Computational Antibody Papers

Filter by tags
protein design
Filter by published year
All
TitleKey points

  • 2025-04-28

    Scaling unlocks broader generation and deeper functional understanding of proteins

    • protein design
    • non-antibody stuff
    • Novel protein generative language model — ProGen3
    • The model can do autoregressive generation N-to-C, C-to-N, and also supports span infilling.
    • The architecture is a Transformer with a Sparse Mixture of Experts (MoE), activating about 27% of parameters per forward pass to improve computational efficiency.
    • They studied how sampling affects training by trying different family-level weighting schemes. Uniform sampling across families (where small and large families have equal chance) gave better diversity and generalization, while unmodified sampling (letting big families dominate) performed worst.
    • They validated the models by showing that generated proteins express well in wet lab experiments (split-GFP assays, spanning both highly novel and moderately novel sequence spaces).
    • They used a large thermostability dataset to align model predictions to stability. This alignment is not standard fine-tuning — instead, preference optimization was applied, teaching the model to prefer sequences predicted to have higher stability. Upon experimental validation, aligned models indeed produced proteins with higher expression and stability.

  • 2025-03-31

    Precise, Specific, and Sensitive De Novo Antibody Design Across Multiple Cases

    • protein design
    • Benchmarking of a proprietary antibody design algorithm.
    • The method generates novel antibodies against a target, for a specific epitopic constraint & can be used to re-design antibodies.
    • Altogether they find good affinity scfv binders for six targets for which they found a complex in the PDB, like PD1 and Her2.
    • The de novo antibody design methods were computationally benchmarked against a curated set of 32 experimentally resolved antibody–antigen complexes using metrics like the G-pass rate and orientation recovery (measured by Fw RMSD). This allowed the authors to compare their method (across different versions) against other approaches.
    • They compare against RFAntibody and dyMEAN but in the computational tasks - reproducing the orientation of an existing antibody.
    • Several rounds of biopanning are employed to enrich for high-affinity, target-specific binders from a pre-designed library and do not involve the introduction of new mutations.
    • They benchmark the developability properties such as monomericity, yield and polyreactivity to show that their antibodies have good properties.
    • They demonstrate that most of their designed binders have less than 50% H3 sequence identity to antibodies in the PDB.
    • Computational details and binders are not given.

  • 2025-03-21

    NanoDesigner: Resolving the complex–CDR interdependency with iterative refinement

    • protein design
    • nanobodies
    • Novel pipeline for computational protein design of nanobodies
    • Several tools are collated and adjusted to nanobody case - IgFold for structure prediction, HDOCK for docking and ABDESIGN, DiffAb and dyMEAN for backbone/sequence prediction.
    • They chiefly perform computational validation showing the performance on the RMSD/DockQ (re-docking) and the amino acid recovery. Results indicate that focusing on nanobodies provides benefit.
    • The entire pipeline can be used for de novo design and optimization.

  • 2025-03-21

    NanoGen: A High-affinity Nanobody Generation Model with Guided Diffusion

    • protein design
    • nanobodies
    • Novel method for nanobody sequence re-design using quite a small network.
    • The model was pre-trained using a large-scale collection of nanobody sequences from the INDI dataset, heavy-chain antibody sequences from the OAS, and antibody complex structures from SabDab. For fine-tuning, affinity data was generated by with 17,500 nanobody–antigen interaction data points—7,500 generated via the ANTIPASTI model and 10,000 through random pairing—with a CD45 patent dataset used for testing. So all computational predictions are not real affinity points.
    • NanoGen uses a two-stage training framework with a shared encoder-decoder architecture based on CNN layers that learns sequence patterns via a Masked Language Modeling task. During generation, a guided discrete diffusion process, augmented with Discrete Bayesian Optimization, is employed to refine the sequence outputs for enhanced binding affinity.
    • The model was tested using sequence recovery (REC) and binding affinity improvement (pKD improvement). Benchmarking involved comparing NanoGen against baseline models such as ESM-2 650M, AbLangHeavy, and nanoBERT under both random masking and CDR-specific masking strategies on the CD45 patent dataset.

  • 2025-03-21

    How well do contextual protein encodings learn structure, function, and evolutionary context?

    • generative methods
    • protein design
    • Novel inverse folding algorithm, studying the effect of pretraining on the effectiveness of Antibody design
    • Authors check multiple inverse folding regimens, pretraining on general proteins, ppi interfaces and antibody-antigen interfaces and likewise finetuning on these.
    • They only use the backbone atoms (N,C,Ca), with special provisions for Cb.
    • They mask portion of the sequence and have the model guess its amino acids.
    • The 37% recovery at 100% masking appears slightly lower than the same feat for proteinMPNN.
    • Pretraining on antibodies still holds a signal towards antibody-antigen complexes, showing the power of such pre-training.

  • 2025-02-17

    Structure-informed language models are protein designers

    • generative methods
    • protein design
    • One of the first studies showing that introducing structure to protein language models, improves the predictive ability.
    • They fed ProteinMPNN (structural) inputs to ESM-1B to show that it improved recovery as opposed to using ESM-1B mask alone.
    • To marry ProteinMPNN and ESM-1B they use an ‘adapter’. Adapters in machine learning are lightweight modules that modify or extend a model’s functionality without retraining all parameters; in LM-DESIGN, a structural adapter integrates structural information into protein sequence predictions by bridging the structure encoder and a pretrained language model (pLM).
    • LM-DESIGN benchmarked against state-of-the-art protein inverse folding models, including ProteinMPNN, PiFold, GVP-Transformer, Structured Transformer, and GVP, while utilizing pretrained language models such as ESM-1b 650M and the ESM-2 series.
    • LM-DESIGN was evaluated on CATH 4.2 and CATH 4.3 datasets using sequence recovery rates and perplexity, compared against baselines.
    • LM-DESIGN outperformed individual models, improving sequence recovery by 4-12% points, surpassing ProteinMPNN and PiFold.

  • 2025-02-17

    IgGM: A Generative Model for Functional Antibody and Nanobody Design

    • protein design
    • generative methods
    • Novel method to design antibodies de novo.
    • Architecturally, it is a mix of language models, diffusion and structure prediction methods.
    • Training happens by noising diffusion, firstly perturbing structure and making the model get it right and afterwards doing the same thing for sequences.
    • After these two steps the model is distilled into a consistency model. This results in a model that can get the final coordinates/sequence in a single step rather than iterative denoising.
    • Method achieves comparable accuracy to many methods out there, such as DiffAb, dyMEAN and others.
    • On docking, the best performance is in the order of 4A iRMSD when using an AlphaFold3 antibody model - so still some challenges remain.
    • No wetlab validation.

  • 2025-02-03

    Benchmarking Inverse Folding Models for Antibody CDR Sequence Design

    • generative methods
    • protein design
    • nanobodies
    • Benchmarking of sequence design methods that are structure-conditioned
    • ESM-IF, LM-Design, ProteinMPNN and AntiFold were benchmarked.
    • On sequence recovery, AntiFold beats others on antibodies, but LM-Design is better when VHHs are considered.
    • AntiFold makes minimal use of the antigen information.
    • ESM-IF and ProteinMPNN have some weak correlation with affinity data.

  • 2024-10-24

    Pretrainable geometric graph neural network for antibody affinity maturation

    • binding prediction
    • protein design
    • GearBind - Novel framework to predict the effect of mutations on an antibody-antigen complex
    • The architecture is graph-based, trained in a contrastive fashion on real atoms and their surroundings versus randomly samples (from rotamer libraries) atoms within the same environment. They use the real proteins from CATH for this purpose. The random points are serving as ‘negatives’ for contrastive learning whereas the real ones as positives.
    • The method shows improvement on previous datasets: SKEMPI and the Absci HER2 dataset.
    • The authors demonstrated the effectiveness of the method by performing in silico affinity maturation on two existing binders.

  • 2024-08-28

    Unsupervised evolution of protein and antibody complexes with a structure-informed language model

    • generative methods
    • protein design
    • Authors demonstrate that using inverse folding, one can affinity mature antibodies, confirmed experimentally.
    • Authors employ ESM-IF as the inverse folding algorithm.
    • They take two existing antibodies, bebletovimab and BD55-5840, both instrumental in COVID-19.
    • They introduce all possible single point mutations to the Vh and Vl regions (about 4300). They pick the best perplexity for experimental characterization.
    • The best perplexity ones have many framework mutations (bebletovimab 10/14 and BD55 5840 3/6). There was only one mutation to CDR-H3 in Bebletovimab.
    • Inverse folding mother achieves much better performance when antigen is used as well.