Ryan L. Kelly
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Ryan L. Kelly has a total of 182 co-authors. Here are the top co-authors by number of shared publications:
- Karl Dane Wittrup
Department of Chemical Engineering Massachusetts Institute of Technology Cambridge Massachusetts USA
6 shared publications Shuang Wu
Department of Antibody Engineering, Genentech, South San Francisco, USA
4 shared publications- Alice Tzeng
Department of Medicine, Brigham and Women's Hospital, Boston, MA
2 shared publications - Benjamin Ethan Steinberg
Neuroscience and Mental Health Program, Hospital for Sick Children, Toronto, Canada
2 shared publications Bettina L. Lee
Department of Physiological Chemistry, Genentech, South San Francisco, USA
2 shared publications- Blayne Amir Sayed
Cell Biology Program, Hospital for Sick Children, Toronto, Canada
2 shared publications
Latest Publications
- Open Access
- 27 November 2025
- Published by Springer Nature in Nature
- Vol. 648 (8093) , E11
Author Correction: Inhibiting membrane rupture with NINJ1 antibodies limits tissue injury
- 27 November 2025
- Published by Springer Nature in Nature
- Vol. 648 (8093) , E11
Open AccessAuthor Correction: Inhibiting membrane rupture with NINJ1 antibodies limits tissue injury
- Open Access
Lab-in-the-loop therapeutic antibody design with deep learning
Therapeutic antibody design is a complex multi-property optimization problem with substantial promise for improvement with the application of machine-learning methods. Towards realizing that promise, we introduce “Lab-in-the-loop,” a new approach that orchestrates state-of-the-art repertoire mining methods, generative machine learning models, multi-task property predictors, active learning ranking and selection, and in vitro experimentation in a semi-autonomous, iterative optimization loop. By automating the design of antibody variants, property prediction, ranking and selection of designs to assay in the lab, and ingestion of in vitro data, we enable an end-to-end approach to developing computationally-informed therapeutic antibody design pipelines. We apply lab-in-the-loop to eleven seed antibodies obtained via animal immunization with four clinically relevant antigen targets: EGFR, IL-6, HER2, and OSM. Over 1,800 unique antibody variants are tested throughout four rounds of iterative optimization identifying 3–100× better binding variants for all targets and 10/11 seeds, with the best binders exceeding 100 pM affinity, demonstrating a process by which end-to-end machine learning can be developed for therapeutic antibody development.
Open AccessLab-in-the-loop therapeutic antibody design with deep learning
Therapeutic antibody design is a complex multi-property optimization problem with substantial promise for improvement with the application of machine-learning methods. Towards realizing that promise, we introduce “Lab-in-the-loop,” a new approach that orchestrates state-of-the-art repertoire mining methods, generative machine learning models, multi-task property predictors, active learning ranking and selection, and in vitro experimentation in a semi-autonomous, iterative optimization loop. By automating the design of antibody variants, property prediction, ranking and selection of designs to assay in the lab, and ingestion of in vitro data, we enable an end-to-end approach to developing computationally-informed therapeutic antibody design pipelines. We apply lab-in-the-loop to eleven seed antibodies obtained via animal immunization with four clinically relevant antigen targets: EGFR, IL-6, HER2, and OSM. Over 1,800 unique antibody variants are tested throughout four rounds of iterative optimization identifying 3–100× better binding variants for all targets and 10/11 seeds, with the best binders exceeding 100 pM affinity, demonstrating a process by which end-to-end machine learning can be developed for therapeutic antibody development.
- 29 January 2024
- Published by Oxford University Press (OUP) in Protein Engineering, Design and Selection
- Vol. 37
High titer expression of antibodies using linear expression cassettes for early-stage functional screening
Antibody discovery processes are continually advancing, with an ever-increasing number of potential binding sequences being identified out of in vivo, in vitro, and in silico sources. In this work we describe a rapid system for high yield recombinant antibody (IgG and Fab) expression using Gibson assembled linear DNA fragments (GLFs). The purified recombinant antibody yields from 1 ml expression for this process are approximately five to ten-fold higher than previous methods, largely due to novel usage of protecting flanking sequences on the 5′ and 3′ ends of the GLF. This method is adaptable for small scale (1 mL) expression and purification for rapid evaluation of binding and activity, in addition to larger scales (30 mL) for more sensitive assays requiring milligram quantities of antibody purified over two columns (Protein A and size exclusion chromatography). When compared to plasmid-based expression, these methods provide nearly equivalent yield of high-quality material across multiple applications, allowing for reduced costs and turnaround times to enhance the antibody discovery process.
- 29 January 2024
- Published by Oxford University Press (OUP) in Protein Engineering, Design and Selection
- Vol. 37
High titer expression of antibodies using linear expression cassettes for early-stage functional screening
Antibody discovery processes are continually advancing, with an ever-increasing number of potential binding sequences being identified out of in vivo, in vitro, and in silico sources. In this work we describe a rapid system for high yield recombinant antibody (IgG and Fab) expression using Gibson assembled linear DNA fragments (GLFs). The purified recombinant antibody yields from 1 ml expression for this process are approximately five to ten-fold higher than previous methods, largely due to novel usage of protecting flanking sequences on the 5′ and 3′ ends of the GLF. This method is adaptable for small scale (1 mL) expression and purification for rapid evaluation of binding and activity, in addition to larger scales (30 mL) for more sensitive assays requiring milligram quantities of antibody purified over two columns (Protein A and size exclusion chromatography). When compared to plasmid-based expression, these methods provide nearly equivalent yield of high-quality material across multiple applications, allowing for reduced costs and turnaround times to enhance the antibody discovery process.
- Top Cited PapersOpen Access
- 17 May 2023
- Published by Springer Nature in Nature
- Vol. 618 (7967) , 1072-1077
Inhibiting membrane rupture with NINJ1 antibodies limits tissue injury
Plasma membrane rupture (PMR) in dying cells undergoing pyroptosis or apoptosis requires the cell-surface protein NINJ1 1 . PMR releases pro-inflammatory cytoplasmic molecules, collectively called damage-associated molecular patterns (DAMPs), that activate immune cells. Therefore, inhibiting NINJ1 and PMR may limit the inflammation that is associated with excessive cell death. Here we describe an anti-NINJ1 monoclonal antibody that specifically targets mouse NINJ1 and blocks oligomerization of NINJ1, preventing PMR. Electron microscopy studies showed that this antibody prevents NINJ1 from forming oligomeric filaments. In mice, inhibition of NINJ1 or Ninj1 deficiency ameliorated hepatocellular PMR induced with TNF plus d -galactosamine, concanavalin A, Jo2 anti-Fas agonist antibody or ischaemia–reperfusion injury. Accordingly, serum levels of lactate dehydrogenase, the liver enzymes alanine aminotransaminase and aspartate aminotransferase, and the DAMPs interleukin 18 and HMGB1 were reduced. Moreover, in the liver ischaemia–reperfusion injury model, there was an attendant reduction in neutrophil infiltration. These data indicate that NINJ1 mediates PMR and inflammation in diseases driven by aberrant hepatocellular death.
- 17 May 2023
- Published by Springer Nature in Nature
- Vol. 618 (7967) , 1072-1077
Top Cited PapersOpen AccessInhibiting membrane rupture with NINJ1 antibodies limits tissue injury
Plasma membrane rupture (PMR) in dying cells undergoing pyroptosis or apoptosis requires the cell-surface protein NINJ1 1 . PMR releases pro-inflammatory cytoplasmic molecules, collectively called damage-associated molecular patterns (DAMPs), that activate immune cells. Therefore, inhibiting NINJ1 and PMR may limit the inflammation that is associated with excessive cell death. Here we describe an anti-NINJ1 monoclonal antibody that specifically targets mouse NINJ1 and blocks oligomerization of NINJ1, preventing PMR. Electron microscopy studies showed that this antibody prevents NINJ1 from forming oligomeric filaments. In mice, inhibition of NINJ1 or Ninj1 deficiency ameliorated hepatocellular PMR induced with TNF plus d -galactosamine, concanavalin A, Jo2 anti-Fas agonist antibody or ischaemia–reperfusion injury. Accordingly, serum levels of lactate dehydrogenase, the liver enzymes alanine aminotransaminase and aspartate aminotransferase, and the DAMPs interleukin 18 and HMGB1 were reduced. Moreover, in the liver ischaemia–reperfusion injury model, there was an attendant reduction in neutrophil infiltration. These data indicate that NINJ1 mediates PMR and inflammation in diseases driven by aberrant hepatocellular death.
- Open Access
- 13 April 2018
- Published by Springer Nature in Nature Communications
- Vol. 9 (1) , 1461
Directed evolution of broadly crossreactive chemokine-blocking antibodies efficacious in arthritis
Chemokine receptors typically have multiple ligands. Consequently, treatment with a blocking antibody against a single chemokine is expected to be insufficient for efficacy. Here we show single-chain antibodies can be engineered for broad crossreactivity toward multiple human and mouse proinflammatory ELR+CXC chemokines. The engineered molecules recognize functional epitopes of ELR+CXC chemokines and inhibit neutrophil activation ex vivo. Furthermore, an albumin fusion of the most crossreactive single-chain antibody prevents and reverses inflammation in the K/BxN mouse model of arthritis. Thus, we report an approach for the molecular evolution and selection of broadly crossreactive antibodies towards a family of structurally related, yet sequence-diverse protein targets, with general implications for the development of novel therapeutics.
- 13 April 2018
- Published by Springer Nature in Nature Communications
- Vol. 9 (1) , 1461
Open AccessDirected evolution of broadly crossreactive chemokine-blocking antibodies efficacious in arthritis
Chemokine receptors typically have multiple ligands. Consequently, treatment with a blocking antibody against a single chemokine is expected to be insufficient for efficacy. Here we show single-chain antibodies can be engineered for broad crossreactivity toward multiple human and mouse proinflammatory ELR+CXC chemokines. The engineered molecules recognize functional epitopes of ELR+CXC chemokines and inhibit neutrophil activation ex vivo. Furthermore, an albumin fusion of the most crossreactive single-chain antibody prevents and reverses inflammation in the K/BxN mouse model of arthritis. Thus, we report an approach for the molecular evolution and selection of broadly crossreactive antibodies towards a family of structurally related, yet sequence-diverse protein targets, with general implications for the development of novel therapeutics.
- Open Access
- 17 August 2017
- Published by Taylor & Francis in mAbs
- Vol. 9 (7) , 1029-1035
Nonspecificity in a nonimmune human scFv repertoire
Efforts to develop effective antibody therapeutics are frequently hampered by issues such as aggregation and nonspecificity, often only detected in late stages of the development process. In this study, we used a high throughput cross-reactivity assay to select nonspecific clones from a naïve human repertoire scFv library displayed on the surface of yeast. Most antibody families were de-enriched; however, the rarely expressed VH6 family was highly enriched among nonspecific clones, representing almost 90% of isolated clones. Mutational analysis of this family reveals a dominant role of CDRH2 in driving nonspecific binding. Homology modeling of a panel of VH6 antibodies shows a constrained β-sheet structure in CDRH2 that is not present in other families, potentially contributing to nonspecificity of the family. These findings confirm the common decision to exclude VH6 from synthetic antibody libraries, and support VH6 polyreactivity as a possible important role for the family in early ontogeny and cause for its overabundance in cases of some forms of autoimmunity.
- 17 August 2017
- Published by Taylor & Francis in mAbs
- Vol. 9 (7) , 1029-1035
Open AccessNonspecificity in a nonimmune human scFv repertoire
Efforts to develop effective antibody therapeutics are frequently hampered by issues such as aggregation and nonspecificity, often only detected in late stages of the development process. In this study, we used a high throughput cross-reactivity assay to select nonspecific clones from a naïve human repertoire scFv library displayed on the surface of yeast. Most antibody families were de-enriched; however, the rarely expressed VH6 family was highly enriched among nonspecific clones, representing almost 90% of isolated clones. Mutational analysis of this family reveals a dominant role of CDRH2 in driving nonspecific binding. Homology modeling of a panel of VH6 antibodies shows a constrained β-sheet structure in CDRH2 that is not present in other families, potentially contributing to nonspecificity of the family. These findings confirm the common decision to exclude VH6 from synthetic antibody libraries, and support VH6 polyreactivity as a possible important role for the family in early ontogeny and cause for its overabundance in cases of some forms of autoimmunity.
- Open Access
- 26 July 2017
- Published by Taylor & Francis in mAbs
- Vol. 9 (7) , 1036-1040
Chaperone proteins as single component reagents to assess antibody nonspecificity
Early stage assays that evaluate monoclonal antibody drug-like properties serve as valuable tools for selection of lead candidates. One liability for clinical development, off-target reactivity, is often assessed by binding to a mixture or panel of noncognate proteins. While robust, these mixes are often ill-defined, and can suffer from issues such as lot-to-lot variability. In this study, we discovered in immunoprecipitation experiments that certain chaperones are present in one of these mixtures;we then explored the use of recombinant chaperone proteins as well-characterized agents to predict antibody nonspecificity. Antibody binding to the heat shock proteins HSP70, HSP90, or trigger factor all served as predictors of cross-interaction propensity, with HSP90 providing the greatest ability to predict antibody clearance rates in mouse. Individual chaperone binding correlates surprisingly closely with binding to complex cell extracts, with the exception of a few “false negatives” (assuming a complex cell extract as the “true” value). As defined reagents, these chaperone reagents present advantages for high throughput assays of nonspecificity.
- 26 July 2017
- Published by Taylor & Francis in mAbs
- Vol. 9 (7) , 1036-1040
Open AccessChaperone proteins as single component reagents to assess antibody nonspecificity
Early stage assays that evaluate monoclonal antibody drug-like properties serve as valuable tools for selection of lead candidates. One liability for clinical development, off-target reactivity, is often assessed by binding to a mixture or panel of noncognate proteins. While robust, these mixes are often ill-defined, and can suffer from issues such as lot-to-lot variability. In this study, we discovered in immunoprecipitation experiments that certain chaperones are present in one of these mixtures;we then explored the use of recombinant chaperone proteins as well-characterized agents to predict antibody nonspecificity. Antibody binding to the heat shock proteins HSP70, HSP90, or trigger factor all served as predictors of cross-interaction propensity, with HSP90 providing the greatest ability to predict antibody clearance rates in mouse. Individual chaperone binding correlates surprisingly closely with binding to complex cell extracts, with the exception of a few “false negatives” (assuming a complex cell extract as the “true” value). As defined reagents, these chaperone reagents present advantages for high throughput assays of nonspecificity.
- Open Access
- 1 December 2016
- Published by Elsevier in Cell Reports
- Vol. 17 (10) , 2503-2511
Temporally Programmed CD8α+ DC Activation Enhances Combination Cancer Immunotherapy
- 1 December 2016
- Published by Elsevier in Cell Reports
- Vol. 17 (10) , 2503-2511
Open AccessTemporally Programmed CD8α+ DC Activation Enhances Combination Cancer Immunotherapy
- Top Cited PapersOpen Access
- 24 October 2016
- Published by Springer Nature in Nature Medicine
- Vol. 22 (12) , 1402-1410
Eradication of large established tumors in mice by combination immunotherapy that engages innate and adaptive immune responses
An immunotherapy consisting of a tumor-antigen targeting antibody, PD-1 blocking antibody, extended half-life recombinant IL-2 and a lymph-node-targeted T cell vaccine mobilized innate and adaptive immunity and eradicated large established tumors in a variety of mouse models. Checkpoint blockade with antibodies specific for cytotoxic T lymphocyte–associated protein (CTLA)-4 or programmed cell death 1 (PDCD1; also known as PD-1) elicits durable tumor regression in metastatic cancer, but these dramatic responses are confined to a minority of patients. This suboptimal outcome is probably due in part to the complex network of immunosuppressive pathways present in advanced tumors, which are unlikely to be overcome by intervention at a single signaling checkpoint. Here we describe a combination immunotherapy that recruits a variety of innate and adaptive immune cells to eliminate large tumor burdens in syngeneic tumor models and a genetically engineered mouse model of melanoma; to our knowledge tumors of this size have not previously been curable by treatments relying on endogenous immunity. Maximal antitumor efficacy required four components: a tumor-antigen-targeting antibody, a recombinant interleukin-2 with an extended half-life, anti-PD-1 and a powerful T cell vaccine. Depletion experiments revealed that CD8+ T cells, cross-presenting dendritic cells and several other innate immune cell subsets were required for tumor regression. Effective treatment induced infiltration of immune cells and production of inflammatory cytokines in the tumor, enhanced antibody-mediated tumor antigen uptake and promoted antigen spreading. These results demonstrate the capacity of an elicited endogenous immune response to destroy large, established tumors and elucidate essential characteristics of combination immunotherapies that are capable of curing a majority of tumors in experimental settings typically viewed as intractable.
- 24 October 2016
- Published by Springer Nature in Nature Medicine
- Vol. 22 (12) , 1402-1410
Top Cited PapersOpen AccessEradication of large established tumors in mice by combination immunotherapy that engages innate and adaptive immune responses
An immunotherapy consisting of a tumor-antigen targeting antibody, PD-1 blocking antibody, extended half-life recombinant IL-2 and a lymph-node-targeted T cell vaccine mobilized innate and adaptive immunity and eradicated large established tumors in a variety of mouse models. Checkpoint blockade with antibodies specific for cytotoxic T lymphocyte–associated protein (CTLA)-4 or programmed cell death 1 (PDCD1; also known as PD-1) elicits durable tumor regression in metastatic cancer, but these dramatic responses are confined to a minority of patients. This suboptimal outcome is probably due in part to the complex network of immunosuppressive pathways present in advanced tumors, which are unlikely to be overcome by intervention at a single signaling checkpoint. Here we describe a combination immunotherapy that recruits a variety of innate and adaptive immune cells to eliminate large tumor burdens in syngeneic tumor models and a genetically engineered mouse model of melanoma; to our knowledge tumors of this size have not previously been curable by treatments relying on endogenous immunity. Maximal antitumor efficacy required four components: a tumor-antigen-targeting antibody, a recombinant interleukin-2 with an extended half-life, anti-PD-1 and a powerful T cell vaccine. Depletion experiments revealed that CD8+ T cells, cross-presenting dendritic cells and several other innate immune cell subsets were required for tumor regression. Effective treatment induced infiltration of immune cells and production of inflammatory cytokines in the tumor, enhanced antibody-mediated tumor antigen uptake and promoted antigen spreading. These results demonstrate the capacity of an elicited endogenous immune response to destroy large, established tumors and elucidate essential characteristics of combination immunotherapies that are capable of curing a majority of tumors in experimental settings typically viewed as intractable.
- Open Access
- 24 July 2016
- Published by Taylor & Francis in mAbs
- Vol. 8 (7) , 1269-1275
Target-independent variable region mediated effects on antibody clearance can be FcRn independent
The importance of the neonatal Fc receptor (FcRn) in extending the serum half-life of monoclonal antibodies (mAbs) is well demonstrated, and has led to the development of multiple engineering approaches designed to alter Fc interactions with FcRn. Recent reports have additionally highlighted the effect of nonspecific interactions on antibody pharmacokinetics (PK), suggesting an FcRn-independent mechanism for mAb clearance. In this report we examine a case study of 2 anti-interleukin-12/23 antibodies, ustekinumab and briakinumab, which share the same target and Fc, but differ in variable region sequences. Ustekinumab displayed near baseline signal in a wide range of early stage developability assays for undesirable protein/protein interactions, while briakinumab showed significant propensity for self- and cross-interactions. This phenotypic difference correlates with faster clearance rates for briakinumab in both human FcRn transgenic and FcRn knockout mice. These findings support a dominant contribution for FcRn-independent clearance for antibodies with high nonspecificity, and highlight a key role for early stage developability screening to eliminate clones with such high nonspecific disposition PK.
- 24 July 2016
- Published by Taylor & Francis in mAbs
- Vol. 8 (7) , 1269-1275
Open AccessTarget-independent variable region mediated effects on antibody clearance can be FcRn independent
The importance of the neonatal Fc receptor (FcRn) in extending the serum half-life of monoclonal antibodies (mAbs) is well demonstrated, and has led to the development of multiple engineering approaches designed to alter Fc interactions with FcRn. Recent reports have additionally highlighted the effect of nonspecific interactions on antibody pharmacokinetics (PK), suggesting an FcRn-independent mechanism for mAb clearance. In this report we examine a case study of 2 anti-interleukin-12/23 antibodies, ustekinumab and briakinumab, which share the same target and Fc, but differ in variable region sequences. Ustekinumab displayed near baseline signal in a wide range of early stage developability assays for undesirable protein/protein interactions, while briakinumab showed significant propensity for self- and cross-interactions. This phenotypic difference correlates with faster clearance rates for briakinumab in both human FcRn transgenic and FcRn knockout mice. These findings support a dominant contribution for FcRn-independent clearance for antibodies with high nonspecificity, and highlight a key role for early stage developability screening to eliminate clones with such high nonspecific disposition PK.