Cell Res takes the progress again! Based on artificial intelligence, Cao Xuetao's team quickly generated a wide range of reactive antibodies against the new crown virus and variants

iNature

The COVID-19 pandemic has been going on for nearly two and a half years, and new variants of interest (VOCs) of SARS-CoV-2 continue to emerge, prompting the development of a wide range of neutralizing antibodies
.
Variants such as delta (B.
1.
617.
2 lineage) and Omicron (BA.
1 and BA.
2) have been reported to show immune evasion
against some current therapeutic antibodies.
Evolving SARS-CoV-2 requires rapid prediction of the binding of antibodies to new variants and the development of a wide range of neutralizing antibodies
.

On September 27, 2022, Cao Xuetao's team from Nankai University/Oxford Research Institute of the Chinese Academy of Medical Sciences 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 research makes use of the advantages of artificial intelligence and big data to independently develop and design an artificial intelligence algorithm that can screen the specific antibodies of the new crown virus and its variants in high throughput: XBCR-net
.

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 virus variants.
In addition, the model developed by the research and subsequent extension algorithms can solve the discovery and optimization of antibodies to cancer, infectious diseases and spontaneous immune diseases, thereby accelerating the development
of therapeutic drugs.

In addition, on September 9, 2022, iNature systematically summarized 12 important achievements/reviews published by Cao Xuetao's team in Cancer Cell, PNAS and STTT in 2022 (click to read).

In the context of widespread COVID-19 vaccination around the world, although the Omicraon variant only causes asymptomatic to moderate clinical symptoms, the spread is extremely widespread, and the case fatality rate remains high
.
At present, even if therapeutic antibody drugs targeting RBD can neutralize the new crown virus and some variants, more than 70% of antibody drugs are still ineffective against new coronavirus variants with higher variations
(such as Omicrane).
The Omichron variant can escape neutralizing antibodies produced by vaccines or previous infections, and BA.
4/5 can even escape antibodies
produced by previous BA.
1 infections.
Therefore, an urgent problem is how to develop therapeutic antibodies
that can target all new coronavirus variants.
Moreover, how to quickly respond to possible variants of the new crown virus in the future and propose targeted treatment methods as soon as possible is also an urgent problem
to be solved.
Artificial intelligence-assisted biology research, especially immunotherapy, is booming and is now a hot spot
in scientific research.
Compared with traditional biological techniques, artificial intelligence has significant advantages in assisting protein structure prediction and target protein screening, which not only greatly reduces the time and cost of experiments, but also improves the explainability and quantitative analysis ability
of biological problems.
This work makes use of the advantages of artificial intelligence and big data, and independently developed and designed an artificial intelligence algorithm that can screen the specific antibodies of the new crown virus and its variants at high throughput: XBCR-net
.
XBCR-net, unlike the transformer-based deep learning algorithm recently published in the journal Immunity, supports simultaneous screening of multi-antigen antibody sequences that predict the probability
of binding an antibody to multiple similar antigens.
The system studied is the first algorithm
that can predict the binding of an antibody to multiple antigens.
As shown in Figure 1, after the amino acid sequences of antigens and antibodies (light chain, heavy chain) are fed into the network, they are first translated by one-hot encoding into a matrix
that can be recognized by the neural network.
After feature extraction by a cavity winding machine, it is pooled globally to the fully connected layer, and the predictive results of antibody antigen binding are given
.
The study leveraged the results of a large number of antibody studies since the COVID-19 pandemic to analyze and summarize nearly 10,000 antibodies specific to the coronavirus subfamily that were fed into XBCR-net for method and supplementary data 1
.
The training results showed that XBCR-net had 80% accuracy in predicting the new coronavirus and its variant-specific antibodies (Figure 1
).
To verify that XBCR-net can predict specific antibodies to new coronavirus variants (not included in the training data), the researchers tested
them with Omickron BA.
1 and BA.
2.
The test results show that the XBCR-net algorithm can accurately predict whether the antibody binds to the Omikron variant (Table 1
).
This is also the first deep learning algorithm
that can predict new antigen-specific antibodies.
To verify whether XBCR-net can assist in therapeutic antibody drug development, the researchers screened single-cell BCR sequencing data from a group of COVID-19 infected patients with XBCR-net and selected the 25 BCR sequences with the highest Omikron specific prediction scores to express purified monoclonal antibodies
.
It was experimentally proven that 20 of these antibodies were specific to Omikron
.
The researchers selected three antibodies (XBN-10, 11, 22) for further experimental verification, which have a very strong affinity and neutralization ability with the new crown virus and most variants (Figure 1).

On this basis, the study found the characteristics of the predicted broad-spectrum antibody of the new coronavirus subfamily through the analysis of life and credit, and summarized
it.
This discovery could also guide the development
of subsequent broad-spectrum neutralizing antibodies.
This study is the first project to use deep learning to screen out broad-spectrum neutralizing antibodies, and the resulting broad-spectrum neutralizing antibodies have a very high affinity, which proves the reliability
of the research methodology.
On this basis, the researchers found that the monoclonal antibody also has a high affinity for the SARS virus, suggesting that the neutralization ability
of the broad-spectrum antibody for the SARS virus and the new coronavirus and its variants can be further improved through antibody engineering optimization.
This work is of great significance
for the control of variants that will emerge in the future.
The deep learning algorithm designed by the researchers can predict the binding of multi-antigen antibodies, making it more widely
available.
Compared to the current model, which only targets single antigen prediction, the model of this study can predict multi-antigen binding, thereby assisting in the discovery
of virus broad-spectrum neutralizing antibodies.
For HIV and influenza viruses, researchers can also train XBCR-net with HIV and influenza-virus-specific antibodies to predict broad-spectrum HIV or influenza virus neutralizing antibodies
.
Based on the current model, the researchers are next preparing to upgrade the basic network, dividing the network into two parts (base and fine-tuning
).
In the base section, more than half a billion pieces of data are used to train a general model, so that one model predicts the action of all antibody antigens
.
The construction of this model can quickly promote the discovery of broad-spectrum neutralizing antibodies for different viruses and respond to emerging infectious diseases
in a short period of time.
Not only that, this calculation method can also be applied to the development of specific antibodies for different cancers, solving problems such as
long development cycle of cancer-specific antibodies and poor specificity of antibodies produced by patients.
In summary, the model developed by the research and subsequent extended algorithms can address the discovery and optimization of antibodies to cancer, infectious diseases and spontaneous immune diseases, thereby accelerating the development
of therapeutic drugs.

Reference message:
—END—Content is [iNature]