission colors and other intriguing photophysical properties. AgN-DNAs hold significant potential as "programmable" emitters for biological imaging due to their combination of small molecular-like sizes, bright and sequence-tuned fluorescence, low toxicities, and cost-effective synthesis. In particular, the potential to extend AgN-DNAs into the second near-infrared region (NIR-II) is promising for deep tissue imaging, which is a major area of interest for advancing biomedical imaging. Achieving this goal requires a deep understanding of the structure-property relationships that govern AgN-DNAs in order to design AgN-DNA emitters with sizes and geometries that support NIR-II emission.In recent years, major advances have been made in understanding the structure and composition of AgN-DNAs, enabling new insights into the correlation of nanocluster structure and photophysical properties. These advances have hinged on combined innovations in mass characterization and crystallography of compositionally pure AgN-DNAs, together with combinatorial experiments and machine learning-guided design. A combined approach is essential due to the major challenge of growing suitable AgN-DNA crystals for diffraction and to the labor-intensive nature of preparing and solving the molecular formulas of atomically precise AgN-DNAs by mass spectrometry. These approaches alone are not feasibly scaled to explore the large sequence space of DNA oligomer templates for AgN-DNAs.This account describes recent fundamental advances in AgN-DNA science that have been enabled by high throughput synthesis and fluorimetry together with detailed analytical studies of purified AgN-DNAs. First, short introductions to nanocluster chemistry and AgN-DNA basics are presented. Then, we review recent large-scale studies that have screened thousands of DNA templates for AgN-DNAs, leading to discovery of distinct classes of these emitters with unique cluster core compositions and ligand chemistries. In particular, the discovery of a new class of chloride-stabilized AgN-DNAs enabled the first ab initio calculations of AgN-DNA electronic structure and present new approaches to stabilize these emitters in biologically relevant conditions. Near-infrared (NIR) emissive AgN-DNAs are also found to exhibit diverse structures and properties. Finally, we conclude by highlighting recent proof-of-principle demonstrations of NIR AgN-DNAs for targeted fluorescence imaging. Continued efforts may future push AgN-DNAs into the tissue transparency window for fluorescence imaging in the NIR-II tissue transparency window.

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.

Selection pressures on evolution of ribonuclease H explored with rigorous free–energy–based design

Understanding natural protein evolution and designing novel proteins are motivating interest in development of high-throughput methods to explore large sequence spaces. In this work, we demonstrate the application of multisite λ dynamics (MSλD), a rigorous free energy simulation method, and chemical denaturation experiments to quantify evolutionary selection pressure from sequence-stability relationships and to address questions of design. This study examines a mesophilic phylogenetic clade of ribonuclease H (RNase H), furthering its extensive characterization in earlier studies, focusing on E. coli RNase H (ecRNH) and a more stable consensus sequence (AncCcons) differing at 15 positions. The stabilities of 32,768 chimeras between these two sequences were computed using the MSλD framework. The most stable and least stable chimeras were predicted and tested along with several other sequences, revealing a designed chimera with approximately the same stability increase as AncCcons, but requiring only half the mutations. Comparing the computed stabilities with experiment for 12 sequences reveals a Pearson correlation of 0.86 and root mean squared error of 1.18 kcal/mol, an unprecedented level of accuracy well beyond less rigorous computational design methods. We then quantified selection pressure using a simple evolutionary model in which sequences are selected according to the Boltzmann factor of their stability. Selection temperatures from 110 to 168 K are estimated in three ways by comparing experimental and computational results to evolutionary models. These estimates indicate selection pressure is high, which has implications for evolutionary dynamics and for the accuracy required for design, and suggests accurate high-throughput computational methods like MSλD may enable more effective protein design.

Hold out the genome: a roadmap to solving the cis-regulatory code.

Gene expression is regulated by transcription factors that work together to read cis-regulatory DNA sequences. The 'cis-regulatory code' - how cells interpret DNA sequences to determine when, where and how much genes should be expressed - has proven to be exceedingly complex. Recently, advances in the scale and resolution of functional genomics assays and machine learning have enabled substantial progress towards deciphering this code. However, the cis-regulatory code will probably never be solved if models are trained only on genomic sequences; regions of homology can easily lead to overestimation of predictive performance, and our genome is too short and has insufficient sequence diversity to learn all relevant parameters. Fortunately, randomly synthesized DNA sequences enable testing a far larger sequence space than exists in our genomes, and designed DNA sequences enable targeted queries to maximally improve the models. As the same biochemical principles are used to interpret DNA regardless of its source, models trained on these synthetic data can predict genomic activity, often better than genome-trained models. Here we provide an outlook on the field, and propose a roadmap towards solving the cis-regulatory code by a combination of machine learning and massively parallel assays using synthetic DNA.

Profiling DNA Ligase Substrate Specificity with a Pacific Biosciences Single-Molecule Real-Time Sequencing Assay.

DNA ligases catalyze the joining of breaks in nucleic acid backbones and are essential enzymes for in vivo genome replication and repair across all domains of life. These enzymes are also critically important to in vitro manipulation of DNA in applications such as cloning, sequencing, and molecular diagnostics. DNA ligases generally catalyze the formation of a phosphodiester bond between an adjacent 5'-phosphate and 3'-hydroxyl in DNA, but they exhibit different substrate structure preferences, sequence-dependent biases in reaction kinetics, and variable tolerance for mismatched base pairs. Information on substrate structure and sequence specificity can inform both biological roles and molecular biology applications of these enzymes. Given the high complexity of DNA sequence space, testing DNA ligase substrate specificity on individual nucleic acid sequences in parallel rapidly becomes impractical when a large sequence space is investigated. Here, we describe methods for investigating DNA ligase sequence bias and mismatch discrimination using Pacific Biosciences Single-Molecule Real-Time (PacBio SMRT) sequencing technology. Through its rolling-circle amplification methodology, SMRT sequencing can give multiple reads of the same insert. This feature permits high-quality top- and bottom-strand consensus sequences to be determined while preserving information on top-bottom strand mismatches that can be obfuscated or lost when using other sequencing methods. Thus, PacBio SMRT sequencing is uniquely suited to measuring substrate bias and enzyme fidelity through multiplexing a diverse set of sequences in a single reaction. The protocols describe substrate synthesis, library preparation, and data analysis methods suitable for measuring fidelity and bias of DNA ligases. The methods are easily adapted to different nucleic acid substrate structures and can be used to characterize many enzymes under a variety of reaction conditions and sequence contexts in a rapid and high-throughput manner. © 2023 New England Biolabs and The Authors. Current Protocols published by Wiley Periodicals LLC. Basic Protocol 1: Preparation of overhang DNA substrates for ligation Basic Protocol 2: Preparation of ligation fidelity libraries Support Protocol 1: Preparation of ligation libraries for PacBio Sequel II sequencing Support Protocol 2: Loading and sequencing of a prepared library on the Sequel II instrument Basic Protocol 3: Computational processing of ligase fidelity sequencing data.

High-Throughput Screen Reveals the Structure-Activity Relationship of the Antimicrobial Lasso Peptide Ubonodin.

The Burkholderia cepacia complex (Bcc) is a group of bacteria including opportunistic human pathogens. Immunocompromised individuals and cystic fibrosis patients are especially vulnerable to serious infections by these bacteria, motivating the search for compounds with antimicrobial activity against the Bcc. Ubonodin is a lasso peptide with promising activity against Bcc species, working by inhibiting RNA polymerase in susceptible bacteria. We constructed a library of over 90 000 ubonodin variants with 2 amino acid substitutions and used a high-throughput screen and next-generation sequencing to examine the fitness of the entire library, generating the most comprehensive data set on lasso peptide activity so far. This screen revealed information regarding the structure-activity relationship of ubonodin over a large sequence space. Remarkably, the screen identified one variant with not only improved activity compared to wild-type ubonodin but also a submicromolar minimum inhibitory concentration (MIC) against a clinical isolate of the Bcc member Burkholderia cenocepacia. Ubonodin and several of the variants identified in this study had lower MICs against certain Bcc strains than those of many clinically approved antibiotics. Finally, the large library size enabled us to develop DeepLasso, a deep learning model that can predict the RNAP inhibitory activity of an ubonodin variant.

Directed Evolution of Laccase for Improved Thermal Stability Facilitated by Droplet-Based Microfluidic Screening System.

Laccases are attractive biocatalysts for industry due to their broad substrate spectrum, the use of oxygen as final electron acceptor, and water as the sole byproduct. Increasing efforts have been devoted to the engineering of laccases to improve their properties. The droplet-based microfluidic screening (DMFS) technology can accelerate the screening procedure and probe the large sequence space. In this study, a DMFS system including a heating step and picoinjection was used to sort large laccase libraries, yielding 12 variants with enhanced thermotolerance. All the obtained amino acid substitutions are distributed on the surface of the laccase. Interestingly, recombination of three identified substitutions of Asp to Asn on the surface resulted in the best variant M20, exhibiting 24.0-fold higher remaining activity at 58.8 °C and 1.9-3.4-fold higher remaining activity after incubation in organic solvents solution (20% (v/v) methanol and ethanol) and ionic liquid solution (20% (v/v) 1-ethyl-3-methylimidazolium ethyl sulfate) for 12 h. Furthermore, molecular dynamic simulations revealed that the recombination of the three beneficial substitutions, Asp98Asn, Asp474Asn, and Asp340Asn on the surface introduced more hydrogen bonds compared to the wild type, which made M20 more thermostable. This study highlighted the importance of the DMFS system for an efficient identification of beneficial long-distance amino acid substitutions.

From Trees to Clouds: PhageClouds for Fast Comparison of ∼640,000 Phage Genomic Sequences and Host-Centric Visualization Using Genomic Network Graphs.

Background: Fast and computationally efficient strategies are required to explore genomic relationships within an increasingly large and diverse phage sequence space. Here, we present PhageClouds, a novel approach using a graph database of phage genomic sequences and their intergenomic distances to explore the phage genomic sequence space.Methods: A total of 640,000 phage genomic sequences were retrieved from a variety of databases and public virome assemblies. Intergenomic distances were calculated with dashing, an alignment-free method suitable for handling massive data sets. These data were used to build a Neo4j® graph database.Results: PhageClouds supported the search of related phages among all complete phage genomes from GenBank for a single query phage in just 10 s. Moreover, PhageClouds expanded the number of closely related phage sequences detected for both finished and draft phage genomes, in comparison with searches exclusively targeting phage entries from GenBank.Conclusions: PhageClouds is a novel resource that will facilitate the analysis of phage genomic sequences and the characterization of assembled phage genomes.

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.

TITAN: T-cell receptor specificity prediction with bimodal attention networks.

MotivationThe activity of the adaptive immune system is governed by T-cells and their specific T-cell receptors (TCR), which selectively recognize foreign antigens. Recent advances in experimental techniques have enabled sequencing of TCRs and their antigenic targets (epitopes), allowing to research the missing link between TCR sequence and epitope binding specificity. Scarcity of data and a large sequence space make this task challenging, and to date only models limited to a small set of epitopes have achieved good performance. Here, we establish a k-nearest-neighbor (K-NN) classifier as a strong baseline and then propose Tcr epITope bimodal Attention Networks (TITAN), a bimodal neural network that explicitly encodes both TCR sequences and epitopes to enable the independent study of generalization capabilities to unseen TCRs and/or epitopes.ResultsBy encoding epitopes at the atomic level with SMILES sequences, we leverage transfer learning and data augmentation to enrich the input data space and boost performance. TITAN achieves high performance in the prediction of specificity of unseen TCRs (ROC-AUC 0.87 in 10-fold CV) and surpasses the results of the current state-of-the-art (ImRex) by a large margin. Notably, our Levenshtein-based K-NN classifier also exhibits competitive performance on unseen TCRs. While the generalization to unseen epitopes remains challenging, we report two major breakthroughs. First, by dissecting the attention heatmaps, we demonstrate that the sparsity of available epitope data favors an implicit treatment of epitopes as classes. This may be a general problem that limits unseen epitope performance for sufficiently complex models. Second, we show that TITAN nevertheless exhibits significantly improved performance on unseen epitopes and is capable of focusing attention on chemically meaningful molecular structures.Availability and implementationThe code as well as the dataset used in this study is publicly available at https://github.com/PaccMann/TITAN.Supplementary information Supplementary data are available at Bioinformatics online.

Incremental Massive Random Access Exploiting the Nested Reed-Muller Sequences

Massive machine-type communication (mMTC) is expected to provide reliable and low-latency connectivity for an enormous number of devices, which turn active sporadically or frequently. In this highly dynamic situation, it is crucial to design efficient random access (RA) procedures to cope both with the flood of simultaneous access requests and with the potential access failures. In this article, by exploiting the large sequence space, the excellent correlation property and especially the elegant nested structure of Reed-Muller (RM) sequences, we propose a new RA scheme, which facilitates both instantaneous access for newly active users and incremental access for the existing users who suffer from detection failures. In particular, when a failure occurs, the user continues accessing the channel employing an expanded RM sequence, which is combined with the previously received ones at the access point (AP) to form a longer sequence so as to attain potentially better detection probability. Furthermore, a recursive detection algorithm is designed for jointly detecting the resultant RM sequences and the channel coefficients of both the newly active users and the existing ones. The performance of the proposed algorithm is analyzed in detail. Our simulation results validate the analysis and show the scheme’s superior access probability, access latency and computational complexity.

High-throughput modeling and scoring of TCR-pMHC complexes to predict cross-reactive peptides.

The binding of T-cell receptors (TCRs) to their target peptide MHC (pMHC) ligands initializes the cell-mediated immune response. In autoimmune diseases such as multiple sclerosis, the TCR erroneously recognizes self-peptides as foreign and activates an immune response against healthy cells. Such responses can be triggered by cross-recognition of the autoreactive TCR with foreign peptides. Hence, it would be desirable to identify such foreign-antigen triggers to provide a mechanistic understanding of autoimmune diseases. However, the large sequence space of foreign antigens presents an obstacle in the identification of cross-reactive peptides. Here, we present an in silico modeling and scoring method which exploits the structural properties of TCR-pMHC complexes to predict the binding of cross-reactive peptides. We analyzed three mouse TCRs and one human TCR isolated from a patient with multiple sclerosis. Cross-reactive peptides for these TCRs were previously identified via yeast display coupled with deep sequencing, providing a robust dataset for evaluating our method. Modeling query peptides in their associated TCR-pMHC crystal structures, our method accurately selected the top binding peptides from sets containing more than a hundred thousand unique peptides. Analyses were performed using custom Python and R scripts available at https://github.com/weng-lab/antigen-predict. Supplementary data are available at Bioinformatics online.

Smart Microbial Cells Couple Catalysis and Sensing to Provide High-Throughput Selection of an Organophosphate Hydrolase.

Enzyme engineering for gain of function requires navigating a large combinatorial sequence space efficiently. Typically, many mutations are needed to get significant improvements, while a single "bad" mutation can inactivate the enzyme. To establish high-throughput screening and achieve enhanced resolution between two variants, genetic libraries of the organophosphate hydrolase enzyme paraoxonase 1 (PON1) were rapidly screened via an engineered positive-feedback circuit: a p-nitrophenol (PNP)-specific transcription factor (TF) regulated expression of PON1, which catalyzed paraoxon breakdown and PNP production. Rare active mutant colonies, picked by simple visual fluorescence of a PON1-green fluorescent protein (GFP) fusion, were characterized. In a single screening round, high (library-scale) throughput enabled the discovery of enhanced paraoxon degradation activity in PON1, including structurally unexpected mutations.

Evaluating Self-Assembly Propensity of Tetra-Peptide using MD and Machine Learning

Evaluation of the self-assembly propensity such as aggregation and amyloid formation of a peptide is important to elucidate protein property. It is also useful in protein engineering for avoiding the aggregation of designed proteins. And, self-assembly structures with high stability are used for nanomaterials. As measuring aggregation propensity of many peptides using experimental methods are difficult, molecular dynamics (MD) simulation has been used for predicting self-assembling peptides. It has been reported that di-peptides (202=400 types) and tri-peptides (203=8000 types) aggregation propensity can be evaluated using MD simulation. However, such methods are difficult to be applied to tetra-peptides due to a large sequence space (204=160,000 types). In this study, we combined the MD simulation and machine learning to accelerate the evaluation of aggregation propensity. We conducted full atomistic MD simulation for 319 peptides and used them for training a machine-learning model to search all possible tetra-peptides. The two scores were used as the measures for self-assembly propensity: AP score related to the accessible surface area of peptides and Beta score related to the possibility of β-sheets structure. We validated that machine learning can accurately predict the result of MD simulation with Pearson's correlation coefficients of 0.88 (AP score) and 0.86 (Beta score). Top 100 predictions suggest that the histidine at C-terminus promotes the formation of a β-sheets structure. We experimentally confirmed that a selected peptide with N-terminal proline predicted to have high AP score and Beta score was aggregated and then changed to fibrous crystals. The crystal structure determined using X-ray diffraction data showed β-sheets structure. These results demonstrate that our framework combining machine learning and MD simulation is a promising strategy for evaluating a property of peptides and proteins.

High-Density DNA and RNA microarrays - Photolithographic Synthesis, Hybridization and Preparation of Large Nucleic Acid Libraries.

Photolithography is a powerful technique for the synthesis of DNA oligonucleotides on glass slides, as it combines the efficiency of phosphoramidite coupling reactions with the precision and density of UV light reflected from micrometer-sized mirrors. Photolithography yields microarrays that can accommodate from hundreds of thousands up to several million different DNA sequences, 100-nt or longer, in only a few hours. With this very large sequence space, microarrays are ideal platforms for exploring the mechanisms of nucleic acid·ligand interactions, which are particularly relevant in the case of RNA. We recently reported on the preparation of a new set of RNA phosphoramidites compatible with in situ photolithography and which were subsequently used to grow RNA oligonucleotides, homopolymers as well as mixed-base sequences. Here, we illustrate in detail the process of RNA microarray fabrication, from the experimental design, to instrumental setup, array synthesis, deprotection and final hybridization assay using a template 25mer sequence containing all four bases as an example. In parallel, we go beyond hybridization-based experiments and exploit microarray photolithography as an inexpensive gateway to complex nucleic acid libraries. To do so, high-density DNA microarrays are fabricated on a base-sensitive monomer that allows the DNA to be conveniently cleaved and retrieved after synthesis and deprotection. The fabrication protocol is optimized so as to limit the number of synthetic errors and to that effect, a layer of β-carotene solution is introduced to absorb UV photons that may otherwise reflect back onto the synthesis substrates. We describe in a step-by-step manner the complete process of library preparation, from design to cleavage and quantification.

Going Beyond the Sequences: TCR Binding Patterns at the Service of Cancer Detection.

Deep sequencing of T-cell receptors enables the comprehensive profiling of lymphocyte populations and the characterization of the repertoire of T-cell responses against tumors, which could be applied to diagnose cancers. Ostmeyer and colleagues introduce a novel approach to characterize TCR patterns correlating with antigen recognition. By projecting the large TCR sequence space into a handful of biophysicochemical descriptors for key residues and seeking TCRs with similar antigen-binding capabilities even in the absence of identical amino acids, this approach presents several advantages over current methods.See related article by Ostmeyer et al., p. 1671.

Approaching protein design with multisite λ dynamics: Accurate and scalable mutational folding free energies in T4 lysozyme.

The estimation of changes in free energy upon mutation is central to the problem of protein design. Modern protein design methods have had remarkable success over a wide range of design targets, but are reaching their limits in ligand binding and enzyme design due to insufficient accuracy in mutational free energies. Alchemical free energy calculations have the potential to supplement modern design methods through more accurate molecular dynamics based prediction of free energy changes, but suffer from high computational cost. Multisite λ dynamics (MSλD) is a particularly efficient and scalable free energy method with potential to explore combinatorially large sequence spaces inaccessible with other free energy methods. This work aims to quantify the accuracy of MSλD and demonstrate its scalability. We apply MSλD to the classic problem of calculating folding free energies in T4 lysozyme, a system with a wealth of experimental measurements. Single site mutants considering 32 mutations show remarkable agreement with experiment with a Pearson correlation of 0.914 and mean unsigned error of 1.19 kcal/mol. Multisite mutants in systems with up to five concurrent mutations spanning 240 different sequences show comparable agreement with experiment. These results demonstrate the promise of MSλD in exploring large sequence spaces for protein design.