Ankur Annapareddy

Bio: Ankur Annapareddy is an academic researcher from University of Texas at Austin. The author has contributed to research in topics: Epitope & Deep learning. The author has an hindex of 3, co-authored 3 publications receiving 19 citations. Previous affiliations of Ankur Annapareddy include United States Army Research Laboratory.

Discovery of Novel Gain-of-Function Mutations Guided by Structure-Based Deep Learning.

Abstract: Despite the promise of deep learning accelerated protein engineering, examples of such improved proteins are scarce. Here we report that a 3D convolutional neural network trained to associate amino...

Synthetic repertoires derived from convalescent COVID-19 patients enable discovery of SARS-CoV-2 neutralizing antibodies and a novel quaternary binding modality

Abstract: The ongoing evolution of SARS-CoV-2 into more easily transmissible and infectious variants has sparked concern over the continued effectiveness of existing therapeutic antibodies and vaccines. Hence, together with increased genomic surveillance, methods to rapidly develop and assess effective interventions are critically needed. Here we report the discovery of SARS-CoV-2 neutralizing antibodies isolated from COVID-19 patients using a high-throughput platform. Antibodies were identified from unpaired donor B-cell and serum repertoires using yeast surface display, proteomics, and public light chain screening. Cryo-EM and functional characterization of the antibodies identified N3-1, an antibody that binds avidly (K d,app = 68 pM) to the receptor binding domain (RBD) of the spike protein and robustly neutralizes the virus in vitro . This antibody likely binds all three RBDs of the trimeric spike protein with a single IgG. Importantly, N3-1 equivalently binds spike proteins from emerging SARS-CoV-2 variants of concern, neutralizes UK variant B.1.1.7, and binds SARS-CoV spike with nanomolar affinity. Taken together, the strategies described herein will prove broadly applicable in interrogating adaptive immunity and developing rapid response biological countermeasures to emerging pathogens.

Rapid characterization of spike variants via mammalian cell surface display

Abstract: The SARS-CoV-2 spike (S) protein is a critical component of subunit vaccines and a target for neutralizing antibodies. Spike is also undergoing immunogenic selection with clinical variants that increase infectivity and partially escape convalescent plasma. Here, we describe spike display, a high-throughput platform to rapidly characterize glycosylated spike ectodomains across multiple coronavirus-family proteins. We assayed ∼200 variant SARS-CoV-2 spikes for their expression, ACE2 binding, and recognition by thirteen neutralizing antibodies (nAbs). An alanine scan of all five N-terminal domain (NTD) loops highlights a public class of epitopes in the N1, N3, and N5 loops that are recognized by most of the NTD-binding nAbs. Some clinical NTD substitutions abrogate binding to these epitopes but are circulating at low frequencies around the globe. NTD mutations in variants of concern B.1.1.7 (United Kingdom), B.1.351 (South Africa), B.1.1.248 (Brazil), and B.1.427/B.1.429 (California) impact spike expression and escape most NTD-targeting nAbs. However, two classes of NTD nAbs still bind B.1.1.7 spikes and neutralize in pseudoviral assays. B.1.351 and B.1.1.248 include compensatory mutations that either increase spike expression or increase ACE2 binding affinity. Finally, B.1.351 and B.1.1.248 completely escape a potent ACE2 peptide mimic. We anticipate that spike display will accelerate antigen design, deep scanning mutagenesis, and antibody epitope mapping for SARS-CoV-2 and other emerging viral threats.

Prevalent, protective, and convergent IgG recognition of SARS-CoV-2 non-RBD spike epitopes.

Abstract: The molecular composition and binding epitopes of the immunoglobulin G (IgG) antibodies that circulate in blood plasma after severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection are unknown. Proteomic deconvolution of the IgG repertoire to the spike glycoprotein in convalescent subjects revealed that the response is directed predominantly (>80%) against epitopes residing outside the receptor binding domain (RBD). In one subject, just four IgG lineages accounted for 93.5% of the response, including an amino (N)-terminal domain (NTD)-directed antibody that was protective against lethal viral challenge. Genetic, structural, and functional characterization of a multidonor class of "public" antibodies revealed an NTD epitope that is recurrently mutated among emerging SARS-CoV-2 variants of concern. These data show that "public" NTD-directed and other non-RBD plasma antibodies are prevalent and have implications for SARS-CoV-2 protection and antibody escape.

Learning inverse folding from millions of predicted structures

Abstract: We consider the problem of predicting a protein sequence from its backbone atom coordinates. Machine learning approaches to this problem to date have been limited by the number of available experimentally determined protein structures. We augment training data by nearly three orders of magnitude by predicting structures for 12M protein sequences using AlphaFold2. Trained with this additional data, a sequence-to-sequence transformer with invariant geometric input processing layers achieves 51% native sequence recovery on structurally held-out backbones with 72% recovery for buried residues, an overall improvement of almost 10 percentage points over existing methods. The model generalizes to a variety of more complex tasks including design of protein complexes, partially masked structures, binding interfaces, and multiple states.

Protein Sequence Design with a Learned Potential

Abstract: The primary challenge of fixed-backbone protein sequence design is to find a distribution of sequences that fold to the backbone of interest. In practice, state-of-the-art protocols often find viable but highly convergent solutions. In this study, we propose a novel method for fixed-backbone protein sequence design using a learned deep neural network potential. We train a convolutional neural network (CNN) to predict a distribution over amino acids at each residue position conditioned on the local structural environment around the residues. Our method for sequence design involves iteratively sampling from this conditional distribution. We demonstrate that this approach is able to produce feasible, novel designs with quality on par with the state-of-the-art, while achieving greater design diversity. In terms of generalizability, our method produces plausible and variable designs for a de novo TIM-barrel structure, showcasing its practical utility in design applications for which there are no known native structures.

Protein sequence design with a learned potential

Abstract: The task of protein sequence design is central to nearly all rational protein engineering problems, and enormous effort has gone into the development of energy functions to guide design. Here, we investigate the capability of a deep neural network model to automate design of sequences onto protein backbones, having learned directly from crystal structure data and without any human-specified priors. The model generalizes to native topologies not seen during training, producing experimentally stable designs. We evaluate the generalizability of our method to a de novo TIM-barrel scaffold. The model produces novel sequences, and high-resolution crystal structures of two designs show excellent agreement with in silico models. Our findings demonstrate the tractability of an entirely learned method for protein sequence design.