Optimization of therapeutic antibodies by predicting antigen specificity from antibody sequence via deep learning

Mason, Derek M; Friedensohn, Simon; Weber, Cédric R.; Jordi, Claudio; Wagner, Bastian; Meng, Simon M; Ehling, Roy; Bonati, Lucia; Dahinden, Jan; Gainza, Pablo; Correia, Bruno E.

doi:10.1038/s41551-021-00699-9

Cited by 213 publications

(304 citation statements)

References 54 publications

Supporting

Mentioning

298

Contrasting

Order By: Relevance

“…Epitope immunogenicity is captured by binding hotspots (Figure 2D), mirroring experimentally observed or predicted epitope clusters (71,80) and hotspots were overlapping with the experimental antibody structures most of the time (Figure S7E). Similar CDRH3 sequences were found to bind different antigens (Figure S11D-H), as found in experimental studies where binders and non-binders to the same antigen could be very similar (32), and very different sequences bound to the same antigen (Figure S11A-C, ( 32)). Previously, we have shown that the structural lattice model used in Absolut!…”

Section: Discussionsupporting

confidence: 79%

Unconstrained generation of synthetic antibody-antigen structures to guide machine learning methodology for real-world antibody specificity prediction

Akbar

Frank

et al. 2021

Preprint

View full text Add to dashboard Cite

Machine learning (ML) is a key technology to enable accurate prediction of antibody-antigen binding, a prerequisite for in silico vaccine and antibody design. Two orthogonal problems hinder the current application of ML to antibody-specificity prediction and the benchmarking thereof: (i) The lack of a unified formalized mapping of immunological antibody specificity prediction problems into ML notation and (ii) the unavailability of large-scale training datasets. Here, we developed the Absolut! software suite that allows the parameter-based unconstrained generation of synthetic lattice-based 3D-antibody-antigen binding structures with ground-truth access to conformational paratope, epitope, and affinity. We show that Absolut!-generated datasets recapitulate critical biological sequence and structural features that render antibody-antigen binding prediction challenging. To demonstrate the immediate, high-throughput, and large-scale applicability of Absolut!, we have created an online database of 1 billion antibody-antigen structures, the extension of which is only constrained by moderate computational resources. We translated immunological antibody specificity prediction problems into ML tasks and used our database to investigate paratope-epitope binding prediction accuracy as a function of structural information encoding, dataset size, and ML method, which is unfeasible with existing experimental data. Furthermore, we found that in silico investigated conditions, predicted to increase antibody specificity prediction accuracy, align with and extend conclusions drawn from experimental antibody-antigen structural data. In summary, the Absolut! framework enables the development and benchmarking of ML strategies for biotherapeutics discovery and design.

show abstract

Section: Discussionsupporting

confidence: 79%

Unconstrained generation of synthetic antibody-antigen structures to guide machine learning methodology for real-world antibody specificity prediction

Akbar

Frank

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

“…S3 for a comparison of the distribution of all 7 million CDR-H3 sequences ["native"] vs the top 1% affinity ones ["native_top"]). Reference CNN model trained on experimental human epidermal growth factor 2 (HER2) CDR-H3 binder and non-binder sequences CDR-H3 sequences that bind (binders) and do not bind (non-binders) to HER2 were obtained from Mason and colleagues (31). They used a total of experimentally validated 11 300 HER2-binders and 27 539 validated HER-2 non-binders to train a convolutional neural network (CNN) classifier that assigns an HER-2 binding probability to a given input CDR-H3 sequence.…”

Section: Methodsmentioning

confidence: 99%

“…One solution to this challenge are experimentally validated ML-classifiers that can screen the potential sequence space for binders. One such classifier for HER-2 binders was previously developed by Mason and colleagues (31). Briefly, this CNN-based classifier classifies CDR-H3 amino acid sequences for their potential to bind HER2; all sequences annotated with a binding probability of p>0.5 are considered binders.…”

Section: Experimental Validation Of Antibody-design Conclusion Drawn From ML Training On Simulated Antibody-antigen Binding Datamentioning

confidence: 99%

“…resolves the current problems of in silico validation of generated sequences (29,30). And (ii) an experimentally validated deep learning classifier that was trained on binders and non-binders to epidermal growth factor 2 (HER2) (31). Our work provides the basis for the ML-driven design of fit-for-purpose antibodies with respect to binding affinity, epitope, and developability.…”

Section: Introductionmentioning

confidence: 98%

See 1 more Smart Citation

In silico proof of principle of machine learning-based antibody design at unconstrained scale

Akbar

Weber

et al. 2021

Preprint

Self Cite

View full text Add to dashboard Cite

Generative machine learning (ML) has been postulated to be a major driver in the computational design of antigen-specific monoclonal antibodies (mAb). However, efforts to confirm this hypothesis have been hindered by the infeasibility of testing arbitrarily large numbers of antibody sequences for their most critical design parameters: paratope, epitope, affinity, and developability. To address this challenge, we leveraged a lattice-based antibody-antigen binding simulation framework, which incorporates a wide range of physiological antibody binding parameters. The simulation framework enables both the computation of antibody-antigen 3D-structures as well as functions as an oracle for unrestricted prospective evaluation of the antigen specificity of ML-generated antibody sequences. We found that a deep generative model, trained exclusively on antibody sequence (1D) data can be used to design native-like conformational (3D) epitope-specific antibodies, matching or exceeding the training dataset in affinity and developability variety. Furthermore, we show that transfer learning enables the generation of high-affinity antibody sequences from low-N training data. Finally, we validated that the antibody design insight gained from simulated antibody-antigen binding data is applicable to experimental real-world data. Our work establishes a priori feasibility and the theoretical foundation of high-throughput ML-based mAb design.HighlightsA large-scale dataset of 70M [3 orders of magnitude larger than the current state of the art] synthetic antibody-antigen complexes, that reflect biological complexity, allows the prospective evaluation of antibody generative deep learningCombination of generative learning, synthetic antibody-antigen binding data, and prospective evaluation shows that deep learning driven antibody design and discovery at an unconstrained level is feasibleTransfer learning (low-N learning) coupled to generative learning shows that antibody-binding rules may be transferred across unrelated antibody-antigen complexesExperimental validation of antibody-design conclusions drawn from deep learning on synthetic antibody-antigen binding dataGraphical abstractWe leverage large synthetic ground-truth data to demonstrate the (A,B) unconstrained deep generative learning-based generation of native-like antibody sequences, (C) the prospective evaluation of conformational (3D) affinity, paratope-epitope pairs, and developability. (D) Finally, we show increased generation quality of low-N-based machine learning models via transfer learning.

show abstract

“…of the immune repertoire of individuals, maps them to sequence representations, a pooling function generates a repertoire representation and an output network predicts the disease class of the repertoire (94); (iii) predict the antigen or epitope to which T-and B-cell receptors can bind: Jurtz et al created NetTCR, a method based on convolutional neural networks trained on TCR CDR3 sequences and binding peptide sequences that can filter out from TCR AIRR-seq data peptides that bind to a certain TCR(95), while the recent work from Mason and colleagues showed that deep learning models can generate in vitro antibodies retaining antigen-binding properties(96) and predict antibody-antigen binding(97). ML was also recently used to understand the differences in BCR between normal and tumor-affected tissues (109), which could be diagnostically used in oncology.…”

mentioning

confidence: 99%

Prospective Artificial Intelligence to Dissect the Dengue Immune Response and Discover Therapeutics

2021

View full text Add to dashboard Cite

Dengue virus (DENV) poses a serious threat to global health as the causative agent of dengue fever. The virus is endemic in more than 128 countries resulting in approximately 390 million infection cases each year. Currently, there is no approved therapeutic for treatment nor a fully efficacious vaccine. The development of therapeutics is confounded and hampered by the complexity of the immune response to DENV, in particular to sequential infection with different DENV serotypes (DENV1–5). Researchers have shown that the DENV envelope (E) antigen is primarily responsible for the interaction and subsequent invasion of host cells for all serotypes and can elicit neutralizing antibodies in humans. The advent of high-throughput sequencing and the rapid advancements in computational analysis of complex data, has provided tools for the deconvolution of the DENV immune response. Several types of complex statistical analyses, machine learning models and complex visualizations can be applied to begin answering questions about the B- and T-cell immune responses to multiple infections, antibody-dependent enhancement, identification of novel therapeutics and advance vaccine research.

show abstract

Optimization of therapeutic antibodies by predicting antigen specificity from antibody sequence via deep learning

Cited by 213 publications

References 54 publications

Unconstrained generation of synthetic antibody-antigen structures to guide machine learning methodology for real-world antibody specificity prediction

Unconstrained generation of synthetic antibody-antigen structures to guide machine learning methodology for real-world antibody specificity prediction

In silico proof of principle of machine learning-based antibody design at unconstrained scale

Prospective Artificial Intelligence to Dissect the Dengue Immune Response and Discover Therapeutics

Contact Info

Product

Resources

About