Biology subjects

Kortemme, T.

Publications and source records attributed to Kortemme, T..

At least 19 recordsLinked to original sources

bioRxiv2026

Generative Design of New-to-nature Biosynthetic Assembly Lines with Genomic Language Modeling

Reprogramming biosynthetic assembly lines can extend biosynthesis beyond the chemical space explored by nature. However, this remains difficult because assembly-line function depends on coordinated interactions across large multidomain enzymes. Here, we couple gLM2, a genomic language model trained on metagenomic sequences, with discrete diffusion and domain-level conditioning to enable generative design and optimization of biosynthetic gene clusters. We apply this approach to a chimeric type I polyketide synthase (PKS) engineered to produce {delta}-valerolactam, a molecule not naturally synthesized by PKSs. Through iterative redesign of two multi-domain regions in the context of the full PKS sequence, gLM2 progressively improved {delta}-valerolactam production, yielding variants with up to 9.4-fold higher titer than the starting enzyme. Together, these results demonstrate that evolutionary sequence information can be learned and applied to complex, multi-domain enzyme design problems, expanding biosynthetic assembly lines to produce molecules outside their natural biosynthetic repertoire.

bioengineering↗

bioRxiv2026

Deep Learning-based Modeling Enhances Efficacy of Natural Ligand CAR Binders Targeting CD70

CD70 is well-recognized as a promising "pan-cancer: chimeric antigen receptor (CAR) T-cell target. Prior work has shown that a "natural ligand" (NL)-based CAR targeting CD70, employing its physiological interaction partner CD27, may have therapeutic advantages over antibody-based CARs. Yet while antibody-based CARs are routinely optimized by affinity maturation of their scFv, whether the binding sequence of an NL CAR can be engineered to improve its function remains unexplored. Here, we combined deep learning with physics-based modeling to redesign residues at the CD27:CD70 interface, identifying a CD27 variant, "N88A", which enhances the efficacy of CD70-targeting CAR T-cells across models of acute myeloid leukemia, multiple myeloma, and renal cell carcinoma. Biophysical approaches, including molecular dynamics simulations, support a mechanism of increased binder conformational freedom underlying potency enhancement. Our work presents CD27N88A CAR T-cells as a promising new therapeutic option and proposes that computational modeling could be applied to enhance efficacy of other NL-based immunotherapies.

cancer biology↗

bioRxiv2026

Programming protein shape as an explicit design layer via CAD blueprint-guided diffusion

In engineering, industrial design principles link geometric form with function. In nature, global protein shape, curvature, and chirality similarly drive critical processes like membrane remodeling and cellular motility. However, explicitly programming these global features remains a challenge for generative AI. Here, we integrate computer-aided design (CAD) blueprints with diffusion models, establishing protein shape as an explicit, programmable layer. We engineered diverse, tunable architectures, including single-chain shapes, scaffolds with unusual twist handedness, and superhelical assemblies, structurally validating 31 of 59 designs via X-ray crystallography or electron microscopy. To demonstrate shape-encoded function, we coupled fluid-flow-optimized helical propellers to an F1-ATPase rotor, yielding a prototype ATP-dependent molecular swimmer. This framework translates industrial design principles to the molecular scale, enabling the programmable engineering of biomimetic architecture with emergent functions. One-sentence summaryIntegrating CAD blueprints with diffusion-based AI translates industrial design principles to the molecular scale, enabling the explicit programming of protein shapes with emergent functions.

bioengineering↗

bioRxiv2026

A Hybrid Physics-Deep Learning Framework for Combinatorial De Novo Design of Small-Molecule Binding Proteins

Engineering small-molecule binding proteins de novo remains a significant challenge as even advanced generative models struggle to model the atom-level details of protein-ligand interactions with sufficient accuracy. Higher experimental success rates have resulted from methods that explicitly scaffold predefined binding interactions into helical bundles. Here we introduce a scaffolding strategy that generalizes to alpha-beta architectures. By screening thousands of combinatorially assembled protein-ligand interactions against diverse de novo backbones with finely varied pocket geometries, the protocol allows for high-fidelity accommodation of target interaction geometries. Our protocol then integrates physics-based and deep learning methods for optimization of interfacial interactions and sequence-structure compatibility, considerably improving in silico design metrics. Applying this method to two chemically similar steroids achieved a notable experimental success rate (4/26 designs bind their targets), and NMR structures of two designs are in good agreement with design models. Our generalizable, atomically precise approach offers a robust framework for small-molecule binder design, effectively eliminating the need for high-throughput screening.

bioengineering↗

bioRxiv2025

All-atom protein design via SE(3) flow matching with ProteinZen

The advent of generative models of protein structure has greatly accelerated our ability to perform de novo protein design, especially concerning design at coarser physical scales such as backbone generation and protein binder design. How-ever, the design of precise placements at atomic scales remains a challenge for existing design methods. One avenue towards higher fidelity atomic-scale design is via generative models with full atomic resolution, but is complicated by the intricacies of simultaneously designing both discrete protein sequence and continuous atomic positionings. In this work we propose a framework to capture this interplay by decomposing residues into collections of oriented rigid bodies, allowing us to apply SE(3) flow-matching for all-atom protein structure generation. Our method, ProteinZen1, generates designs with high sequence-structure consistency while retaining competitive diversity and novelty on both unconditional and conditional generation tasks. We demonstrate competitive performance for unconditional monomer design and state-of-the-art performance on various forms of motif scaffolding, including full-atom motif scaffolding and motif scaffolding without specifying motif segment spacing or relative sequence order.

bioengineering↗

bioRxiv2025

Structure-guided engineering of CCL27 enhances natural ligand CAR T-cells against CCR10 for multiple myeloma

Despite the success of BCMA CAR-Ts, many multiple myeloma patients relapse and require additional therapeutic options. Our group previously identified the chemokine receptor CCR10 as a potential alternate target to address this need. Here, we validated CCR10 expression on primary myeloma tumors and sought to develop CAR T-cells against CCR10, utilizing its natural ligand CCL27 as a CAR binding element. However, CARs based on the native CCL27 sequence were ineffective. We thus utilized computational modeling and structure-guided engineering to inform rational mutations along the CCL27-CCR10 interface, exploiting a hydrophobic pocket on CCR10. This effort identified CCL27 mutants with an additional N-terminal aromatic amino acid that dramatically improved the efficacy of CCL27-based CAR-Ts to near that of current anti-BCMA CAR-Ts. We validated key amino acid contacts at the CCL27-CCR10 interface, which contribute to increased CAR binding avidity, predicted to be influenced by increased Van der Waals interactions. Lastly, we found that the CCL27 mutants have no toxicity in the hematopoietic compartment. This work illustrates the potential of engineering natural ligand CAR-Ts beyond their wild-type sequences and underscores the translational potential of engineered CCL27 mutant CAR-Ts.

cancer biology↗

bioRxiv2025

A combinatorial mutational map of active non-native protein kinases by deep learning guided sequence design

Mapping protein sequence-function landscapes has either been limited to small steps (only few mutations) or to sequences similar to those already explored by evolution to maintain activity. Here, we overcome both limitations by applying deep-learning guided redesign to a natural protein tyrosine kinase to generate novel, functional sequences with highly combinatorial mutations. Using cell-free assays, we measure the activities and concentrations of 537 redesigned sequences, which differ from the wild-type by an average of 37 mutations while retaining activity in 85% of variants. These sequences sample 436 unique mutations at 76 different positions throughout the kinase domain. A simple regression model identifies key sequence determinants of function and predicts the function of unseen sequences. Our approach demonstrates how integrating deep-learning guided redesign, functional measurement at scale, and interpretable computational modelling enables functional exploration of highly combinatorial and sparse sequence-function landscapes at mutational scales not possible before.

biochemistry↗

bioRxiv2025

An improved model for prediction of de novo designed proteins with diverse geometries

Nature uses structural variations on protein folds to fine-tune the geometries of proteins for diverse functions, yet deep learning-based de novo protein design methods generate highly regular, idealized protein fold geometries that fail to capture natural diversity. Here, using physics-based design methods, we generated and experimentally validated a dataset of 5,996 stable, de novo designed proteins with diverse non-ideal geometries. We show that deep learning-based structure prediction methods applied to this set have a systematic bias towards idealized geometries. To address this problem, we present a fine-tuned version of Alphafold2 that is capable of recapitulating geometric diversity and generalizes to a new dataset of thousands of geometrically diverse de novo proteins from 5 fold families unseen in fine-tuning. Our results suggest that current deep learning-based structure prediction methods do not capture some of the physics that underlie the specific conformational preferences of proteins designed de novo and observed in nature. Ultimately, approaches such as ours and further informative datasets should lead to improved models that reflect more of the physical principles of atomic packing and hydrogen bonding interactions and enable improved generalization to more challenging design problems.

bioinformatics↗

bioRxiv2024

Deep learning guided design of dynamic proteins

Deep learning has greatly advanced design of highly stable static protein structures, but the controlled conformational dynamics that are hallmarks of natural switch-like signaling proteins have remained inaccessible to de novo design. Here, we describe a general deep-learning-guided approach for de novo design of dynamic changes between intra-domain geometries of proteins, similar to switch mechanisms prevalent in nature, with atom-level precision. We solve 4 structures validating the designed conformations, show microsecond transitions between them, and demonstrate that the conformational landscape can be modulated by orthosteric ligands and allosteric mutations. Physics-based simulations are in remarkable agreement with deep-learning predictions and experimental data, reveal distinct state-dependent residue interaction networks, and predict mutations that tune the designed conformational landscape. Our approach demonstrates that new modes of motion can now be realized through de novo design and provides a framework for constructing biology-inspired, tunable and controllable protein signaling behavior de novo.

bioengineering↗

bioRxiv2024

An integrative approach to protein sequence design through multiobjective optimization

With recent methodological advances in the field of computational protein design, in particular those based on deep learning, there is an increasing need for frameworks that allow for coherent, direct integration of different models and objective functions into the generative design process. Here we demonstrate how evolutionary multiobjective optimization techniques can be adapted to provide such an approach. With the established Non-dominated Sorting Genetic Algorithm II (NSGA-II) as the optimization framework, we use AlphaFold2 and ProteinMPNN confidence metrics to define the objective space, and a mutation operator composed of ESM-1v and ProteinMPNN to rank and then redesign the least favorable positions. Using the multistate design problem of the foldswitching protein RfaH as an in-depth case study, we show that the evolutionary multiobjective optimization approach leads to significant reduction in the bias and variance in RfaH native sequence recovery, compared to a direct application of ProteinMPNN. We suggest that this improvement is due to three factors: (i) the use of an informative mutation operator that accelerates the sequence space exploration, (ii) the parallel, iterative design process inherent to the genetic algorithm that improves upon the ProteinMPNN autoregressive sequence decoding scheme, and (iii) the explicit approximation of the Pareto front that leads to optimal design candidates representing diverse tradeoff conditions. We anticipate this approach to be readily adaptable to different models and broadly relevant for protein design tasks with complex specifications. Author summaryProteins are the fundamental building blocks of life, and engineering them has broad applications in medicine and biotechnology. Computational methods that seek to model and predict the protein sequence-structure-function relationship have seen significant advancement from the recent development in deep learning. As more models become available, it remains an open question how to effectively combine them into a coherent computational design approach. One approach is to perform computational design with one model, and filter the design candidates with the others. In this work, we demonstrate how an optimization algorithm inspired by evolution can be adapted to provide an alternative framework that outperforms the post hoc filtering approach, especially for problems with multiple competing design specifications. Such a framework may enable researchers to more effectively integrate the strengths of different modeling approaches to tackle more complex design problems.

biophysics↗

bioRxiv2024

Targeting high-risk multiple myeloma genotypes with optimized anti-CD70 CAR-T cells

Despite the success of BCMA-targeting CAR-Ts in multiple myeloma, patients with high-risk cytogenetic features still relapse most quickly and are in urgent need of additional therapeutic options. Here, we identify CD70, widely recognized as a favorable immunotherapy target in other cancers, as a specifically upregulated cell surface antigen in high risk myeloma tumors. We use a structure-guided design to define a CD27-based anti-CD70 CAR-T design that outperforms all tested scFv-based CARs, leading to >80-fold improved CAR-T expansion in vivo. Epigenetic analysis via machine learning predicts key transcription factors and transcriptional networks driving CD70 upregulation in high risk myeloma. Dual-targeting CAR-Ts against either CD70 or BCMA demonstrate a potential strategy to avoid antigen escape-mediated resistance. Together, these findings support the promise of targeting CD70 with optimized CAR-Ts in myeloma as well as future clinical translation of this approach. One sentence summaryStructure-optimized CD27-based CAR-T cells targeting CD70 are a promising therapeutic option for high-risk multiple myeloma patients who are most likely to relapse on current BCMA-targeting cellular therapies.

cancer biology↗

bioRxiv2023

Structure-conditioned masked language models for protein sequence design generalize beyond the native sequence space

Machine learning has revolutionized computational protein design, enabling significant progress in protein backbone generation and sequence design. Here, we introduce Frame2seq, a structure-conditioned masked language model for protein sequence design. Frame2seq generates sequences in a single pass, achieves 49.1% sequence recovery on the CATH 4.2 test dataset, and accurately estimates the error in its own predictions, outperforming the autoregressive ProteinMPNN model with over six times faster inference. To probe the ability of Frame2seq to generate novel designs beyond the native-like sequence space it was trained on, we experimentally test 26 Frame2seq designs for de novo backbones with low identity to the starting sequences. We show that Frame2seq successfully designs soluble (22/26), monomeric, folded, and stable proteins (17/26), including a design with 0% sequence identity to native. The speed and accuracy of Frame2seq will accelerate exploration of novel sequence space across diverse design tasks, including challenging applications such as multi-objective optimization.

bioengineering↗

bioRxiv2023

Multi-input drug-controlled switches of mammalian gene expression based on engineered nuclear hormone receptors

Protein-based switches that respond to different inputs to regulate cellular outputs, such as gene expression, are central to synthetic biology. For increased controllability, multi-input switches that integrate several cooperating and competing signals for the regulation of a shared output are of particular interest. The nuclear hormone receptor (NHR) superfamily offers promising starting points for engineering multi-input-controlled responses to clinically approved drugs. Starting from the VgEcR/RXR pair, we demonstrate that novel (multi-)drug regulation can be achieved by exchange of the ecdysone receptor (EcR) ligand binding domain (LBD) for other human NHR-derived LBDs. For responses activated to saturation by an agonist for the first LBD, we show that outputs can be boosted by an agonist targeting the second LBD. In combination with an antagonist, output levels are tunable by up to three simultaneously present small-molecule drugs. Such high-level control validates NHRs as a versatile, engineerable platform for programming multi-drug-controlled responses.

synthetic biology↗

bioRxiv2022

Computational pipeline provides mechanistic understanding of Omicron variant of concern neutralizing engineered ACE2 receptor traps

The SARS-CoV-2 Omicron variant, with 15 mutations in Spike receptor binding domain (Spike-RBD), renders virtually all clinical monoclonal antibodies against WT SARS-CoV-2 ineffective. We recently engineered the SARS-CoV-2 host entry receptor, ACE2, to tightly bind WT-Spike-RBD and prevent viral entry into host cells ("receptor traps"). Here we determine cryo-EM structures of our receptor traps in complex with full length Spike. We develop a multi-model pipeline combining Rosetta protein modeling software and cryo-EM to allow interface energy calculations even at limited resolution and identify interface side chains that allow for high affinity interactions between our ACE2 receptor traps and Spike-RBD. Our structural analysis provides a mechanistic rationale for the high affinity (0.53 - 4.2nM) binding of our ACE2 receptor traps to Omicron-RBD confirmed with biolayer interferometry measurements. Finally, we show that ACE2 receptor traps potently neutralize Omicron- and Delta-pseudotyped viruses, providing alternative therapeutic routes to combat this evolving virus.

biochemistry↗

bioRxiv2022

Complete mutational mapping of a GTPase switch in vivo reveals novel allosteric regulation

Allosteric regulation is central to protein function in cellular networks1. However, despite technological advances2,3 most studies of allosteric effects on function are conducted in heterologous environments2,4,5, limiting the discovery of allosteric mechanisms that rely on endogenous binding partners or posttranslational modifications to modulate activity. Here we report an approach that enables probing of new sites of allosteric regulation at residue-level resolution in essential eukaryotic proteins in their native biological context by comprehensive mutational scanning. We apply our approach to the central GTPase Gsp1/Ran. GTPases are highly regulated molecular switches that control signaling, with switching occurring via catalyzed GTP hydrolysis and nucleotide exchange. We find that 28% of 4,315 assayed mutations in Gsp1/Ran are highly deleterious, showing a toxic response identified by our assay as gain-of-function (GOF). Remarkably, a third of all positions enriched for GOF mutations (20/60) are outside the GTPase active site. Kinetic analysis shows that these distal sites are allosterically coupled to the active site, including a novel cluster of sites that alter the nucleotide preference of Gsp1 from GDP to GTP. We describe multiple distinct mechanisms by which allosteric mutations alter Gsp1/Ran cellular function by modulating GTPase switching. Our systematic discovery of new regulatory sites provides a functional map relevant to other GTPases such as Ras that could be exploited for targeting and reprogramming critical biological processes.

biophysics↗

bioRxiv2021

Ligand-induced changes in dynamics mediate long-range allostery in the lac repressor

Allostery, broadly defined as a proteins functional response to distal perturbations, is fundamental to biological regulation. In classical models, allosteric ligand binding produces a defined set of structural changes in the protein, resulting in a different low-energy conformation. Proteins that undergo ligand-induced allostery with few observable structural changes therefore frustrate interpretations by classical models. Here we used hydrogen-deuterium exchange with mass spectrometry (HDX/MS) to map the allosteric effects in a paradigm ligand-responsive allosteric transcription factor, the lac repressor (LacI). X-ray crystal structures of the core domain of LacI bound to different small molecule ligands, or the DNA operator, show less than 1.5 [A] difference in the protein all-atom root-mean-square-deviation (RMSD) between any two structures. Despite this high degree of similarity among static structures, our HDX/MS experiments reveal widespread and unexpected differences in the flexibility of secondary structures in the LacI core domain in each functional state. We propose a model in which ligand binding allosterically switches the functional response of the repressor by selectively changing the dynamics of particular secondary structure elements relative to each other, shifting the conformational ensemble of the protein between mutually incompatible DNA-bound and inducer-bound states. Our model also provides a mechanistic context for the altered functions of thousands of documented LacI mutants. Furthermore, our approach provides a platform for characterizing and engineering allosteric responses in proteins.

biophysics↗

bioRxiv2021

Accurately positioning functional residues with robotics-inspired computational protein design

Accurate positioning of functional residues is critical for the design of new protein functions, but has remained difficult because of the prevalence of irregular local geometries in active sites. Here we introduce two computational methods that build local protein geometries from sequence with atomic accuracy: fragment kinematic closure (FKIC) and loophash kinematic closure (LHKIC). FKIC and LHKIC integrate two approaches: robotics-inspired kinematics of protein backbones and insertion of peptide fragments, and show up to 140-fold improvements in native-like predictions over either approach alone. We then integrate these methods into a new design protocol, pull-into-place (PIP), to position functionally important sidechains via design of new structured loop conformations. We validate PIP by remodeling a sizeable active site region in an enzyme and confirming the engineered new conformations of two designs with crystal structures. The described methods can be applied broadly to the design of many new protein geometries and functions.

bioengineering↗

bioRxiv2021

Ensuring scientific reproducibility in bio-macromolecular modeling via extensive, automated benchmarks

Each year vast international resources are wasted on irreproducible research. The scientific community has been slow to adopt standard software engineering practices, despite the increases in high-dimensional data, complexities of workflows, and computational environments. Here we show how scientific software applications can be created in a reproducible manner when simple design goals for reproducibility are met. We describe the implementation of a test server framework and 40 scientific benchmarks, covering numerous applications in Rosetta bio-macromolecular modeling. High performance computing cluster integration allows these benchmarks to run continuously and automatically. Detailed protocol captures are useful for developers and users of Rosetta and other macromolecular modeling tools. The framework and design concepts presented here are valuable for developers and users of any type of scientific software and for the scientific community to create reproducible methods. Specific examples highlight the utility of this framework and the comprehensive documentation illustrates the ease of adding new tests in a matter of hours.

bioinformatics↗