Antibody structure prediction using interpretable deep learning

Abstract

Therapeutic antibodies make up a rapidly growing segment of the biologics market. However, rational design of antibodies is hindered by reliance on experimental methods for determining antibody structures. In recent years, deep learning methods have driven significant advances in general protein structure prediction. Here, we present DeepAb, a deep learning method for predicting accurate antibody FV structures from sequence. We evaluate DeepAb on two benchmark sets - one balanced for structural diversity and the other composed of clinical-stage therapeutic antibodies - and find that our method consistently outperforms the leading alternatives. Previous deep learning methods have operated as “black boxes” and offered few insights into their predictions. By introducing a directly interpretable attention mechanism, we show that our network attends to physically important residue pairs. For example, in prediction of one CDR H3 residue conformation, the network attends to proximal aromatics and a key hydrogen bonding interaction that constrain the loop conformation. Finally, we present a novel mutant scoring metric derived from network confidence and show that for a particular antibody, all ten of the top-ranked mutations improve binding affinity. These results suggest that this model will be useful for a broad range of antibody prediction and design tasks.

Deciphering the language of antibodies using self-supervised learning

“…A rapid, accurate, paratope prediction method can shed light on binding properties, and is valuable for therapeutic antibody engineering (42,43,49). While we focus on paratope prediction here, the AntiBERTa model can potentially be fine-tuned for other tasks such as antibody structure prediction and humanisation (13,35).…”

“…It has so far proven challenging to predict a BCR's binding specificity and function from its amino acid sequence alone. Most work focuses on analysing the third CDR (CDR3) of the BCR heavy chain, as it is the greatest determinant of binding; however, predicting CDR3 structure and function is notoriously difficult (11)(12)(13). Sequence-dissimilar CDR3s can adopt similar structures (14) and recognise similar regions of a target molecule (15), while small changes in CDR3 sequence can change structure and binding properties (16,17).…”

“…To date, two LMs have been deployed for BCRs and antibodies: DeepAb (13) and Sapiens (35). DeepAb is a bi-directional long short-term memory (LSTM) network that is pre-trained on 100k paired BCR sequences from the Observed Antibody Space (36,37).…”

Progress and challenges for the machine learning-based design of fit-for-purpose monoclonal antibodies

“…At present, the Observed Antibody Space (OAS) 57 database contains over one billion antibody sequences curated from 79 studies, while the iReceptor database contains almost four billion sequences and 6013 repertoires from three remote repositories, 49 research labs, and 60 studies. 53 Such large sequence datasets have been used, for example, to generate latent representations of phenotypically similar antibodies, 58 , 59 prior to training ML models on small-scale structural datasets.

Figure 2.

…”

“… 27 Such efforts have begun to increase the number of datasets to a level where the benchmarking of data-intensive methods, such as deep learning to study antibody-antigen binding at the paratope-epitope level as well as deep learning-based antibody sequence generation, started to become feasible. 27 , 54 More generally, large-scale 3D-atomistic resolution data generation may represent the next major step where abundantly available antibody sequence data will be leveraged to obtain large quantities of antibody-antigen complexes via recent advances in computational structural biology methods such as antibody modeling, 59–63 molecular docking, 64–67 and molecular dynamics. 68 , 69 …”