It is highly likely that the COVID-19 pandemic will last longer due to the continuous evolution of SARS-CoV-2 and the emergence of variants that are resistant to vaccines and neutralizing antibodies. The ACE2 receptor-binding domain (RBD), a prime target site for neutralizing antibodies, plays a significant role in the selection and emergence of SARS-CoV-2 variants. A machine-learning-guided protein engineering technique known as deep mutational learning (DML) is developed here. It is used to accurately predict how ACE2 binding and antibody escape will be affected by a vast sequence space of combinatorial mutations consisting of billions of RBD variants. A multitude of evolutionary pathways could lead to SARS-CoV-2 variants that are highly diverse. A DML profile may guide the development of therapeutic antibody treatments and vaccines for COVID-19 by predicting current and future variants, even highly mutated variants like Omicron.  

The original founder strain, Wu-Hu-1, has been replaced by variants of SARS-CoV-2, which are more transmissible and/or immune-evading (antibody escape). It is common for these variants to have at least one mutation in the RBD, which can directly affect their ability to bind to ACE2. The N501Y mutation is present in Alpha (B.1.1.7), Beta (B.1.351), and Gamma (P.1) variants, suggesting that these mutations may have been selected for by higher affinity binding to ACE2. The binding and neutralization of neutralizing antibodies to variants are often reduced, especially with monoclonal antibody therapeutics and vaccines, including ones induced from Wu-Hu-1 spike proteins. It has been found that many neutralizing antibodies to SARS-CoV-2 share structural and sequence characteristics, which has led to their classification into four standard classes based on their targeted RBD epitopes.  

In addition to escaping several classes of neutralizing antibodies bound to a variety of RBD epitopes, Omicron variants have proven to have severe immune evasion properties. As a result, nearly all clinically approved antibody therapies are no longer neutralizing Omicron (BA.1 / B.1.1.529), including Eli Lilly’s (LY-CoV16+LyCoV555) multi-class antibodies, Regeneron’s (REGN10933+REGN10987) and AstraZeneca’s (AZD8895+AZD1061). The FDA has subsequently revoked the approval of (Eli Lilly and Regeneron) or modified the dosage of (AstraZeneca). Despite having crossreactivity to SARS-CoV-2, S309 has reduced but still potent neutralizing activity against Omicron (BA.1), which was originally isolated from B cells of a patient infected with SARS-CoV-1. This is most likely due to its binding to a highly conserved epitope found across all genetically diverse sarbecoviruses.  

The clinical approval for S309 was lost in early 2022 as the Omicron sublineage BA.2 showed substantial escape from S309. The only clinically approved antibody, LY-CoV1404 (bebtelovimab), was discovered by binding to the original Wu-Hu-1 RBD in August 2022 and appeared to be highly effective in neutralizing BA.1 and BA.2 variants as well as emerging sublineages BA.4 and BA.5. In an effort to determine the impacts of single-position substitutions on binding to ACE2 and escape from monoclonal or serum antibodies, Bloom and colleagues performed yeast surface display and deep mutational scanning (DMS) on the entire 201 amino acid RBD of SARS-CoV-2. Despite DMS’s ability to detect single mutation profiles of the RBD, several previously circulating variants (e.g., Beta, Gamma, and Delta) possess multiple mutations of the RBD, and Omicron and its sublineages possess 21 RBD mutations (BA.1.12.1), underscoring the urgent need to investigate how combinatorial mutations affect the genome.  

Using deep sequencing and machine learning, deep mutational learning integrates yeast display screening of RBD mutagenesis libraries. Through DML, we comprehensively examine combinational RBD mutations in relation to ACE2 binding and escape from a panel of neutralizing antibodies, including clinical therapeutics and other broadly neutralizing antibodies. Mutations in RBD that retain ACE2 binding while escaping many different types of neutralizing antibodies are revealed by DML. Furthermore, DML may be useful in evaluating and selecting the most promising antibody therapeutics for clinical development by predicting their robustness to prospective SARSCoV-2 variants. 

Deep Mutational Learning for ACE2 Mutations Prediction
Image description: Various steps performed to depict the bining of ACE2 and antibodies to RBD variants 
Image source:

The emergence of new variants of SARS-CoV-2 poses a perpetual risk because it is endemic and constantly evolving. DML is a machine learning-guided protein engineering method that is used in this study to determine whether combinatorial mutations in the SARS-CoV-2 RBD affect ACE2 binding and antibody escape. The DML method makes highly accurate predictions across billions of sequences of RBD variants using machine learning models trained on thousands of labeled variants obtained from library screening, much greater than what can be achieved by experimental screening alone.  

Improved machine learning models, perhaps including structural knowledge, could improve prediction across longer lengths of the RBD by combining future library designs, more elaborate screening strategies, and more elaborate screening thresholds. Recent studies, for example, used machine learning to predict the apparent affinities of ACE2 binding to single and multiple mutations of RBD. A second important consideration is that the DML library was based on the Wu-Hu-1 RBD, whereas Omicron or its sublineages are common worldwide.  

The DML libraries were designed by rationally using DMS data published previously on the RBD in order to improve the probability of isolating variants that retain binding to ACE2, which is crucial for generating sufficient machine learning training data. Because of this, some positions were left fixed since the single mutation DMS data suggested that mutations at these positions would result in loss of binding to ACE2. It was generally effective in covering the mutational sequence space of most SARS-CoV-2 variants with this approach, but it also led to some limitations since some of the fixed positions in the library design (e.g., 486, 496) are mutated in Omicron and its sublineages, further emphasizing the significance of epistasis or combinatorial mutations for SARS-CoV-2. This position is highly associated with antibody escape, including to BA.1-specific antibodies, as it is mutated in Omicron BA.4 and BA.5 variants (F486V). In order to design future mutagenesis libraries for DML, epistatic effects will need to be taken into account, and not only single-mutation DMS data should be used. The RBD was also divided into three regions so that RBM-1, -2, and -3 libraries could be constructed to limit the size of the combinatorial libraries. In Omicron, mutations in RBM-1 and RBM-2 leave us unable to investigate the epistatic effects of mutations across RBM sites.

Article Source: Reference Paper

Learn More:

Top Bioinformatics Books

Learn more to get deeper insights into the field of bioinformatics.

Top Free Online Bioinformatics Courses ↗

Freely available courses to learn each and every aspect of bioinformatics.

Latest Bioinformatics Breakthroughs

Stay updated with the latest discoveries in the field of bioinformatics.

Website | + posts

Srishti Sharma is a consulting Scientific Content Writing Intern at CBIRT. She's currently pursuing M. Tech in Biotechnology from Jaypee Institute of Information Technology. Aspiring researcher, passionate and curious about exploring new scientific methods and scientific writing.


Please enter your comment!
Please enter your name here