Motivation The precise targeting of antibodies and other protein therapeutics is required for their proper function and the elimination of deleterious off-target effects. Often the molecular structure of a therapeutic target is unknown and randomized methods are used to design antibodies without a model that relates antibody sequence to desired properties. Results Here, we present Ens-Grad, a machine learning method that can design complementarity determining regions of human Immunoglobulin G antibodies with target affinities that are superior to candidates derived from phage display panning experiments. We also demonstrate that machine learning can improve target specificity by the modular composition of models from different experimental campaigns, enabling a new integrative approach to improving target specificity. Our results suggest a new path for the discovery of therapeutic molecules by demonstrating that predictive and differentiable models of antibody binding can be learned from high-throughput experimental data without the need for target structural data. Availability and implementation Sequencing data of the phage panning experiment are deposited at NIH’s Sequence Read Archive (SRA) under the accession number SRP158510. We make our code available at https://github.com/gifford-lab/antibody-2019. Supplementary information Supplementary data are available at Bioinformatics online.

A large unmet medical need exists for safer antithrombotic drugs because all currently approved anticoagulant agents interfere with hemostasis, leading to an increased risk of bleeding. Genetic and pharmacologic evidence in humans and animals suggests that reducing factor XI (FXI) levels has the potential to effectively prevent and treat thrombosis with a minimal risk of bleeding. We generated a fully human antibody (MAA868) that binds the catalytic domain of both FXI (zymogen) and activated FXI. Our structural studies show that MAA868 traps FXI and activated FXI in an inactive, zymogen-like conformation, explaining its equally high binding affinity for both forms of the enzyme. This binding mode allows the enzyme to be neutralized before entering the coagulation process, revealing a particularly attractive anticoagulant profile of the antibody. MAA868 exhibited favorable anticoagulant activity in mice with a dose-dependent protection from carotid occlusion in a ferric chloride–induced thrombosis model. MAA868 also caused robust and sustained anticoagulant activity in cynomolgus monkeys as assessed by activated partial thromboplastin time without any evidence of bleeding. Based on these preclinical findings, we conducted a first-in-human study in healthy subjects and showed that single subcutaneous doses of MAA868 were safe and well tolerated. MAA868 resulted in dose- and time-dependent robust and sustained prolongation of activated partial thromboplastin time and FXI suppression for up to 4 weeks or longer, supporting further clinical investigation as a potential once-monthly subcutaneous anticoagulant therapy.

Prion immunotherapy may hold great potential, but antibodies against certain PrP epitopes can be neurotoxic. Here, we identified > 6,000 PrP‐binding antibodies in a synthetic human Fab phage display library, 49 of which we characterized in detail. Antibodies directed against the flexible tail of PrP conferred neuroprotection against infectious prions. We then mined published repertoires of circulating B cells from healthy humans and found antibodies similar to the protective phage‐derived antibodies. When expressed recombinantly, these antibodies exhibited anti‐PrP reactivity. Furthermore, we surveyed 48,718 samples from 37,894 hospital patients for the presence of anti‐PrP IgGs and found 21 high‐titer individuals. The clinical files of these individuals did not reveal any enrichment of specific pathologies, suggesting that anti‐PrP autoimmunity is innocuous. The existence of anti‐prion antibodies in unbiased human immunological repertoires suggests that they might clear nascent prions early in life. Combined with the reported lack of such antibodies in carriers of disease‐associated PRNP mutations, this suggests a link to the low incidence of spontaneous prion diseases in human populations.

No abstract

The precise targeting of antibodies and other protein therapeutics is required for their proper function and the elimination of deleterious off-target effects. Often the molecular structure of a therapeutic target is unknown and randomized methods are used to design antibodies without a model that relates antibody sequence to desired properties. Here we present a machine learning method that can design human Immunoglobulin G (IgG) antibodies with target affinities that are superior to candidates from phage display panning experiments within a limited design budget.We also demonstrate that machine learning can improve target-specificity by the modular composition of models from different experimental campaigns, enabling a new integrative approach to improving target specificity. Our results suggest a new path for the discovery of therapeutic molecules by demonstrating that predictive and differentiable models of antibody binding can be learned from high-throughput experimental data without the need for target structural data. SignificanceAntibody based therapeutics must meet both affinity and specificity metrics, and existing in vitro methods for meeting these metrics are based upon randomization and empirical testing. We demonstrate that with sufficient target-specific training data machine learning can suggest novel antibody variable domain sequences that are superior to those observed during training. Our machine learning method does not require any target structural information. We further show that data from disparate antibody campaigns can be combined by machine learning to improve antibody specificity.