Search bioRxiv⌕ Search

Biology subjects

Kuhlman, B.

Publications and source records attributed to Kuhlman, B..

At least 19 recordsLinked to original sources
bioRxiv2026

Deep learning based design of buried hydrogen bond networks with HBDesigner

Accurate design of hydrogen-bonding (H-bonding) interactions is a longstanding goal in protein design, as they can facilitate specific protein-protein interactions while improving the solubility of the proteins in the unbound state. Despite this, computational design of H-bond networks remains underexplored in the deep learning era. Here, we present HBDesigner, a novel algorithm for H-bond network design. Through a combination of deep learning-based sampling and atomistic energy scoring, HBDesigner outperforms existing tools in designing connected H-bond networks onto protein scaffolds. We demonstrate the usefulness of HBDesigner by creating monomeric proteins with buried polar interactions and homodimers with extended interface H-bond networks, and by installing specificity into a family of homologous heterodimers where prior design tools fail to do so. The ability to design H-bond networks into arbitrary protein scaffolds should be broadly useful for a wide range of design applications.

bioengineering↗
bioRxiv2026

De novo masking domains that gate TNF-family ligand assembly and activity

Tumor necrosis factor family ligands (TNFLs) are central regulators of immunity and promising agents for cancer therapy, but their clinical use is often limited by dose-limiting systemic toxicity. Conditional activation and genetic fusion with antibodies could improve safety and pharmacokinetics, yet these features are difficult to combine because TNFLs form obligate homotrimers that are structurally mismatched with antibody architectures. Here we use AI-enabled protein design to create de novo protein masks that conditionally control TNFL assembly. The masks are genetically fused to TNFLs through protease-sensitive linkers and prevent trimer formation by competitively binding the TNFL trimerization interface. We demonstrate with TNF, OX40L, and 4-1BBL that protease-mediated release of the mask promotes trimer assembly, receptor binding and biological activity. These monomeric, switchable TNFLs are readily incorporated into full-length IgG formats, enabling plug-and-play construction of conditionally activatable antibody fusions. These fusions can be designed either to release activated soluble TNFLs upon protease treatment or, by reordering the fusion architecture, to trigger antibody multimerization and signaling by membrane-type TNFLs. When this multimerization strategy is applied to antibody-drug conjugates, conditional trimerization enhances cell killing by [~]30-fold, thereby improving the therapeutic window and enabling multiple strategies for selective tumor microenvironment targeting.

bioengineering↗
bioRxiv2026

Enhancing ML-based binder design with high-throughput screening: a comparison of mRNA and yeast display technologies

Recent advances in machine learning (ML)-based protein design methods have enabled the rapid in silico generation of large libraries of miniprotein binders with minimal manual input. While computational design capacity has scaled rapidly, experimental validation methods have lagged, creating a bottleneck in binder discovery pipelines. Here, we apply mRNA display to screen an ML-designed miniprotein binder library and directly compare its performance with the more widely used yeast surface display platform using a single shared DNA library. We screened 2,009 designs targeting the platelet receptor TLT-1 and 3,159 designs targeting the immune receptor B7-H3 across both platforms. While both selection methods reliably identified functional binders, we found that mRNA display preferentially enriched binders with slower dissociation rates. In addition, mRNA display achieved higher library coverage than yeast display, likely rescuing functional designs that are penalized in a cell-based expression system. Biophysical characterization of selected binders from both platforms revealed strong binding affinities and high thermal stabilities. These results showcase the power of integrating ML-based computational design tools with rapid in vitro selection technologies, providing a scalable framework for therapeutic miniprotein discovery. IMPORTANCEMiniprotein binders offer major advantages as next-generation therapeutics, including small size, high stability, and efficient production. In this work, we conduct a side-by-side comparison of mRNA and yeast display as platforms for high-throughput evaluation of de novo miniprotein binders. The binders generated here serve as starting points for therapeutics targeting TLT-1 or B7-H3, two clinically relevant molecules.

bioengineering↗
bioRxiv2026

Computational design of orthogonal TCR α/β interfaces for dual-TCR therapeutics

T-cell receptors (TCRs) recognize peptides presented by MHC, enabling access to intracellular targets that are largely inaccessible to antibodies and difficult to target with small molecules. Despite this potential, their inherent cross-reactivity limits tumor specificity, while single-antigen targeting provides limited coverage of intratumoral heterogeneity. Dual-TCR therapeutics comprising two distinct TCRs could enhance tumor specificity via combinatorial recognition while broadening coverage across heterogeneous antigens. However, practical development of dual-TCR therapeutics has been limited by /{beta} subunit mispairing that prevents efficient production and creates undesired binding properties. Here, we develop orthogonal TCR /{beta} interfaces that prevent subunit mispairing. Using computational multistate design and second-site suppressor strategies implemented in Rosetta, we identified over 250 TCR variants for experimental screening to assess protein stability and pairing fidelity. The top-performing designs achieved approximately 95% correct pairing, as validated by mass spectrometry and X-ray crystallography. Focusing mutations on constant domains and conserved framework regions of variable domains enabled broad applicability across diverse TCRs while preserving antigen recognition. Using these orthogonal interfaces, we developed trispecific T-cell engagers (TriTEs) that target two cancer-testis antigens and CD3 on T cells, demonstrating enhanced potency under dual-antigen engagement (EC50 of 380 fM) while maintaining high activity when targeting cells displaying a single antigen (EC50s of 48 pM and 20 pM). This orthogonal TCR interface technology establishes a generalizable platform for engineering multi-specific immune therapeutics targeting diverse cancer antigens.

bioengineering↗
bioRxiv2025

De novo design of protein binders that target DELE1 to inhibit the mitochondrial stress response

Mitochondrial stress activates the integrated stress response (ISR) through the mitochondrial protein DELE1, which relays stress signals to the cytosolic kinase HRI to induce ATF4. Dysregulation of DELE1-mediated signaling has been implicated in pathological conditions, yet molecular strategies to modulate DELE1 remain unavailable. Here, we report de novo designed proteins that bind DELE1, block its oligomerization, and inhibit DELE1-mediated ISR activation. Several designs form stable complexes with DELE1 and disrupt its oligomerization in vitro while preserving DELE1s ability to bind HRI. In cells, these designs suppress ATF4 induction during mitochondrial stress and impair the recovery of elongated mitochondrial morphology following transient insult. Crystal structure of a representative binder, together with structural modeling and targeted mutagenesis, confirm that the designed proteins engage a critical interface required for DELE1 oligomerization. These findings establish DELE1 as a druggable target and demonstrate that de novo designed proteins offer precise tools to modulate this pathway, providing a foundation for future therapeutic exploration. SIGNIFICANCEMitochondrial stress activates the integrated stress response through the signaling protein DELE1, but no molecular tools have been available to directly target DELE1 and selectively modulate this pathway. We developed de novo designed protein binders that recognize a critical oligomerization interface in DELE1, disrupt its assembly, and suppress mitochondrial stress-induced ISR activation in cells. These binders also impair recovery of mitochondrial network morphology following transient stress, linking DELE1 assembly to adaptive remodeling. Our study establishes DELE1 as a tractable and druggable target in mitochondrial stress signaling and demonstrates that de novo protein design can generate precise modulators of intracellular stress-response pathways.

biochemistry↗
bioRxiv2025

Computational Stabilization of the Human VH Germline Repertoire to Enable Conditional Multi-Specific Therapeutic Development

Protein-based biologic therapies, particularly antibody-like therapeutics, have emerged as a major modality to treat nearly all chronic and infectious diseases. Antagonists have dominated the first wave of antibody and antibody-like biologics, whereas agonism has generally been challenging due to systemic activation leading to issues with therapeutic index and problems with pleiotropic activity. Additionally, agonists that use native proteins such as cytokines can be challenging to produce at scale given these proteins evolved to act locally and are not designed for large scale manufacturing. To address these challenges, we generated a biologics platform comprised of stabilized human VH domains (VH-Select) which encompass the entire human germline repertoire with the goal of building multispecific biologics denoted Tentacles that use avidity-based binding to achieve conditional activity directed to specific cell types or tissues. Stabilizing disulfides and point mutations were identified computationally with Rosetta, evaluated in vitro, and combined into designs with 3-5 amino acid substitutions for each of the seven germline families (VH1-VH7). Computational design was also employed to reduce dimerization from both VL and homotypic VH-VH interactions. Optimization of specific sequences improved expression by greater than 600-fold and thermostability by more than 20{degrees}C. Each of the germline variants were screened for low HLA class II binding and incorporated into a library which demonstrated significant improvements in cellular protein production, thereby increasing sequence diversity for screening campaigns. These VH-Select scaffolds are useful for the discovery of novel binders used to build multispecific Tentacles designed for cis-interactions that achieve cell- and tissue-specific activation. As an example of the utility of the platform, we generated a set of Tentacles that use VH-Select binders to conditionally agonize IL2R{gamma}{beta} and 41BB on PD1+, LAG3+, or CD8+ T cells. These Tentacles demonstrate promising manufacturability, antibody-like exposure in vivo, and strong anti-tumor activity in a humanized tumor model. Overall, we believe the incorporation of VH-Select binders into multispecific Tentacles has the potential to create a host of conditionally active biologics to treat various chronic and acute diseases.

bioengineering↗
bioRxiv2025

Enhancing enzymatic bioconjugation efficiency via installation of a substrate recruitment domain

Enzyme mediated bioconjugation provides a method for easy and rapid formation of protein-protein and protein-small molecule conjugates under mild conditions. Promiscuous enzymes are of particular interest because they can catalyze conjugation reactions on a broad set of substrates. However, this promiscuity carries the risk of undesirable off-target modifications. To mitigate this effect, we used computational design to install a substrate recruitment domain (SRD) onto the promiscuous enzyme, tyrosinase. The redesigned tyrosinase, called D42, preferentially modifies tyrosine residues within the peptide core (core) linked to a 6-amino acid recognition motif/sequence (RS) specific for the SRD. Incorporation of the recognition sequence along with a neighboring tyrosine in peptides or proteins allows for rapid D42-mediated conversion of the tyrosine to an orthoquinone, which can be selectively modified with a variety of nucleophiles. We demonstrate the utility of our design system by rapidly installing cytotoxic molecules on a monoclonal antibody. For Table of Contents Only O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=72 SRC="FIGDIR/small/684804v1_ufig1.gif" ALT="Figure 1"> View larger version (20K): org.highwire.dtl.DTLVardef@1a3b274org.highwire.dtl.DTLVardef@18eeedorg.highwire.dtl.DTLVardef@ee0aaeorg.highwire.dtl.DTLVardef@1ff3fa1_HPS_FORMAT_FIGEXP M_FIG C_FIG

biochemistry↗
bioRxiv2025

Intensity Modulation of Trichromatic Split Fluorescent Proteins for Live Cell Mapping

Multiplexed live cell-labeling technologies are essential tools for investigating complex biological systems, yet existing approaches face limitations in achieving reliable discrimination of multiple cell populations. Here, we present color and intensity modulation with split fluorescent proteins for cell labeling (Caterpie), a live cell labeling system that enables high-fidelity identification of multiple cell populations through rational design of split fluorescent proteins. Drawing inspiration from human trichromatic vision, we engineered optimized split fluorescent protein variants--split CFP2, split mNG3A, and split sfCherry3C--and developed a comprehensive library of the 11th {beta}-strand (FP11) tags comprising variable numbers of tandem repeats. Structure-guided protein engineering enhanced the performance of split mNG3A and split sfCherry3C through targeted modifications of their C-terminal sequences and binding interfaces, respectively. By systematically varying the composition and arrangement of FP11 repeats, we selected 20 distinct FP11 tags that enable robust cell population discrimination with 97% accuracy using conventional fluorescence microscopy. In these FP11 tag-expressing cells, we expressed multiple EGF ligands and their receptors and cultured them in bulk to examine the effects of EGF signaling on chemotaxis within the cell population. The results showed that chemotaxis changes according to the intensity of EGFR signaling. The Caterpie system holds significant potential for applications requiring precise identification and tracking of multiple cell populations in complex biological systems.

cell biology↗
bioRxiv2025

PRAME Epitopes are T-Cell Immunovulnerabilities in BRD4::NUTM1 Initiated NUT Carcinoma

BackgroundNUT carcinoma is a rare but highly lethal solid tumor without an effective standard of care. NUT carcinoma is caused by bromodomain-containing NUTM1 fusion oncogenes, most commonly BRD4::NUTM1. BRD4::NUTM1 recruits p300 to acetylate H3K27 forming expansive stretches of hyperacetylated chromatin called "megadomains" with the overexpression of corresponding oncogenes, including MYC. We hypothesized that transcriptional dysregulation caused by BRD4::NUTM1 would lead to the generation of cancer-specific antigens that could be therapeutically actionable. MethodsWe integrated genomics, computational antigen prediction software, targeted immunopeptidomics using single- and double-labeled peptide standards, and gain/loss-of-function genetic experiments on a panel of cell lines (N=5), a patient derived xenograft, a tissue microarray (N=77), and patient samples from the Tempus AI Sequencing Database harboring evidence of NUTM1 fusions (N=165). We created an PRAME425 T-cell receptor x SP34 CD3 bispecific molecule modeled after brenetafusp, an PRAME425 T-cell receptor bispecific T-cell engager, as well as PRAME425 TCR T-cells based on anzutresgene autoleucel and we applied these products to NUT carcinoma cells in vitro. ResultsWe identified PRAME as the most commonly expressed cancer/testis antigen in patient samples harboring the three canonical NUT carcinoma fusions (BRD4::NUTM1, BRD3::NUTM1, and NSD3::NUTM1). Additionally, 56% (43/77) of NUT carcinoma tissue microarray samples stained positive for PRAME. BRD4::NUTM1 expression in HEK 293T cells enhanced PRAME levels and BRD4::NUTM1 knockout in NUT carcinoma cells reduced PRAME levels. Immunopeptidomics detected more PRAME-derived HLA ligands (N=9) than all other cancer/testis antigens combined (N=5). Targeted mass spectrometry detected the HLA-A*02:01/SLLQHLIGL (PRAME425) epitope in 100% (4/4) of HLA-A*02+, PRAME+ NUT carcinoma samples at higher levels (>0.01 fM) than HLA-A*02:01/RLDQLLRHV (PRAME312) or HLA-A*02:01/YLHARLREL (PRAME462). The PRAME425 T-cell receptor x SP34 CD3 bispecific molecule and PRAME425 TCR T-cells each exhibited potent, T-cell mediated cytotoxicity against PRAME+ NUT carcinoma cells. ConclusionsPRAME is highly and frequently expressed in NUT carcinoma and the most common oncoprotein causing NUT carcinoma, BRD4::NUTM1, contributes to these high PRAME levels. PRAME epitopes presented by HLA Class I are a previously unrecognized therapeutic vulnerability for NUT carcinoma that warrant clinical trials testing PRAME targeted immunotherapies in this neglected patient population. What is already known on this topicNUT carcinoma is a devastating malignancy that is recalcitrant to cytotoxic chemotherapy, T-cell checkpoint blockade, and targeted therapies in the form of bromodomain inhibitors. What this study addsNUT carcinoma tumors are high in the cancer/testis gene PRAME. The oncogene most commonly causing NUT carcinoma, BRD4::NUTM1, contributes to these high levels. NUT carcinoma cells present PRAME epitopes on HLA Class I molecules and are susceptible to PRAME-directed, T-cell mediated cytotoxicity. How this study might affect research, practice or policyOur results argue for phase I/II clinical trials testing PRAME immunotherapies like brenetafusp or anzutresgene autoleucel in PRAME+ NUT carcinoma patients. Graphical Abstract O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=149 SRC="FIGDIR/small/642090v3_ufig1.gif" ALT="Figure 1"> View larger version (38K): org.highwire.dtl.DTLVardef@495654org.highwire.dtl.DTLVardef@c2b975org.highwire.dtl.DTLVardef@1de507org.highwire.dtl.DTLVardef@a76d33_HPS_FORMAT_FIGEXP M_FIG C_FIG

cancer biology↗
bioRxiv2024

Design of a Water-Soluble CD20 Antigen with Computational Epitope Scaffolding

The poor solubility of integral membrane proteins in water frequently hinders studies with these proteins, presenting challenges for structure determination and binding screens. For instance, the transmembrane protein CD20, which is an important target for treating B-cell malignancies, is not soluble in water and cannot be easily screened against potential protein binders with techniques like phage display or yeast display. Here, we use de novo protein design to create a water-soluble mimic of the CD20 dimer ("soluble CD20"). Soluble CD20 replaces the central transmembrane helix of CD20 with a water-soluble helix that dimerizes to form a coiled coil that structurally matches the dimer interface of native CD20 and presents the central extracellular loop of CD20 in a binding competent conformation. Unlike peptides derived from CD20, soluble CD20 binds tightly to monoclonal antibodies that recognize quaternary epitopes on the extracellular face of CD20. We demonstrate that soluble CD20 is easy to produce, remains folded above 60{degrees}C, and is compatible with binder screening via yeast display. Our results highlight the ability of computational protein design to scaffold conformational epitopes from membrane proteins for use in binding and protein engineering studies. ImportanceThis study demonstrates that computational protein design can be used to create water soluble variants of transmembrane proteins. The water-soluble CD20 mimic created here is compatible with high throughput binding screens and should be useful for developing next generation anti-CD20 therapeutics.

biophysics↗
bioRxiv2024

Assessing substrate scope of the cyclodehydratase LynD by mRNA display-enabled machine learning models

Many of the biosynthetic pathways for ribosomal synthesized and post-translationally modified peptide (RiPP) natural products make use of multi-domain enzymes with separate recruitment and catalysis domains that separately bind and modify peptide substrates. This "division of labor" allows RiPP enzymes to use relatively open and promiscuous active sites to perform chemistry at multiple residues within a peptide substrate seemingly regardless of the surrounding context. Defining, measuring, and predicting the seemingly broad substrate promiscuity of RiPPs necessitates high throughput assays, capable of assessing activity against very large libraries of peptides. Using mRNA display, a high throughput peptide display technology, we examine the substrate promiscuity of the RiPP cyclodehydratase, LynD. The vast substrate profiling that can be done with mRNA display enables the construction of deep learning models for accurate prediction of substrate processing by LynD. These models further inform on epistatic interactions involved in enzymatic processing. This work will facilitate the further elucidation of other RiPP enzymes and enable their use in the modification of mRNA display libraries for selection of modified peptide-based inhibitors and therapeutics.

biochemistry↗
bioRxiv2024

Protein stability models fail to capture epistatic interactions of double point mutations

There is strong interest in accurate methods for predicting changes in protein stability resulting from amino acid mutations to the protein sequence. Recombinant proteins must often be stabilized to be used as therapeutics or reagents, and destabilizing mutations are implicated in a variety of diseases. Due to increased data availability and improved modeling techniques, recent studies have shown advancements in predicting changes in protein stability when a single point mutation is made. Less focus has been directed toward predicting changes in protein stability when there are two or more mutations, despite the significance of mutation clusters for disease pathways and protein design studies. Here, we analyze the largest available dataset of double point mutation stability and benchmark several widely used protein stability models on this and other datasets. We identify a blind spot in how predictors are typically evaluated on multiple mutations, finding that, contrary to assumptions in the field, current stability models are unable to consistently capture epistatic interactions between double mutations. We observe one notable deviation from this trend, which is that epistasis-aware models provide marginally better predictions on stabilizing double point mutations. We develop an extension of the ThermoMPNN framework for double mutant modeling as well as a novel data augmentation scheme which mitigates some of the limitations in available datasets. Collectively, our findings indicate that current protein stability models fail to capture the nuanced epistatic interactions between concurrent mutations due to several factors, including training dataset limitations and insufficient model sensitivity. SignificanceProtein stability is governed in part by epistatic interactions between energetically coupled residues. Prediction of these couplings represents the next frontier in protein stability modeling. In this work, we benchmark protein stability models on a large dataset of double point mutations and identify previously overlooked limitations in model design and evaluation. We also introduce several new strategies to improve modeling of epistatic couplings between protein point mutations.

bioinformatics↗
bioRxiv2024

Multivalent administration of dengue E dimers on liposomes elicits type-specific neutralizing responses without immune interference

The four serotypes of dengue virus (DENV1-4) are a major health concern putting 50% of the global population at risk of infection. Crucially, DENV vaccines must be tetravalent to provide protection against all four serotypes because immunity to only one serotype can enhance infections caused by heterologous serotypes. Uneven replication of live-attenuated viruses in tetravalent vaccines can lead to disease enhancement instead of protection. Subunit vaccines are a promising alternative as the vaccine components are not dependent on viral replication and antigen doses can be controlled to achieve a balanced response. Here, we show that a tetravalent subunit vaccine of dengue envelope (E) proteins computationally stabilized to form native-like dimers elicits type-specific neutralizing antibodies in mice against all four serotypes. The immune response was enhanced by displaying the E dimers on liposomes embedded with adjuvant, and no interference was detected between the four components.

microbiology↗
bioRxiv2024

Stabilized dengue virus 2 envelope subunit vaccine redirects the neutralizing antibody response to all E-domains

The four-dengue virus (DENV) serotypes cause several hundred million infections annually. Several live-attenuated tetravalent dengue vaccines (LAVs) are at different stages of clinical testing and regulatory approval. A major hurdle faced by the two leading LAVs is uneven replication of vaccine serotypes stimulating a dominant response to one serotype at the expense of the other three, leading to the potential for vaccine antibody (Ab) enhanced more severe infections by wild type DENV serotypes that fail to replicate in the vaccine. Protein subunit vaccines are a promising alternative since antigen dosing can be precisely controlled. However, DENV envelope (E) protein subunit vaccines have not performed well to date, possibly due to differences between the monomeric structure of soluble E and the E homodimer of the viral surface. Previously, we have combined structure-guided computational and experimental approaches to design and produce DENV2 E antigens that are stable homodimers at 37 and stimulate higher levels of neutralizing Abs (NAbs) than the WT E antigen in mice. The goal of this study was to evaluate if DENV2 E homodimers stimulate NAbs that target different epitopes on E protein compared to the WT E monomer. Using DENV4/2 chimeric viruses and Ab depletion methods, we mapped the WT E-elicited NAbs to simple epitopes on domain III of E. In contrast, the stable E homodimer stimulated a more complex response towards all three surface-exposed domains of the E protein. Our findings highlight the impact of DENV2 E oligomeric state on the quality and specificity of DENV NAbs, and the promise of DENV E homodimers as subunit vaccines. ImportanceThe ideal dengue virus (DENV) vaccine should elicit a balanced and highly protective immune response against all 4 DENV serotypes. Current tetravalent live-attenuated DENV vaccines have faced challenges due to uneven replication of vaccine virus strains stimulating a strong immune response to one serotype and weak responses to the other three. Protein subunit vaccines provide novel opportunities to stimulate a balanced response because dosing can be precisely controlled and independent of vaccine virus replication. Here, we compare immune responses elicited by a new DENV serotype 2 protein vaccine designed to match the structure of proteins on the viral surface. We find that proteins designed to match the viral surface stimulate better immune responses targeting multiple sites on the viral surface compared to previous protein vaccines. Our results justify further testing and development of these second-generation DENV protein subunit vaccines.

microbiology↗
bioRxiv2023

Invariant point message passing for protein side chain packing and design

Protein side chain packing (PSCP) is a fundamental problem in the field of protein engineering, as high-confidence and low-energy conformations of amino acid side chains are crucial for understanding (and designing) protein folding, protein-protein interactions, and protein-ligand interactions. Traditional PSCP methods (such as the Rosetta Packer) often rely on a library of discrete side chain conformations, or rotamers, and a forcefield to guide the structure to low-energy conformations. Recently, deep learning (DL) based methods (such as DLPacker, AttnPacker, and DiffPack) have demonstrated state-of-the-art predictions and speed in the PSCP task. Building off the success of geometric graph neural networks for protein modeling, we present the Protein Invariant Point Packer (PIPPack) which effectively processes local structural and sequence information to produce realistic, idealized side chain coordinates using {chi}-angle distribution predictions and geometry-aware invariant point message passing (IPMP). On a test set of [~]1,400 high-quality protein chains, PIPPack is highly competitive with other state-of-the-art PSCP methods in rotamer recovery and per-residue RMSD but is significantly faster.

biophysics↗
bioRxiv2023

Transfer learning to leverage larger datasets for improved prediction of protein stability changes

Amino acid mutations that lower a proteins thermodynamic stability are implicated in numerous diseases, and engineered proteins with enhanced stability are important in research and medicine. Computational methods for predicting how mutations perturb protein stability are therefore of great interest. Despite recent advancements in protein design using deep learning, in silico prediction of stability changes has remained challenging, in part due to a lack of large, high-quality training datasets for model development. Here we introduce ThermoMPNN, a deep neural network trained to predict stability changes for protein point mutations given an initial structure. In doing so, we demonstrate the utility of a newly released mega-scale stability dataset for training a robust stability model. We also employ transfer learning to leverage a second, larger dataset by using learned features extracted from a deep neural network trained to predict a proteins amino acid sequence given its three-dimensional structure. We show that our method achieves competitive performance on established benchmark datasets using a lightweight model architecture that allows for rapid, scalable predictions. Finally, we make ThermoMPNN readily available as a tool for stability prediction and design.

bioinformatics↗
bioRxiv2023

In silico evolution of protein binders with deep learning models for structure prediction and sequence design

There has been considerable progress in the development of computational methods for designing protein-protein interactions, but engineering high-affinity binders without extensive screening and maturation remains challenging. Here, we test a protein design pipeline that uses iterative rounds of deep learning (DL)-based structure prediction (AlphaFold2) and sequence optimization (ProteinMPNN) to design autoinhibitory domains (AiDs) for a PD-L1 antagonist. Inspired by recent advances in therapeutic design, we sought to create autoinhibited (or masked) forms of the antagonist that can be conditionally activated by proteases. Twenty-three de novo designed AiDs, varying in length and topology, were fused to the antagonist with a protease sensitive linker, and binding to PD-L1 was tested with and without protease treatment. Nine of the fusion proteins demonstrated conditional binding to PD-L1 and the top performing AiDs were selected for further characterization as single domain proteins. Without any experimental affinity maturation, four of the AiDs bind to the PD-L1 antagonist with equilibrium dissociation constants (KDs) below 150 nM, with the lowest KD equal to 0.9 nM. Our study demonstrates that DL-based protein modeling can be used to rapidly generate high affinity protein binders. Significance statementProtein-protein interactions are critical to most processes in biology, and improved methods for designing protein binders will enable the creation of new research reagents, diagnostics, and therapeutics. In this study, we show that a deep learning-based method for protein design can create high-affinity protein binders without the need for extensive screening or affinity maturation.

biophysics↗