Antibodies are popularly used in therapeutics due to their drug-like properties, such as minimal immunogenic effects and increased target selectivity. The first step for the development of an antibody drug is the identification of a lead molecule. A lead molecule is randomly searched by screening large libraries of varied antibody variants against a target antigen. This step is followed by the molecule development process, which is also a time-intensive step. Hence, it is important to develop quicker methods for antibody drug development.
In the conventional process of developing antibodies, the common techniques used for the initial screening are yeast display, phage display, and immunization coupled with hybridoma screening or B-cell sequencing. The screening step is followed by the molecular development processes.
As stated above, both steps are time-consuming and resource-intensive. Additionally, these methods often result in the development of sub-optimal lead molecules. This shortcoming can be resolved by applying generative artificial intelligence (AI).
Application of the generative AI method to design de novo antibodies in a zero-shot and controllable fashion could significantly reduce the production time of therapeutic antibody development.
As many protein sequence and structure databases are available, applying AI methods to design protein therapeutics could be compelling. The protein sequences are used for model training. Importantly, recent AI-based studies have shown that models trained on these data could be effectively used for the de novo design of specific protein classes.
Even though the clinical efficacy of antibody-based therapeutics has been established, no methods that involve the de novo design of antibodies with wet lab validation are available.
About the study
A recent study, posted in the bioRxiv* preprint server, used generative AI models to develop de novo design antibodies against three distinct targets in a zero-shot fashion. A zero-shot designing method involves designing an antibody to bind to an antigen without follow-up optimization. The newly designed process has been termed de novo, meaning proteins (antibodies) were designed from first principles or from scratch.
The main aim of this study is the experimental validation of a generative AI approach to developing antibodies for therapeutic purposes. The authors designed complementary determining regions (CDRs), which are the key determinants of antibody functions and their binding to the antigen. Importantly, the current study integrated innovative generative modeling ideas with high-throughput experimentation capabilities in the wet lab.
Recent progress in DNA synthesis and sequencing, along with fluorescence-activated cell sorting techniques and E. coli-based antibody expression, has enabled the evaluation of hundreds of thousands of individual designs rapidly and simultaneously.
The current study screened over 400,00 antibody variants that were designed for binding to human epidermal growth factor receptor 2 (HER2) via high-throughput wet lab capabilities. Subsequently, 421 binders were characterized using surface plasmon resonance (SPR). Among these, three proteins were identified that revealed greater binding capacity than the therapeutic antibody trastuzumab.
The binder proteins were highly diverse and could assume variable structural conformations. These proteins also revealed minimal sequence identity to known antibodies. Importantly, these binders exhibited high scores on the previously introduced Naturalness metric, which indicated the possibility of containing favorable developmental characteristics and low immunogenicity.
Compared to conventional processes, the zero-shot nature of the AI-based design has significantly reduced the time to identify lead molecules. In addition, the controllable nature of this model enables the development of therapeutic proteins with optimized immunogenicity and minimal developability risks.
This strategy can be used for more sophisticated antibody development of higher therapeutic relevance, containing specific epitope targeting sites. It is an important step toward future de novo antibody design with comparable or greater binding and natural sequence characteristics.
The high Naturalness scores of the binder proteins obtained from the AI-based model indicate that their designed antibody sequences have higher biological activity. The designed sequences differed significantly from other sequences, indicating that the model can design diverse solution sets of binding molecules.
For instance, the de novo HCDR3 binders exhibited increased backbone conformational variability along with the conservation of important contact positions with the HER2 antigen. One of the advantages of this technique is its generalizability, which was demonstrated by its application to distinct antigens.
To summarize, generative AI-designed antibodies will significantly decrease the timelines for antibody production because this technique enables the production of molecules with favorable characteristics that do not require further optimization. Furthermore, the controllability of AI-designed antibodies will help develop customized molecules for specific disease targets.
This generative design could be further advanced in the future to enable the de novo design of all CDRs and framework regions. This would also enhance the capacity to diversify possible binding solutions. Furthermore, the scale and speed for wet lab validation for AI-generated designs will increase based on the advancements that would enable a decrease in the time and cost of DNA synthesis.
Declaration of Competing Interest
The authors of the research paper (referenced below) are current or former employees, contractors, interns, or executives of Absci Corporation and may hold shares in Absci Corporation. Methods and compositions described in the source manuscript (referenced below) are the subject of one or more pending patent applications.
bioRxiv publishes preliminary scientific reports that are not peer-reviewed and, therefore, should not be regarded as conclusive, guide clinical practice/health-related behavior, or treated as established information.
- Shanehsazzadeh, A. et al. (2023) "Unlocking de novo antibody design with generative artificial intelligence". bioRxiv. doi: 10.1101/2023.01.08.523187. https://www.biorxiv.org/content/10.1101/2023.01.08.523187v1
Posted in: Business / Finance | Medical Science News | Medical Research News
Tags: Antibodies, Antibody, Antigen, Artificial Intelligence, Cell, Cell Sorting, DNA, DNA Synthesis, E. coli, Efficacy, Fluorescence, Growth Factor, Immunization, Molecule, Protein, Receptor, Therapeutics, Trastuzumab, Yeast
Dr. Priyom Bose
Priyom holds a Ph.D. in Plant Biology and Biotechnology from the University of Madras, India. She is an active researcher and an experienced science writer. Priyom has also co-authored several original research articles that have been published in reputed peer-reviewed journals. She is also an avid reader and an amateur photographer.
Source: Read Full Article