Christian Andrew Choe
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Christian Andrew Choe has a total of 145 co-authors. Here are the top co-authors by number of shared publications:
- Po-Ssu Huang
Department of Bioengineering, Stanford University, Stanford, CA, USA
7 shared publications - Rhiju Das
Department of Biochemistry, Stanford University School of Medicine, Stanford, CA, USA
6 shared publications Eterna Participants
Eterna Massive Open Laboratory, USA
5 shared publicationsJohn J. Nicol
Department of Biochemistry, Stanford University School of Medicine, Stanford, USA
4 shared publications- Jonathan Romano
Eterna Massive Open Laboratory, USA
4 shared publications - Gina El Nesr
Biophysics Program, Stanford University, Stanford, USA
3 shared publications
Latest Publications
- 27 August 2026
- Published by American Association for the Advancement of Science (AAAS) in Science
- Vol. 393 (6814) , 931-937
De novo design of RNA pseudoknots with deep learning
RNA design has been hindered by the limited accuracy of three-dimensional (3D) structure prediction. In this study, we show that intricate RNA structures can be generated with current deep learning tools through accurate de novo design of pseudoknot secondary structures. In an Eterna competition involving 57 pseudoknots, generative artificial intelligence (AI) methods matched experienced human designers in solving most blind challenges, evaluated by single nucleotide–resolution chemical mapping, compensatory mutagenesis, and cryo–electron microscopy. AI-generated molecules with accurate secondary structures formed well-ordered 3D folds stabilized by noncanonical tertiary interactions not modeled during design. Success was guided by an RNet foundation model trained on prior chemical mapping data, suggesting that some difficult RNA design tasks may be tractable without first solving RNA 3D structure prediction.
- 27 August 2026
- Published by American Association for the Advancement of Science (AAAS) in Science
- Vol. 393 (6814) , 931-937
De novo design of RNA pseudoknots with deep learning
RNA design has been hindered by the limited accuracy of three-dimensional (3D) structure prediction. In this study, we show that intricate RNA structures can be generated with current deep learning tools through accurate de novo design of pseudoknot secondary structures. In an Eterna competition involving 57 pseudoknots, generative artificial intelligence (AI) methods matched experienced human designers in solving most blind challenges, evaluated by single nucleotide–resolution chemical mapping, compensatory mutagenesis, and cryo–electron microscopy. AI-generated molecules with accurate secondary structures formed well-ordered 3D folds stabilized by noncanonical tertiary interactions not modeled during design. Success was guided by an RNet foundation model trained on prior chemical mapping data, suggesting that some difficult RNA design tasks may be tractable without first solving RNA 3D structure prediction.
- Open Access
- 15 July 2026
- Published by Springer Nature in Nature Communications
- Vol. 17 (1)
ADAPT-M: a workflow for rapid, quantitative in vitro measurements of enriched protein libraries
Protein-protein interactions underpin most cellular processes, and engineered binders present powerful tools for probing biology and developing novel therapeutics. However, scalable, quantitative characterization of large numbers of candidates remains a major bottleneck. Here we show that ADAPT-M ( A ffinity D etermination by A daptation of P ro T ein binders for M icrofluidics) enables rapid, parallel measurement of binding affinities and dissociation behavior directly from enriched display libraries in under one week, without requiring gene synthesis or hands-on protein purification. Applied to a computationally designed library targeting the SARS-CoV-2 Omicron BA.1 receptor binding domain, ADAPT-M recovered most highly enriched variants and revealed that many display-enriched binders lacked measurable binding in vitro, highlighting limitations of screening alone. ADAPT-M enabled quantitative characterization of dozens of binders in parallel and selection of lead candidates for structural analysis. Unexpectedly, structural and mutational studies revealed that designed binding interfaces were preserved despite engaging alternative epitopes. By bridging screening and scalable in vitro validation, ADAPT-M accelerates protein binder discovery and supports data-driven protein engineering.
- 15 July 2026
- Published by Springer Nature in Nature Communications
- Vol. 17 (1)
Open AccessADAPT-M: a workflow for rapid, quantitative in vitro measurements of enriched protein libraries
Protein-protein interactions underpin most cellular processes, and engineered binders present powerful tools for probing biology and developing novel therapeutics. However, scalable, quantitative characterization of large numbers of candidates remains a major bottleneck. Here we show that ADAPT-M ( A ffinity D etermination by A daptation of P ro T ein binders for M icrofluidics) enables rapid, parallel measurement of binding affinities and dissociation behavior directly from enriched display libraries in under one week, without requiring gene synthesis or hands-on protein purification. Applied to a computationally designed library targeting the SARS-CoV-2 Omicron BA.1 receptor binding domain, ADAPT-M recovered most highly enriched variants and revealed that many display-enriched binders lacked measurable binding in vitro, highlighting limitations of screening alone. ADAPT-M enabled quantitative characterization of dozens of binders in parallel and selection of lead candidates for structural analysis. Unexpectedly, structural and mutational studies revealed that designed binding interfaces were preserved despite engaging alternative epitopes. By bridging screening and scalable in vitro validation, ADAPT-M accelerates protein binder discovery and supports data-driven protein engineering.
- Open Access
De novo design of RNA pseudoknots with deep learning
RNA design has been hindered by the limited accuracy of 3D structure prediction. Here, we show that intricate RNA structures can be generated with current deep learning tools through accurate de novo design of pseudoknot secondary structures. In an Eterna competition involving 57 pseudoknots, generative AI methods matched experienced human designers in solving most blind challenges, evaluated by single-nucleotide-resolution chemical mapping, compensatory mutagenesis, and cryogenic electron microscopy. AI-generated molecules with accurate secondary structures formed well-ordered 3D folds stabilized by noncanonical tertiary interactions not modeled during design. Success was guided by an RNet foundation model trained on prior chemical mapping data, suggesting that some difficult RNA design tasks may be tractable without first solving RNA 3D structure prediction.
Open AccessDe novo design of RNA pseudoknots with deep learning
RNA design has been hindered by the limited accuracy of 3D structure prediction. Here, we show that intricate RNA structures can be generated with current deep learning tools through accurate de novo design of pseudoknot secondary structures. In an Eterna competition involving 57 pseudoknots, generative AI methods matched experienced human designers in solving most blind challenges, evaluated by single-nucleotide-resolution chemical mapping, compensatory mutagenesis, and cryogenic electron microscopy. AI-generated molecules with accurate secondary structures formed well-ordered 3D folds stabilized by noncanonical tertiary interactions not modeled during design. Success was guided by an RNet foundation model trained on prior chemical mapping data, suggesting that some difficult RNA design tasks may be tractable without first solving RNA 3D structure prediction.
- Open Access
- 12 November 2025
- Published by Springer Nature in Nature Chemistry
- Vol. 17 (12) , 1839-1852
Compact RNA sensors for increasingly complex functions of multiple inputs
Designing single molecules that compute general functions of input molecular partners is a major unsolved challenge in molecular design. Here we demonstrate that high-throughput, iterative experimental testing of diverse RNA designs crowdsourced from the online game Eterna yields sensors of increasingly complex functions of input oligonucleotide concentrations. After designing single-input RNA sensors with activation ratios beyond our detection limits, we created logic gates, including challenging XOR and XNOR gates, and sensors that respond to the ratio of two inputs. Finally, we describe the OpenTB challenge, which elicited 85-nucleotide sensors that compute a score for diagnosing active tuberculosis based on the ratio of products of three gene segments. Building on OpenTB design strategies, we created an algorithm, Nucleologic, that produces similarly compact sensors for the three-gene score based on RNA and DNA. These results expand the possibilities for using compact, single-molecule sensors in a range of applications previously constrained by design complexity.
- 12 November 2025
- Published by Springer Nature in Nature Chemistry
- Vol. 17 (12) , 1839-1852
Open AccessCompact RNA sensors for increasingly complex functions of multiple inputs
Designing single molecules that compute general functions of input molecular partners is a major unsolved challenge in molecular design. Here we demonstrate that high-throughput, iterative experimental testing of diverse RNA designs crowdsourced from the online game Eterna yields sensors of increasingly complex functions of input oligonucleotide concentrations. After designing single-input RNA sensors with activation ratios beyond our detection limits, we created logic gates, including challenging XOR and XNOR gates, and sensors that respond to the ratio of two inputs. Finally, we describe the OpenTB challenge, which elicited 85-nucleotide sensors that compute a score for diagnosing active tuberculosis based on the ratio of products of three gene segments. Building on OpenTB design strategies, we created an algorithm, Nucleologic, that produces similarly compact sensors for the three-gene score based on RNA and DNA. These results expand the possibilities for using compact, single-molecule sensors in a range of applications previously constrained by design complexity.
- Open Access
ADAPT-M: A workflow for rapid, quantitative in vitro measurements of enriched protein libraries
Protein-protein interactions underpin most cellular interactions, and engineered binders present powerful tools for probing biology and developing novel therapeutics. One bottleneck in binder generation is the scalable, quantitative characterization of these interactions. We present ADAPT-M ( A ffinity D etermination by A daptation of P ro T ein binders for M icrofluidics), a streamlined workflow that connects yeast surface display (YSD) with in vitro affinity and kinetic measurements using the high-throughput STAMMPPING microfluidic platform. ADAPT-M quantifies K d s and dissociation kinetic parameters for hundreds of enriched protein variants in under one week without requiring hands-on protein purification. We applied ADAPT-M to a computationally designed library targeting the SARS-CoV-2 Omicron BA.1 receptor binding domain, successfully recovering and measuring K d s for most highly enriched YSD variants. Measurements correlate strongly with biolayer interferometry and yeast titration assays. ADAPT-M further enabled selection of lead candidates for structural and mutational analysis, which revealed designed paratopes were preserved despite binding to off-target epitopes. By bridging YSD screening and in vitro validation, ADAPT-M accelerates protein binder discovery and supports data-driven protein engineering.
Open AccessADAPT-M: A workflow for rapid, quantitative in vitro measurements of enriched protein libraries
Protein-protein interactions underpin most cellular interactions, and engineered binders present powerful tools for probing biology and developing novel therapeutics. One bottleneck in binder generation is the scalable, quantitative characterization of these interactions. We present ADAPT-M ( A ffinity D etermination by A daptation of P ro T ein binders for M icrofluidics), a streamlined workflow that connects yeast surface display (YSD) with in vitro affinity and kinetic measurements using the high-throughput STAMMPPING microfluidic platform. ADAPT-M quantifies K d s and dissociation kinetic parameters for hundreds of enriched protein variants in under one week without requiring hands-on protein purification. We applied ADAPT-M to a computationally designed library targeting the SARS-CoV-2 Omicron BA.1 receptor binding domain, successfully recovering and measuring K d s for most highly enriched YSD variants. Measurements correlate strongly with biolayer interferometry and yeast titration assays. ADAPT-M further enabled selection of lead candidates for structural and mutational analysis, which revealed designed paratopes were preserved despite binding to off-target epitopes. By bridging YSD screening and in vitro validation, ADAPT-M accelerates protein binder discovery and supports data-driven protein engineering.
- Open Access
RNAGym: Large-scale Benchmarks for RNA Fitness and Structure Prediction
Understanding RNA structure and predicting the functional consequences of mutations are fundamental challenges in computational biology with broad implications for therapeutic development and synthetic biology. Current evaluation of machine learning-based RNA models suffers from disparate experimental datasets and inconsistent performance assessments across different RNA families. To address these challenges, we introduce RNAGym, a large-scale benchmarking framework specifically designed for three core tasks–RNA fitness, secondary structure, and tertiary structure prediction. The framework integrates extensive datasets, including 70 standardized deep mutational scanning assays covering over a million mutations across diverse RNA types; 901k chemical-mapping reactivity profiles for secondary structure; and 215 diverse tertiary structures curated from the PDB. RNAGym is designed to facilitate a systematic comparison of RNA models, offering an essential resource to enhance the understanding and development of these models.
Open AccessRNAGym: Large-scale Benchmarks for RNA Fitness and Structure Prediction
Understanding RNA structure and predicting the functional consequences of mutations are fundamental challenges in computational biology with broad implications for therapeutic development and synthetic biology. Current evaluation of machine learning-based RNA models suffers from disparate experimental datasets and inconsistent performance assessments across different RNA families. To address these challenges, we introduce RNAGym, a large-scale benchmarking framework specifically designed for three core tasks–RNA fitness, secondary structure, and tertiary structure prediction. The framework integrates extensive datasets, including 70 standardized deep mutational scanning assays covering over a million mutations across diverse RNA types; 901k chemical-mapping reactivity profiles for secondary structure; and 215 diverse tertiary structures curated from the PDB. RNAGym is designed to facilitate a systematic comparison of RNA models, offering an essential resource to enhance the understanding and development of these models.
- Open Access
- 13 December 2024
- Published by Springer Nature in Nature Biotechnology
- p. 1-11
Targeting peptide antigens using a multiallelic MHC I-binding system
Identifying highly specific T cell receptors (TCRs) or antibodies against epitopic peptides presented by class I major histocompatibility complex (MHC I) proteins remains a bottleneck in the development of targeted therapeutics. Here, we introduce targeted recognition of antigen–MHC complex reporter for MHC I (TRACeR-I), a generalizable platform for targeting peptides on polymorphic HLA-A*, HLA-B* and HLA-C* allotypes while overcoming the cross-reactivity challenges of TCRs. Our TRACeR–MHC I co-crystal structure reveals a unique antigen recognition mechanism, with TRACeR forming extensive contacts across the entire peptide length to confer single-residue specificity at the accessible positions. We demonstrate rapid screening of TRACeR-I against a panel of disease-relevant HLAs with peptides derived from human viruses (human immunodeficiency virus, Epstein–Barr virus and severe acute respiratory syndrome coronavirus 2), and oncoproteins (Kirsten rat sarcoma virus, paired-like homeobox 2b and New York esophageal squamous cell carcinoma 1). TRACeR-based bispecific T cell engagers and chimeric antigen receptor T cells exhibit on-target killing of tumor cells with high efficacy in the low nanomolar range. Our platform empowers the development of broadly applicable MHC I-targeting molecules for research, diagnostic and therapeutic applications.
- 13 December 2024
- Published by Springer Nature in Nature Biotechnology
- p. 1-11
Open AccessTargeting peptide antigens using a multiallelic MHC I-binding system
Identifying highly specific T cell receptors (TCRs) or antibodies against epitopic peptides presented by class I major histocompatibility complex (MHC I) proteins remains a bottleneck in the development of targeted therapeutics. Here, we introduce targeted recognition of antigen–MHC complex reporter for MHC I (TRACeR-I), a generalizable platform for targeting peptides on polymorphic HLA-A*, HLA-B* and HLA-C* allotypes while overcoming the cross-reactivity challenges of TCRs. Our TRACeR–MHC I co-crystal structure reveals a unique antigen recognition mechanism, with TRACeR forming extensive contacts across the entire peptide length to confer single-residue specificity at the accessible positions. We demonstrate rapid screening of TRACeR-I against a panel of disease-relevant HLAs with peptides derived from human viruses (human immunodeficiency virus, Epstein–Barr virus and severe acute respiratory syndrome coronavirus 2), and oncoproteins (Kirsten rat sarcoma virus, paired-like homeobox 2b and New York esophageal squamous cell carcinoma 1). TRACeR-based bispecific T cell engagers and chimeric antigen receptor T cells exhibit on-target killing of tumor cells with high efficacy in the low nanomolar range. Our platform empowers the development of broadly applicable MHC I-targeting molecules for research, diagnostic and therapeutic applications.
- Open Access
Ribonanza: deep learning of RNA structure through dual crowdsourcing
Prediction of RNA structure from sequence remains an unsolved problem, and progress has been slowed by a paucity of experimental data. Here, we present Ribonanza, a dataset of chemical mapping measurements on two million diverse RNA sequences collected through Eterna and other crowdsourced initiatives. Ribonanza measurements enabled solicitation, training, and prospective evaluation of diverse deep neural networks through a Kaggle challenge, followed by distillation into a single, self-contained model called RibonanzaNet. When fine tuned on auxiliary datasets, RibonanzaNet achieves state-of-the-art performance in modeling experimental sequence dropout, RNA hydrolytic degradation, and RNA secondary structure, with implications for modeling RNA tertiary structure.
Open AccessRibonanza: deep learning of RNA structure through dual crowdsourcing
Prediction of RNA structure from sequence remains an unsolved problem, and progress has been slowed by a paucity of experimental data. Here, we present Ribonanza, a dataset of chemical mapping measurements on two million diverse RNA sequences collected through Eterna and other crowdsourced initiatives. Ribonanza measurements enabled solicitation, training, and prospective evaluation of diverse deep neural networks through a Kaggle challenge, followed by distillation into a single, self-contained model called RibonanzaNet. When fine tuned on auxiliary datasets, RibonanzaNet achieves state-of-the-art performance in modeling experimental sequence dropout, RNA hydrolytic degradation, and RNA secondary structure, with implications for modeling RNA tertiary structure.
- Open Access
Compact RNA sensors for increasingly complex functions of multiple inputs
Designing single molecules that compute general functions of input molecular partners represents a major unsolved challenge in molecular design. Here, we demonstrate that high-throughput, iterative experimental testing of diverse RNA designs crowdsourced from Eterna yields sensors of increasingly complex functions of input oligonucleotide concentrations. After designing single-input RNA sensors with activation ratios beyond our detection limits, we created logic gates, including challenging XOR and XNOR gates, and sensors that respond to the ratio of two inputs. Finally, we describe the OpenTB challenge, which elicited 85-nucleotide sensors that compute a score for diagnosing active tuberculosis, based on the ratio of products of three gene segments. Building on OpenTB design strategies, we created an algorithm Nucleologic that produces similarly compact sensors for the three-gene score based on RNA and DNA. These results open new avenues for diverse applications of compact, single molecule sensors previously limited by design complexity.
Open AccessCompact RNA sensors for increasingly complex functions of multiple inputs
Designing single molecules that compute general functions of input molecular partners represents a major unsolved challenge in molecular design. Here, we demonstrate that high-throughput, iterative experimental testing of diverse RNA designs crowdsourced from Eterna yields sensors of increasingly complex functions of input oligonucleotide concentrations. After designing single-input RNA sensors with activation ratios beyond our detection limits, we created logic gates, including challenging XOR and XNOR gates, and sensors that respond to the ratio of two inputs. Finally, we describe the OpenTB challenge, which elicited 85-nucleotide sensors that compute a score for diagnosing active tuberculosis, based on the ratio of products of three gene segments. Building on OpenTB design strategies, we created an algorithm Nucleologic that produces similarly compact sensors for the three-gene score based on RNA and DNA. These results open new avenues for diverse applications of compact, single molecule sensors previously limited by design complexity.
- Open Access
Deep Generative Design of Epitope-Specific Binding Proteins by Latent Conformation Optimization
A bstract: Designing de novo binding proteins against arbitrary epitopes using a single scaffold, as seen with natural antibodies, remains an unsolved challenge in protein design. Current design methods are unable to capture the structural dynamics of flexible loops nor search loop conformational space in a principled way. Here we present Sculptor, a deep generative design algorithm that creates epitope-specific protein binders. The Sculptor algorithm constitutes a joint search over the positions, interactions, and generated conformations of a fold, and crafts a backbone to complement a user-specified epitope. Sequences are designed onto generated backbones using a combination of a residue-wise interaction database, a convolutional sequence design module, and Rosetta. Instead of relying on static structures, we capture the local conformational landscape of a single fold using molecular dynamics, and demonstrate that a model trained on such dense conformational data can generate backbones tailor-fit to an epitope. We use Sculptor to design binders against a conserved epitope on venom toxins implicated in neuromuscular paralysis, and obtain a multi-toxin binder from a small naïve library – a promising step towards creating broadly neutralizing binders. This study constitutes a novel application of deep generative modeling for epitope-targeted design, leveraging conformational dynamics to achieve function.
Open AccessDeep Generative Design of Epitope-Specific Binding Proteins by Latent Conformation Optimization
A bstract: Designing de novo binding proteins against arbitrary epitopes using a single scaffold, as seen with natural antibodies, remains an unsolved challenge in protein design. Current design methods are unable to capture the structural dynamics of flexible loops nor search loop conformational space in a principled way. Here we present Sculptor, a deep generative design algorithm that creates epitope-specific protein binders. The Sculptor algorithm constitutes a joint search over the positions, interactions, and generated conformations of a fold, and crafts a backbone to complement a user-specified epitope. Sequences are designed onto generated backbones using a combination of a residue-wise interaction database, a convolutional sequence design module, and Rosetta. Instead of relying on static structures, we capture the local conformational landscape of a single fold using molecular dynamics, and demonstrate that a model trained on such dense conformational data can generate backbones tailor-fit to an epitope. We use Sculptor to design binders against a conserved epitope on venom toxins implicated in neuromuscular paralysis, and obtain a multi-toxin binder from a small naïve library – a promising step towards creating broadly neutralizing binders. This study constitutes a novel application of deep generative modeling for epitope-targeted design, leveraging conformational dynamics to achieve function.