Scientists are using AI to build mutation-proof antibodies for SARS-CoV-2

Scientists are using AI to build mutation-proof antibodies for SARS-CoV-2

Scientists used AI to create mutation-resistant antibodies that outperformed conventional drug design and offer a powerful new tool against rapidly evolving viruses like SARS-CoV-2.

In a recent study published in the journalScientific reportsThe researchers tested and used several cutting-edge technologies, including machine learning, protein structure modeling, natural language processing and protein sequence language modeling, to design antibodies that can neutralize more than 1,300 SARS-COV-2 strains (including mutants). The design included 64 key mutations in the receptor binding domain of the Spike protein (RBD) and focused on this critical region for viral entry. The antibody templates used were CR3022, casiirivimab (Regen 10,933) and imdevimab (Regen 10,987), which are known monoclonal antibodies against coronaviruses.

Study results showed strong reactivity between the new antibodies and SARS-CoV-2 strains, including Delta (10 antibodies) and Omicron (1 antibody). Notably, 14% of the first batch of antibodies and 40% of the second batch showed “triple cross-binding,” meaning they bound to the receptor binding domain (RBD) of wild-type, Delta, and Omicron strains in ELISA assays. In particular, the present approach was shown to be time-consuming and more cost-effective than traditional structure-based approaches. It can revolutionize future drug design and development, particularly for rapidly evolving pathogens that require frequent modifications to account for their rapid mutation rates. While the study demonstrated adaptability by responding to the emergence of Omicron with a second round of antibody design, its predictive ability for entirely new and unknown future variants is still speculative and has not been directly demonstrated.

background

The severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) that caused the COVID-19 pandemic remains one of the worst in human history, claiming more than 7 million lives since its discovery in late 2019. Fortunately, government-enforced social distancing measures, in combination with widespread anti-COVID-19 immunization interventions, substantially curbed disease spread.

Unfortunately, SARS-COV-2 is a rapidly evolving family of viruses, and new strains resistant to previously approved antibody therapeutics have now emerged. A classic example of this is the resistance demonstrated by strains b.1.427 and b.1.429 to bamlanivimab and etesevimab due to their L452R substitutions.

While ongoing research races are underway to investigate the origins of novel, increasingly resistant SARS-CoV-2 strains, traditional antibody discovery approaches are labor-intensive, inefficient and expensive. By leveraging recent computational and technological advances in artificial intelligence (AI) models such as graphics neural networks (GNNs) and speech-based networks (natural language processing architectures), researchers can design and screen faster and more efficiently than ever before.

About the study

The present study aims to evaluate the viability of AI-based approaches for modeling antibody-antigen binding and screening for antibodies with broad-spectrum neuter capabilities. It demonstrates the application of AI models in rapid discovery of therapeutics to counter future pandemics and highlight their potential across medical fields.

“Our study describes the use of a deep learning model to computationally design effective and broad spectrum mutations against different strains of the virus’s spike protein, and subsequent wet lab experimentation confirms the results.”

The study developed several in-house “antibody affinity maturation AI models.” These models were based on both GNN and voice-based network architectures. The GNN architecture specifically enabled modeling the relationships between amino acid residues as graphs and captures both local and global sequence features relevant to antibody-antigen binding. All models were trained using data sets from four curated datasets: 1. Skempi database, 2. Observed antibody space, 3. Antibody-Bind database (AB-Bind), and 4. UniProt.

After training, model accuracy and performance were evaluated using a combined dataset from Skempi and AB-Bind databases. Accuracy and scalability were evaluated using a Leave-5-Out (L5O) approach.

The CoVID-19 neutralizing antibodies were identified by first collating GISAID database data (1300 SARS-COV-2 strains), selecting templates in silico cross-binding antibody assays and generating in silico mutant libraries (mutations in the template). Machine learning models were then used to discover antibodies with broad-spectrum binding to several of the 1,300 SARS-CoV-2 strains delivered. Because the S1 protein is essential for antigen-antibody binding, antibodies resistant to mutations in viral S1 proteins were identified.

Wet laboratory assays (enzyme-linked immunosorbent assays [ELISAs] and coronavirus cytotoxicity assays) were subsequently performed experimentally to validate computational findings. Following the emergence of Omicron, researchers conducted a second round of computational antibody design to improve antibody affinity specifically against Omicron, demonstrating the reactive adaptability of their approach to emerging variants.

SARS-COV-2 cross-binding sequence selection and virus mutation data curation. Step 2: AI-based antibody binding prediction and cross-variant binding selection for potential candidate sequences for future variants. Step 3: Measure the binding ability of the antibody using an ELISA-based assay; and measuring the neutralization capacity of the antibody using neutralization and cytopathic effect assays (CPE).

Study results

The model accuracy evaluations (conducted using Spearman ranking coefficients) revealed that the graph-based model outperforms the language-based approaches. Remarkably, both graphical and language-based models matched or exceeded the current commercial (non-machine learning) structure-based approach Discovery Studio.

“Unlike Discovery Studio, which uses a physical model derived from primary, secondary and tertiary protein structure to calculate binding affinity, our model learns the mapping between antibody sequence and binding affinity from a large amount of experimental data.”

The benefits of neural network results extended further, with the graph-based model (Pearson = 0.6) outperforming the most conventional in silico approaches (Discovery Studio Pearson = 0.45).

Wet lab experiments confirmed these results. The AI-designed antibody sequences with the highest predicted binding abilities were synthesized. Encouragingly, most of these antibodies were observed to bind and frequently reach a supersaturated state at higher concentrations on B.1, Delta and Omicron SARS-CoV-2 strains.

Coronavirus cytopathic assays showed 10 antibodies capable of neutralizing Vero-E6 host cells infected with Delta strains and one antibody capable of neutralizing cells infected with Omicron strains. However, the study also found that strong binding in ELISA assays did not always correspond to neutralizing ability in cell-based assays, indicating that binding affinity alone does not guarantee neutralization. This discrepancy may be due to differences in the structure of the spike protein in the plate (ELISA) compared to live virus, as well as the specific epitope location and antibody conformation.

It is important to note that while these results are promising, the study was limited to in vitro experiments. No in vivo (animal or human) efficacy studies have been conducted and further investigations such as epitope mapping and conformational dynamics studies will be required for more precise antibody design and validation.

While the study focused on achieving broad neutralization, the authors acknowledge that there may be a trade-off between broad cross-reactivity and therapeutic specificity, which could limit utility in some clinical contexts.

Conclusions

The present study highlights the benefits of using AI structure-free deep neural networks to discover and screen therapeutic antibodies. These computational models significantly outperformed the traditional structure-based non-machine learning structure platforms in terms of cost, efficiency and accuracy. AI models have the added advantage of iteratively improving initially discovered antibodies to compensate for mutations in rapidly evolving pathogens.

“Because our approach combines flexibility and high throughput at low computational costs, this may also be beneficial for other applications of the technology.”

Sources: