Explore antibody repertoire in the era of AI

Abstract

The diverse antibodies of adaptive immunity comprise an antibody repertoire that combats various pathogens. This repertoire is shaped by both intrinsic antibody gene diversification and extrinsic cellular selection. Conversely, an antibody repertoire contains multiple layers of immunological information, including the history of pathogen exposure. High-throughput sequencing-based antibody repertoire cloning approaches have revealed unexpected features of adaptive immunity. However, our understanding of antibody repertoire data is still in its infancy. In this review, we introduce the emerging concepts and discuss the application of deep learning approaches to understanding antibody repertoires. First, we introduce the definition and functional features of antibody clonotype. Next, we review the evolution of antibody clonotypes and discuss potential antibody repertoire-directed vaccination approaches. Lastly, we summarize the application of deep learning in predicting antibody binding, generating specific antibodies, and making immunologic diagnoses. Recently, artificial intelligence (AI) has made revolutionary progress in biology. Leveraging high-dimensional antibody repertoire information, deep learning models have the potential to transform our understanding of antibody repertoire.

1. Introduction

Antibodies are essential components of the immune system that recognize a wide array of pathogens due to their extraordinary diversity in humoral immunity. They function through antigen recognition by their variable domains and through effector functions by their constant domains [1–3] (Figure 1A). The pool of antibodies expressed by an individual is called antibody repertoire, which provides full protection from environmental pathogens. Since the 1980s, antibody research has progressed from screening for specific monoclonal antibodies to exploring the functionality of antibody repertoires [4–7]. In the era of artificial intelligence (AI), the field has advanced to leveraging machine learning (ML) to analyze antibody repertoires in physiological and pathological settings [8–10]. Thus, the rich information embedded in antibody repertoires is beginning to emerge.

The development of intrinsic antibody diversity involves two steps: an antigen-independent step during the early stage of B cell development and an antigen-dependent step during the humoral immune response [3,11,12]. In the former primary diversification step, the variable regions of the heavy and light chains are assembled through V(D)J recombination [13] (Figure 1A). This process involves joining germline gene segments, inserting and/or deleting nucleotides in junctional regions, and pairing the heavy and light chains, which ultimately enables naïve B cells to express up to 10^15 distinct antibodies [13,14]. The assembled variable region is structurally organized into four framework regions (FWRs) and three hypervariable complementarity-determining regions (CDRs). The flexible, loop-structured CDRs primarily participate in antigen binding. Of these, CDR3 spans the junctional region and exhibits the greatest degree of diversity. Some of these naïve antibodies possess an inherent capacity for antigen recognition. In the latter secondary diversification step, antibody genes are further diversified in the antigen-activated mature B cells. With the help from other immune cells, B cells can form germinal center (GC) structures upon antigen stimulation and enhance affinity for their target antigens through the accumulation of somatic hypermutation (SHM) in the variable regions [11,15–17]. In addition, class switching recombination (CSR) further increases the degree of antibody diversity by altering the constant region, conferring different effector functions to the same antibody variable region [17–19]. In this antigen-dependent step, B cell clonal lineages that recognize various pathogens are generated and affinity-matured B cells are developed.