【CELL RES】The latest research of Academician Cao Xuetao's team! ACNN-based deep learning framework: predictable broad reactive antibodies against the coronavirus and its omiron variant

This article is the original of the translational medicine network, please indicate the source when reprinting

Author: Mia

The COVID-19 pandemic has been going on for nearly two and a half years, and new variants (VOCs) of SARS-CoV-2 are constantly emerging, urging people to develop broadly neutralizing antibodies
.
Variants such as delta (B.
1.
617.
2 line) and Omicron (BA.
1 and BA.
2) have been reported to exhibit immune evasion
against some of the current therapeutic antibodies.

On September 27, Academician Cao Xuetao's team published a research paper
entitled "Deep learning-based rapid generation of broadly reactive antibodies against SARS-CoV-2 and its Omicron variant" online at Cell Research 。 The paper reports a deep learning framework based on a hollow convolutional neural network (ACNN): a cross-reactive B-cell receptor network (XBCR-net) that can predict broadly reactive antibodies
against SARS-CoV-2 and VOCs directly from single-cell BCR sequences.
ACNN-based XBCR-net has higher accuracy, precision, and recall rates than other frameworks
.

#MOESM1

ACNN-based deep learning framework XBCR-net develops a wide range of reactive antibodies

 01 

Evolving SARS-CoV-2 requires us to quickly predict the binding of antibodies to new variants and develop widely neutralizing antibodies
.
Given the application of deep learning in antibody engineering and optimization, the researchers speculate whether it is possible to quickly design and generate a wide range of reactive antibodies
against SARS-CoV-2 variants through deep learning.

The deep learning framework XBCR-net based on a cavitary convolutional neural network (ACNN) consists of two parts, the first part extracting the relevant features of antibody-antigen interactions through three-branch ACNNs; The second part predicts the probability
of binding an antibody to an antigen (14 different RBD sequences) by the residual structural Multi-Layer Perceptron.

Verify the process

 02 

To assess XBCR-net's suitability to unknown VOCs, the researchers tested
them using RBD and 142 strains of anti-Omicron monoclonal antibodies (including therapeutic antibodies LY-CoV016, AZD-1061, REGN10933, and S309) of the new variants of Omicron (BA.
1, BA.
2, and BA.
4).
。 XBCR-net predicted that 102 of the 142 binders would be positive and 116 of the 142 unbound agents (anti-SARs-CoV-2 antibodies that do not bind to Omicron) would be negative, illustrating the usefulness
of XBCR-net in predicting antibodies bound to Omicron without prior knowledge.

The researchers then used XBCR-net to predict SARS-CoV-2 wild-type (WT) and VOC conjugates in single-cell BCR datasets in COVID patients who had never been infected with the Omicron variant (GSE171703
).
Based on 80% HCDR3 sequence similarity, the researchers identified 153 and 89 clusters from the predicted SARS-CoV-2 conjugates and Omicron variant conjugates
.

As shown, three of these clusters are larger than 8 in size and are expected to cross-react
to both the SARS-CoV-2 and Omicron variants.
Cluster 1 is highly converged on well-studied public clones encoded by IGHV1-58, including the therapeutic antibody Tixagevimab
.
The other two clusters also belong to the common anti-SARS-CoV-2 clone encoding by IGHV3-30 and IGHV4-59 V regions
, respectively.
There are also 4 clusters of antibodies that bind to SARS-CoV-2 but do not bind
to SARS-CoV.

XBCR-net predicts that of the 6743 BCRs, 336 BCRs have cross-reactive for the RBD region of WT SARS-CoV-2 and its VOCs, while only 54 BCRs have cross-reactive
for the RBD region of SARS-CoV.

The researchers selected 10 IGHV3-30 antibodies and 15 antibodies with different IGHV uses from a list of filtered antibodies
.
All 25 strains of monoclonal antibodies had significant binding to
the RBD of WT SARS-CoV-2 compared to the 1 μg/mL negative control antibody.
Consistent with the Omicron validation dataset, 20 of the 25 monoclonal antibodies also cross-reacted
to RBD in the SARS-CoV-2 Omicron variant at 1 μg/mL.
Interestingly, all of the IGHV3-30 antibodies in the study were able to bind to the Omicron variant
.
To further empirically verify XBCR-net, the researchers also applied it to cloned 25-monoclonal antibody SARS-CoV binding
.
The end result also demonstrates that XBCR-net can extrapolate BCR cross-reactivity to emerging variants
without the need for additional training data.

Research summary

 03 

As SARS-CoV-2 continues to evolve, treatments for new variants need to be updated quickly to make clinical decisions
.
From the prediction of XBCR-net, it can be found that monoclonal antibodies derived from IGHV3-30 and IGKV1-13 encoded clusters can bind to SARS-CoV and Omicron variants in addition to SARS-CoV-2, indicating that IGHV3-30, IGKV1-13 encoding clusters can be further developed as widely neutralizing antibodies
against SARS-CoV and SARS-CoV-2.

All in all, XBCR-net can predict a wide range of reactive antibodies against newly discovered SARS-CoV-2 variants without prior knowledge of the new variant-specific antibodies, helping to rapidly generate antibodies
against the SARS-CoV-2 variant and other emerging viruses.

Resources:

#MOESM1

Note: This article is intended to introduce medical research advances and cannot be used as a reference for
treatment options.
For health guidance, please visit a regular hospital
.

Recommended Live/Events

09/27 14:00-17:30 Shanghai

Innovative technology to facilitate the summit forum on accurate diagnosis and treatment of tumors

Scan the code to participate for free

November 01-02 09:00-17:30 Chongqing

The 1st Southwest Single Cell Omics Technology Application Forum

Scan the code to participate for free

November 25-27 09:00-17:30 Shanghai

The 4th Shanghai International Cancer Congress

Scan the code to participate