Computational Antibody Papers
Filter by tags
language models
Filter by published year
All
TitleKey points
2025-09-12
ImmunoMatch learns and predicts cognate pairing of heavy and light immunoglobulin chains
- language models
- Novel model to predict the heavy/light chain compatibility
- Data: H/L with the same single-cell barcode; negatives = swap L chains between pairs but only if CDRL3 length matches; balanced set of 233,880 pairs with a 90/10 train–test split.
- Training: Full VH+VL into AntiBERTa2 with a classification head; fine-tuned 3 epochs, lr 2×10⁻⁵, weight decay 0.01; κ/λ-specific variants trained identically. Final AUC-ROC 0.75 (withheld) and 0.66 (external); κ/λ models: 0.885/0.831.
- Baselines: (i) V/J gene-usage → logistic reg. & XGBoost ≈ 0.50–0.52 acc.; (ii) CDRH3+CDRL3 CNNs → moderate; (iii) ESM-2 improves with fine-tuning but AntiBERTa2 FT is best.
- It seems to do better than just ‘matching to the database’. Weak gene-usage baselines, explicit control of CDRL3 length in negatives, external generalisation, and sensitivity to interface residues (CDRH1/2 & framework) in therapeutic-antibody tests argue the model learns sequence-level pairing rules, not just V/L distributions.
- Demonstration showing how large language models (LLMs) can be adapted to reduce the computational cost of molecular dynamics (MD).
- They use the FoldToken encoding to discretize protein 3D conformations into tokens compatible with Mistral, and fine-tune the LLM on short MD trajectories of a single state. The model is then able to generate new sequences of conformations by predicting the next frame from previous frames.
- After fine-tuning, the model can extend trajectories beyond the training data. Starting from a native state, it can discover alternative conformations potential for bypassing kinetic barriers that normally require long MD runs.
- The approach is system-specific (requires an MD trajectory for each protein), does not yet encode thermodynamics/kinetics explicitly, and relies on the choice of structural tokenization.
2025-07-25
NextGenPLM: A Novel Structure-Infused Foundational Protein Language Model for Antibody Discovery and Optimization
- language models
- epitope prediction
- Novel protein language model with applications to epitope prediction and ranking hits in campaigns.
- NextGenPLM introduces a modular, multimodal transformer that fuses frozen pretrained protein language models with structural information via spectral contact-map embeddings, enabling efficient modeling of multi-chain antibody–antigen complexes without requiring full 3D folding of antibodies.
- The model was benchmarked on 112 diverse antibody–antigen complexes against state-of-the-art structure predictors (Chai-1 and Boltz-1x), matching their contact-map and epitope prediction accuracy while achieving ~100× higher throughput (4 complexes/sec vs. ~1 min/complex).
- The model was experimentally validated through an internal affinity-maturation campaign. Using its predictions to rank antibody variants led to designs that achieved up to 17× binding affinity improvements over the wild-type, as confirmed by surface plasmon resonance (SPR) assays.
- A novel antibody-specific language model, trained on paired human antibody data, and explicitly designed for practical antibody engineering applications.
- The model was trained on a carefully curated dataset of productive, paired sequences, prioritizing biological fidelity over sheer volume or data heterogeneity.
- It uses a masked language modelling (MLM) objective. The initial version was based on RoBERTa, while later versions introduced custom architectural modifications tailored to antibody sequences.
- The model was benchmarked on recapitulating clinical humanization decisions and outperformed prior models such as Sapiens and AntiBERTa.
- It was applied to redesign an existing therapeutic antibody, generating variants with retained or improved affinity, reduced predicted liabilities, and confirmed in vitro performance, including CHO expression and binding assays.
2025-06-24
Benchmark for Antibody Binding Affinity Maturation and Design
- binding prediction
- language models
- generative methods
- databases
- Benchmark of machine learning models for antibody-antigen binding affinity.
- A curated dataset of over 150,000 antibody-antigen complexes with associated experimental affinity values is compiled from literature.
- The benchmark compares a wide range of model types: language models, inverse folding models, graph-based, and diffusion-based generative models.
- Inverse folding models that are globally structure-aware perform best.
- General protein models like ESM-IF and ProteinMPNN outperform antibody-specific models such as AntiFold, DiffAb, and dyMEAN.
- Surprisingly, ESM-3 underperforms relative to ESM-IF, despite incorporating structural signals and improving upon earlier ESM models.
- Novel LLM (MINT) that natively encapsulates protein protein interactions.
- MINT (Multimeric INteraction Transformer) extends the ESM-2 protein language model by incorporating a cross-chain attention mechanism. This allows it to process multiple protein sequences simultaneously while preserving inter-sequence relationships and contextual information critical for modeling protein-protein interactions.
- MINT was trained on a large, curated subset of the STRING database, consisting of 96 million high-quality physical protein-protein interactions and 16.4 million unique protein sequences. The training employed a masked language modeling objective adapted for multimeric inputs.
- MINT was benchmarked on several general protein interaction tasks including binary interaction classification, binding affinity prediction (PDB-Bind), and mutational impact prediction (e.g., SKEMPI and MutationalPPI). It consistently outperformed existing PLMs, achieving state-of-the-art performance on multiple datasets such as a 29% improvement over baselines in SKEMPI.
- MINT outperformed antibody-specific models (e.g., IgBert, IgT5, and AbMap) on the FLAB benchmark and SARS-CoV-2 antibody mutant binding prediction tasks. It showed >10% performance improvement on three FLAB datasets and a 14% gain in low-data settings (0.5% training data) for SARS-CoV-2 binding predictions.
- AbMAP - Language model transfer learning framework with applications to antibody engineering.
- Authors address the process of dichotomy of language models in antibodies - either one uses a bare-bones protein model like ESM or only antibody model like Antiberty/IgLM. Normal protein models will not capture hypervariability of CDRs whereas antibody models would focus too much on the framework. They focus solely on CDRs + flanking regions as a solution.
- They show their applicability to three off the shelf models with structure template finding as well as low-n generative modeling.
2025-03-31
IgCraft: A versatile sequence generation framework for antibody discovery and engineering
- generative methods
- language models
- Novel generative model for antibodies that allows one to fill in, inpaint inverse fold etc.
- The model employs Bayesian Flow Networks which is somewhat similar to diffusion.
- The model is trained on data from OAS - unpaired data as a first pass and paired data as a second pass.
- Models are benchmarked on a range of computational metrics, chiefly sequence recovery (for infiling/inverse folding).
- Developability is checked by computational prediction of solubility (CamSol) and humanness (AbNativ)
2025-02-20
Few-shot in-context learning with large language models for antibody characterization
- language models
- Demonstration that general purpose language models - like GPT3.5 - can reason about antibody - engineering tasks.
- Authors explore the topic of in context learning - e.g. few shot learning where several examples are given and on the basis of that the model needs to provide a prediction for a new case.
- They tested an array of general purpose models, such as GPTs, LLamas, Mistrals etc.
- They tested on three antibody tasks - mouse/human discrimination, specificity prediction (from ngs) and isotype identification. In theory not that difficult tasks, but remember we are dealing witha general purpose language model.
- They literally prompt the model with examples on, say mouse antibodies, human antibodies and provide a next one to predict.
- They find that the predictions are not bad, especially in few shot scenario (16 examples or so).
- In one test it even achieved accuracy on par with AntiBERTy.
2025-02-17
DyAb: sequence-based antibody design and property prediction in a low-data regime
- binding prediction
- language models
- Method to employ low-N data for biologic engineering.
- Assuming we have a dataset of ~100 affinity data points, we can choose (100 choose 2) pairs where we know which one has a larger readout than the other (e.g. stronger affinity) giving combinatorially larger amount of data points to train on.
- The architecture used is CNN on top of a language model.
- Benchmarked on three internal campaigns, Il6, EGFR and an undisclosed target.