Biology subjects

Carter, H.

Publications and source records attributed to Carter, H..

At least 19 recordsLinked to original sources

bioRxiv2026

Predicting Neoantigen Immunogenicity from In Vivo Immune Editing

Neoantigen immunogenicity prediction is fundamental to personalized cancer vaccines, tumor-infiltrating lymphocyte (TIL) therapy, and TCR-T cell engineering. Existing computational predictors rely primarily on in-vitro correlates of peptide presentation or models trained against assay-based reactivity, and they are typically validated within a single therapeutic setting. We reasoned that the most direct evidence of neoantigen immunogenicity is longitudinal in-vivo elimination: under immune checkpoint blockade (ICB), subclones bearing recognized neoantigens are selectively depleted over time. Here, we present the Neoantigen Elimination Model (NEMo), a two-compartment (CD8 and CD4) machine learning classifier trained on the in-vivo editing (IVE) of neoantigens across serially sequenced, ICB-treated tumors. By using mechanistically inspired NeoPrecis features designed to capture determinants of immunogenicity beyond MHC binding affinity, NEMo recovered assay-confirmed immunogenic neoantigens across four independent, unseen clinical settings -- pre-existing immunogenicity screening, personalized cancer vaccines, TIL therapy, and a radiotherapy +/- ICB ctDNA cohort -- and stratified progression-free survival more strongly than ELISPOT-confirmed reactivity. The editing signal further revealed an immune-evasion architecture in which oncogenic drivers and neoantigens restricted to lost or silenced HLA alleles are systematically spared from editing.

cancer biology↗

bioRxiv2026

Evaluating Aggregated Gene Level eQTL Scores

Genetic feature engineering, used in methods such as transcriptome-wide association study, supports gene-trait association testing by aggregating single variants into gene-level features predictive of expression. To evaluate how different model architectures, LD filtering thresholds, and variant prioritization methods affect expression prediction quality, we trained over 3 million models and evaluated their performance in independent cohorts. Using the best performing models to impute expression and immunotherapy response as an example trait, we found a significant association with the reactive oxygen species pathway (p=0.032). Our model training workflow will support genetic feature engineering towards improved complex trait modeling.

bioinformatics↗

bioRxiv2026

Cell-Specific Modulation of the Aryl Hydrocarbon Receptor by Kynurenine in Pulmonary Fibrosis Requires Microenvironmental Crosstalk

BackgroundIdiopathic pulmonary fibrosis (IPF) is a progressive, chronic lung disease with limited therapeutic options. Tryptophan metabolism is significantly dysregulated during lung fibrogenesis, with the metabolite kynurenine (kyn) accumulating in lung tissue and driving pathology via the aryl hydrocarbon receptor (AHR). This study evaluates the cell-specific contributions of kyn-mediated AHR signaling across different pulmonary cell types to clarify its role in disease progression. MethodsUsing a murine model of bleomycin-induced pulmonary fibrosis, lung tryptophan metabolites were profiled via liquid chromatography-mass spectrometry. The functional and transcriptomic impacts of kyn administration and AHR modulation were subsequently characterized across three distinct cellular compartments: CD103+ dendritic cells (DCs), fibroblasts, and alveolar epithelial cells (AECs). ResultsKyn levels were elevated in fibrotic lungs, and exogenous kyn selectively exacerbated collagen deposition during the fibrogenic phase rather than altering acute injury. In vitro monocultures of primary lung fibroblasts and AECs revealed negligible functional responses to kyn or AHR inhibition regarding myofibroblast differentiation, migration, or epithelial barrier disruption. Intriguingly, primary tissue-resident CD103+ DCs exhibited a hyperinflammatory, non-canonical AHR signaling profile in vivo. While ex vivo monoculture rapidly reverted these DCs to an anti-inflammatory, canonical AHR state, directly co-culturing DCs with fibrotic primary lung fibroblasts successfully restored the pathogenic, non-canonical signaling phenotype characterized by augmented IL-6 production and suppressed canonical targets. ConclusionsPathogenic AHR signaling in pulmonary fibrosis is highly cell-context dependent and driven by complex cell-cell interactions. Reductionist monocultures fail to replicate tissue- level dendritic cell phenotypes, highlighting the necessity of co-culture models and providing a cautionary note for the systemic clinical use of AHR-targeted therapeutics.

immunology↗

bioRxiv2025

Immunomodulatory nanoparticles elicit antifibrotic monocyte activation to resolve murine pulmonary fibrosis

Organ fibrosis presents a substantial disease burden with few therapeutic options. Innate immunity mediates fibrinogenesis, but also plays a major role in fibrinolysis. Here, we show that immunomodulatory nanoparticles (NPs) can harness this endogenous antifibrotic capacity by catalyzing monocyte activation leading to resolution of bleomycin-induced pulmonary fibrosis in vivo. Cargo-free NPs comprised of the degradable biopolymer poly(lactide-co-glycolide) (PLG) induce a transcriptional shift toward antifibrotic immune activation in profibrotic M2 macrophages (M{Phi}s) in vitro. NPs stimulate M2 M{Phi}s toward a glycolytic, rather than fatty acid oxidative, metabolism; suppress canonical M2 markers like arginase-1 (Arg1) and periostin (Postn); and upregulate collagenases, hyaluronidases and immunoregulatory factors. When delivered intravenously in vivo, NPs reverse established bleomycin-induced pulmonary fibrosis and invert the trajectory of over 1,000 genes from pre- to post-treatment according to bulk RNA-sequencing. NPs also suppress profibrotic signaling and increase expression of repair-associated pathways like peroxisome proliferator-activated receptor gamma (PPAR-{gamma}), nuclear retinoic acid receptor (RAR), vascular endothelial growth factor (VEGF), and sphingolipid signaling in fibrotic lungs. Flow cytometry confirms that NPs induce monocyte recruitment to fibrotic lungs via enhanced integrin expression. Altogether, NPs induce a robust pro-regenerative signature comprised of ECM degradation, inflammation resolution, and tissue repair pathways, concomitant with increased NP+ monocyte recruitment to fibrotic lungs. This work demonstrates that monocytes are not intrinsically profibrotic, but rather, their effects are context-dependent, and they retain a capacity for fibrotic resolution under conditions that can be induced by materials with translational potential. Significance StatementOrgan fibrosis can follow from tissue injury and substantially impairs organ function, but it is incurable and has limited therapeutic options. Myeloid immune cells drive fibrosis onset and progression, but they can also mediate fibrosis resolution. We used polymeric NPs made from clinically translatable biomaterials to harness the antifibrotic capacity of immune cells as a fibrosis therapy. We found that intravenous NPs can reprogram myeloid cells to acquire an antifibrotic phenotype in a mouse model of pulmonary fibrosis. NPs increased myeloid activation and pulmonary recruitment while decreasing fibrosis, showing that directed immune activation, instead of suppression, could be an effective therapeutic strategy for fibrosis.

bioengineering↗

bioRxiv2025

Widespread Epistasis between Cancer Driver Mutations and Allele-Specific Copy Number Variations

Cancer driver mutations alone are often insufficient to fully explain tumorigenesis. We demonstrate that these mutations cooperate with somatic copy number variations (CNVs) in a tissue-specific pattern of genomic epistasis. Analyzing 93,462 tumors, we identified 54 gene-cancer type pairs with significant co-occurrence of somatic mutations and CNVs. Our new Binoculars algorithm, which resolved phased DNA/RNA reads, revealed frequent preferential amplification in oncogenic mutation alleles, including AKT1 p.E17K, BRAF p.V600E, KRAS p.G12C/D/V, NRAS p.Q61K, and a fraction of gain-of-function TP53 p.R175H. Conversely, deletions selectively targeted the reference alleles, leading to loss of heterozygosity of IDH1 p.R132H and tumor suppressor mutations, including CDKN2A and TP53 truncations. Lung cancer patients carrying co-occurrences of somatic mutation-CNVs in TP53 and KRAS showed poorer survival than those carrying the same gene mutations. These findings reveal epistasis of cancer mutations and CNVs at an allelic resolution, suggesting specific genomic events to enhance patient stratification and therapeutic targeting.

genomics↗

bioRxiv2025

Translating clinical gene sequencing into a foundational representation of tumor subtype

While gene sequencing is routine in cancer care, translating sequences into treatment decisions remains a challenge. Here we introduce MutationProjector, an AI foundation model that transforms tumor mutation profiles into a compact representation of cancer subtype, with broad implications for diagnosis and therapy. MutationProjector is pre-trained by integrating genomic alterations from >30,000 tumors with extensive molecular knowledge, yielding a model that accurately reconstructs held-out genetic profiles (demonstrating strong generalization) and determines subtype representations from altered molecular pathways (enabling model interpretability). We evaluate MutationProjector in independent tasks related to prediction of immunotherapy response, prediction of chemotherapy response, and classification of metastasis, recording leading performance in all areas. Each task identifies key biomarkers of interest, including KMT2A and KRAS-STK11 alterations which govern immunotherapy response.

systems biology↗

bioRxiv2025

NeoPrecis: Enhancing Immunotherapy Response Prediction through Integration of Qualified Immunogenicity and Clonality-Aware Neoantigen Landscapes

Despite the transformative impact of cancer immunotherapy, the need for improved patient stratification remains critical due to suboptimal response rates. While neoantigens are central to anti-tumor immunity, current metrics like tumor mutation burden are limited by their neglect of immunogenicity and tumor heterogeneity. We present NeoPrecis, a computational framework designed to refine neoantigen characterization across MHC-I and MHC-II pathways and integrate tumor clonality to improve immunotherapy response prediction. NeoPrecis features an interpretable T-cell recognition model that reveals the critical influence of MHC molecules on TCR recognition beyond mere antigen presentation. Benefit HLA alleles identified through model-driven contribution analysis exhibit significant predictive power for patient outcomes in immune checkpoint inhibitor treatment (melanoma: p-value = 0.04; NSCLC: p-value = 0.01). Applying NeoPrecis to immunotherapy-treated tumors, we show the clonality-aware neoantigen landscape improves response prediction in melanoma and heterogeneous NSCLC, achieving 11% and 20% improvement of AUROC compared to TMB respectively. Heterogeneous NSCLCs, more common among never smokers, retain more subclonal neoantigens due to lower immunoediting pressure, where NeoPrecis better captures the varying prevalence of neoantigens. We propose NeoPrecis as a more comprehensive evaluative framework for neoantigen assessment by incorporating both immunogenicity and tumor clonality, offering insights into the link between collective quality of neoantigen landscapes and immunotherapy response.

bioinformatics↗

bioRxiv2025

GAME: Genomic API for Model Evaluation

The rapid expansion of genomics datasets and the application of machine learning has produced sequence-to-activity genomics models with ever-expanding capabilities. However, benchmarking these models on practical applications has been challenging because individual projects evaluate their models in ad hoc ways, and there is substantial heterogeneity of both model architectures and benchmarking tasks. To address this challenge, we have created GAME, a system for large-scale, community-led standardized model benchmarking on user-defined evaluation tasks. We borrow concepts from the Application Programming Interface (API) paradigm to allow for seamless communication between pre-trained models and benchmarking tasks, ensuring consistent evaluation protocols. Because all models and benchmarks are inherently compatible in this framework, the continual addition of new models and new benchmarks is easy. We also developed a Matcher module powered by a large language model (LLM) to automate ambiguous task alignment between benchmarks and models. Containerization of these modules enhances reproducibility and facilitates the deployment of models and benchmarks across computing platforms. By focusing on predicting underlying biochemical phenomena (e.g. gene expression, open chromatin, DNA binding), we ensure that tasks remain technology-independent. We provide examples of benchmarks and models implementing this framework, and anticipate that the community will contribute their own, leading to an ever-expanding and evolving set of models and evaluation tasks. This resource will accelerate genomics research by illuminating the best models for a given task, motivating novel functional genomic benchmarks, and providing a more nuanced understanding of model abilities.

bioinformatics↗

bioRxiv2025

GenVarLoader: An accelerated dataloader for applying deep learning to personalized genomics

Deep learning sequence models trained on personalized genomics can improve variant effect prediction, however, applications of these models are limited by computational requirements for storing and reading large datasets. We address this with GenVarLoader, which stores personalized genomic data in new memory-mapped formats with optimal data locality to achieve [~]1,000x faster throughput and [~]2,000x better compression compared to existing alternatives.

genomics↗

bioRxiv2024

Dendritic Cell - Fibroblast Crosstalk via TLR9 and AHR Signaling Drives Lung Fibrogenesis

Idiopathic pulmonary fibrosis (IPF) is characterized by progressive scarring and loss of lung function. With limited treatment options, patients succumb to the disease within 2-5 years. The molecular pathogenesis of IPF regarding the immunologic changes that occur is poorly understood. We characterize a role for non-canonical aryl-hydrocarbon receptor signaling (ncAHR) in dendritic cells (DCs) that leads to production of IL-6 and IL-17, promoting fibrosis. TLR9 signaling in myofibroblasts is shown to regulate production of TDO2 which converts tryptophan into the endogenous AHR ligand kynurenine. Mice with augmented ncAHR signaling were created by crossing floxed AHR exon-2 deletion mice (AHR{Delta}ex2) with mice harboring a CD11c-Cre. Bleomycin was used to study fibrotic pathogenesis. Isolated CD11c+ cells and primary fibroblasts were treated ex-vivo with relevant TLR agonists and AHR modulating compounds to study how AHR signaling influenced inflammatory cytokine production. Human datasets were also interrogated. Inhibition of all AHR signaling rescued fibrosis, however, AHR{Delta}ex2 mice treated with bleomycin developed more fibrosis and DCs from these mice were hyperinflammatory and profibrotic upon adoptive transfer. Treatment of fibrotic fibroblasts with TLR9 agonist increased expression of TDO2. Study of human samples corroborate the relevance of these findings in IPF patients. We also, for the first time, identify that AHR exon-2 floxed mice retain capacity for ncAHR signaling.

immunology↗

bioRxiv2024

Integrated germline and somatic features reveal divergent immune pathways driving ICB response

Immune Checkpoint Blockade (ICB) has revolutionized cancer treatment, however mechanisms determining patient response remain poorly understood. Here we used machine learning to predict ICB response from germline and somatic biomarkers and interpreted the learned model to uncover putative mechanisms driving superior outcomes. Patients with higher T follicular helper infiltrates were robust to defects in the class-I Major Histocompatibility Complex (MHC-I). Further investigation uncovered different ICB responses in MHC-I versus MHC-II neoantigen reliant tumors across patients. Despite similar response rates, MHC-II reliant responses were associated with significantly longer durable clinical benefit (Discovery: Median OS=63.6 vs. 34.5 months P=0.0074; Validation: Median OS=37.5 vs. 33.1 months, P=0.040). Characteristics of the tumor immune microenvironment reflected MHC neoantigen reliance, and analysis of immune checkpoints revealed LAG3 as a potential target in MHC-II but not MHC-I reliant responses. This study highlights the value of interpretable machine learning models in elucidating the biological basis of therapy responses. Statement of SignificanceImmune checkpoint blockade works only in a fraction of patients for reasons that are still not fully understood. Our study reveals heterogeneity in the immune responses of ICB responders that correlates with characteristics of the neoantigen landscape. This heterogeneity is accompanied by differences in the duration of clinical benefit as well as by differences as to which immune checkpoint gene serves as a biomarker of ICB response. These findings suggest possible new strategies for improving ICB responses. HighlightsO_LIWe used machine learning to study ICB response across 708 patients from 8 studies across 3 tumor types (melanoma, RCC, and NSCLC). C_LIO_LICombining germline and somatic features improves prediction of ICB response C_LIO_LIInteractions between germline and somatic features reveal mechanisms contributing to ICB sensitivity. C_LIO_LIMHC-I vs. MHC-II reliance implicates LAG3 as a prognostic biomarker in the context of CD4 T cell driven responses. C_LIO_LIMHC-II neoantigen reliant responses provide superior durable clinical benefit in response to ICB. C_LI

cancer biology↗

bioRxiv2023

In vitro validated antibody design against multiple therapeutic antigens using generative inverse folding

Deep learning approaches have demonstrated the ability to design protein sequences given backbone structures [1, 2, 3, 4, 5]. While these approaches have been applied in silico to designing antibody complementarity-determining regions (CDRs), they have yet to be validated in vitro for designing antibody binders, which is the true measure of success for antibody design. Here we describe IgDesign, a deep learning method for antibody CDR design, and demonstrate its robustness with successful binder design for 8 therapeutic antigens. The model is tasked with designing heavy chain CDR3 (HCDR3) or all three heavy chain CDRs (HCDR123) using native backbone structures of antibody-antigen complexes, along with the antigen and antibody framework (FWR) sequences as context. For each of the 8 antigens, we design 100 HCDR3s and 100 HCDR123s, scaffold them into the native antibodys variable region, and screen them for binding against the antigen using surface plasmon resonance (SPR). As a baseline, we screen 100 HCDR3s taken from the models training set and paired with the native HCDR1 and HCDR2. We observe that both HCDR3 design and HCDR123 design outperform this HCDR3-only baseline. IgDesign is the first experimentally validated antibody inverse folding model. It can design antibody binders to multiple therapeutic antigens with high success rates and, in some cases, improved affinities over clinically validated reference antibodies. Antibody inverse folding has applications to both de novo antibody design and lead optimization, making IgDesign a valuable tool for accelerating drug development and enabling therapeutic design. The data generated in this study serve as a useful benchmark of diverse antibody-antigen interactions. We use this data to benchmark self-consistency RMSD (scRMSD), using ABodyBuilder2 [6], ABodyBuilder3 [7], and ESMFold [8], as a metric for assessing binding. We open source the code for IgDesign and the SPR datasets. Code and datasets can be found at https://github.com/AbSciBio/igdesign.

synthetic biology↗

bioRxiv2023

Interface-guided phenotyping of coding variants in the transcription factor RUNX1 with SEUSS

Understanding the consequences of single amino acid substitutions in cancer driver genes remains an unmet need. Perturb-seq provides a tool to investigate the effects of individual mutations on cellular programs. Here we deploy SEUSS, a Perturb-seq like approach, to generate and assay mutations at physical interfaces of the RUNX1 Runt domain. We measured the impact of 115 mutations on RNA profiles in single myelogenous leukemia cells and used the profiles to categorize mutations into three functionally distinct groups: wild-type (WT)-like, loss-of-function (LOF)-like and hypomorphic. Notably, the largest concentration of functional mutations (non-WT-like) clustered at the DNA binding site and contained many of the more frequently observed mutations in human cancers. Hypomorphic variants shared characteristics with loss of function variants but had gene expression profiles indicative of response to neural growth factor and cytokine recruitment of neutrophils. Additionally, DNA accessibility changes upon perturbations were enriched for RUNX1 binding motifs, particularly near differentially expressed genes. Overall, our work demonstrates the potential of targeting protein interaction interfaces to better define the landscape of prospective phenotypes reachable by amino acid substitutions.

bioinformatics↗

bioRxiv2023

Epigenetic Germline Variants Predict Cancer Prognosis and Risk and Distribute Uniquely in Topologically Associating Domains

Cancer is a highly heterogeneous disease caused by genetic and epigenetic alterations in normal cells. A recent study uncovered methylation quantitative trait loci (meQTLs) associated with different levels of local DNA methylation in cancers. Here, we investigated whether the distribution of cancer meQTLs reflected functional organization of the genome in the form of chromatin topologically associated domains (TADs), and evaluated whether cancer meQTLs near known driver genes have the potential to influence cancer risk or progression. At TAD boundaries, we observed differences in the distribution of meQTLs when one or both of the adjacent TADs was transcriptionally active, with higher densities near inactive TADs. Furthermore, we found differences in cancer meQTL distributions in active versus inactive TADs and observed an enrichment of meQTLs in active TADs near tumor suppressors, whereas there was a depletion of such meQTLs near oncogenes. Several meQTLs were associated with cancer risk in the UKBioBank, and we were able to reproduce breast cancer risk associations in the DRIVE cohort. Survival analysis in TCGA implicated a number of meQTLs in 13 tumor types. In 10 of these, polygenic meQTL scores were associated with increased hazard in a CoxPH analysis. Risk and survival-associated meQTLs tended to affect cancer genes involved in DNA damage repair and cellular adhesion and reproduced cancer-specific associations reported in prior literature. In summary, this study provides evidence that genetic variants that influence local DNA methylation are affected by chromatin structure and can impact tumor evolution.

bioinformatics↗

bioRxiv2023

Functional Genomics of Gastrointestinal Escherichia coli Isolated from Patients with Cancer and Diarrhea

We describe the epidemiology and clinical characteristics of 29 patients with cancer and diarrhea in whom Enteroaggregative Escherichia coli (EAEC) was initially identified by GI BioFire panel multiplex. E. coli strains were successfully isolated from fecal cultures in 14 of 29 patients. Six of the 14 strains were identified as EAEC and 8 belonged to other diverse E. coli groups of unknown pathogenesis. We investigated these strains by their adherence to human intestinal organoids, cytotoxic responses, antibiotic resistance profile, full sequencing of their genomes, and annotation of their functional virulome. Interestingly, we discovered novel and enhanced adherence and aggregative patterns for several diarrheagenic pathotypes that were not previously seen when co-cultured with immortalized cell lines. EAEC isolates displayed exceptional adherence and aggregation to human colonoids compared not only to diverse GI E. coli, but also compared to prototype strains of other diarrheagenic E. coli. Some of the diverse E. coli strains that could not be classified as a conventional pathotype also showed an enhanced aggregative and cytotoxic response. Notably, we found a high carriage rate of antibiotic resistance genes in both EAEC strains and diverse GI E. coli isolates and observed a positive correlation between adherence to colonoids and the number of metal acquisition genes carried in both EAEC and the diverse E. coli strains. This work indicates that E. coli from cancer patients constitute strains of remarkable pathotypic and genomic divergence, including strains of unknown disease etiology with unique virulomes. Future studies will allow for the opportunity to re-define E. coli pathotypes with greater diagnostic accuracy and into more clinically relevant groupings.

microbiology↗

bioRxiv2023

Incongruence between transcriptional and vascular pathophysiological cell states

The Notch pathway is a major regulator of transcriptional specification and vascular biology. Previous studies have suggested that targeting the ligand Dll4 or the Notch-receptors results in similar molecular and angiogenesis outcomes. Here, we analyzed single and compound genetic mutants for all Notch signaling members and found very significant differences in the way ligands and receptors regulate vascular homeostasis. Loss of Notch receptors, leads to minor vascular pathology featuring hypermitogenic MAPK-driven cell-cycle arrest and senescence. In contrast, loss of Dll4 triggers a strong Myc-driven switch towards cell proliferation and sprouting and major organ pathology. Targeting of Myc completely suppressed the proliferative and tip-cell angiogenic states induced by Dll4 loss-of-function, however, this did not avoid vascular pathology. Only VEGF blockade prevented the pathology induced by Dll4 loss, but without fully suppressing its transcriptional and metabolic programs. This study shows incongruence between single-cell transcriptional states and adult vascular phenotypes and related pathophysiology.

physiology↗

bioRxiv2023

Melanoma clonal subline analysis uncovers heterogeneity-driven immunotherapy resistance mechanisms

Intratumoral heterogeneity (ITH) can promote cancer progression and treatment failure, but the complexity of the regulatory programs and contextual factors involved complicates its study. To understand the specific contribution of ITH to immune checkpoint blockade (ICB) response, we generated single cell-derived clonal sublines from an ICB-sensitive and genetically and phenotypically heterogeneous mouse melanoma model, M4. Genomic and single cell transcriptomic analyses uncovered the diversity of the sublines and evidenced their plasticity. Moreover, a wide range of tumor growth kinetics were observed in vivo, in part associated with mutational profiles and dependent on T cell-response. Further inquiry into melanoma differentiation states and tumor microenvironment (TME) subtypes of untreated tumors from the clonal sublines demonstrated correlations between highly inflamed and differentiated phenotypes with the response to anti-CTLA-4 treatment. Our results demonstrate that M4 sublines generate intratumoral heterogeneity at both levels of intrinsic differentiation status and extrinsic TME profiles, thereby impacting tumor evolution during therapeutic treatment. These clonal sublines proved to be a valuable resource to study the complex determinants of response to ICB, and specifically the role of melanoma plasticity in immune evasion mechanisms.

cancer biology↗

bioRxiv2023

Deep learning-based codon optimization with large-scale synonymous variant datasets enables generalized tunable protein expression

Increasing recombinant protein expression is of broad interest in industrial biotechnology, synthetic biology, and basic research. Codon optimization is an important step in heterologous gene expression that can have dramatic effects on protein expression level. Several codon optimization strategies have been developed to enhance expression, but these are largely based on bulk usage of highly frequent codons in the host genome, and can produce unreliable results. Here, we develop deep contextual language models that learn the codon usage rules from natural protein coding sequences across members of the Enterobacterales order. We then fine-tune these models with over 150,000 functional expression measurements of synonymous coding sequences from three proteins to predict expression in E. coli. We find that our models recapitulate natural context-specific patterns of codon usage and can accurately predict expression levels across synonymous sequences. Finally, we show that expression predictions can generalize across proteins unseen during training, allowing for in silico design of gene sequences for optimal expression. Our approach provides a novel and reliable method for tuning gene expression with many potential applications in biotechnology and biomanufacturing.

synthetic biology↗