Computational Antibody Papers
Filter by tags
developability
Filter by published year
All
TitleKey points
2026-10-20
kitAb: an automated framework for assay-specific antibody developability modeling and large-scale landscape analysis
- developability
- Developability prediction framework integrating 59 interpretable sequence- and structure-based descriptors with automated machine learning to deliver fast, assay-specific developability predictions and repertoire-scale biophysical profiling.
- The AutoML pipeline was benchmarked on 10 public experimental datasets spanning 46 endpoints for thermostability, aggregation, polyreactivity, and viscosity (n=21–241), encompassing diverse clinical-stage panels (Ginkgo, Jain 2017/2023/2024, Kraft, AB21) and single-parent variant libraries (Hutchinson, Jetha 2019, PDGF38). Evaluated via sequence-aware nested cross-validation (80% identity clustering), kitAb matched PROPERMAB (mean Spearman =0.36) while exceeding sequence baselines (=0.27) and TAP (=0.18), performing best on direct physicochemical readouts such as cIEF (=0.97) and HIC (=0.59–0.71).
- High-throughput profiling was benchmarked on 3,438,179 paired natural human antibodies (curated from OAS and PairedAbNGS across naïve, memory, and plasma B-cell stages) alongside 706 Phase II+ clinical therapeutics from Thera-SAbDab. This comparison established that therapeutic antibodies occupy a distinct biophysical subspace characterized by elevated global positive charge, restricted CDR positive patches, and high aromatic exposure without contiguous hydrophobic clustering, with specific germline combinations (notably IGHV3 - IGKV1) aligning closest to clinical profiles.
- On a single-CPU timing benchmark using the PDGF38 AB-001 structure across 50 repeats, kitAb completed descriptor generation in 4.21 seconds versus 45.98 seconds for PROPERMAB, so ~11-fold acceleration. Predictions proved robust across three structural backends (ABodyBuilder2, ABodyBuilder3, and FlashABB), maintaining low within-method minimization variability (median 0.07–0.08).
2026-10-20
Monoclonal Antibody Coformulations Reveal Three Interaction Patterns Governing Stability
- developability
- Systematic evaluation of 45 binary monoclonal antibody coformulations reveals that high-concentration viscosity is governed by three distinct cross-interaction patterns: minimal, attractive, and repulsive.
- Coformulation can unexpectedly reduce viscosity below that of individual parental antibodies while conformational stability remains preserved.
- Across 1:1 combinations of ten IgG1 mAbs (eight low-viscosity and two high-viscosity) at 150 mg/mL, mixtures partitioned into minimal (viscosity intermediate between parental mAbs; 17 of 28 low-viscosity pairs), attractive (viscosity exceeding both parents; 3 pairs), and repulsive regimes (viscosity lower than both parents; 8 pairs). Coformulation also demonstrated viscosity-rescue potential: pairing high-viscosity mAb H1 with low-viscosity counterparts consistently reduced viscosity below the 20 cP manufacturing threshold, whereas mAb H2 primarily triggered attractive, viscosity-increasing interactions.
- Dilute-solution diffusion interaction parameters (kd at 2–10 mg/mL) displayed only partial correspondence with high-concentration viscosity, highlighting that dilute colloidal measurements cannot reliably predict concentrated mixture rheology. Concurrently, conformational stability evaluated by differential scanning fluorimetry exhibited simple convergence behavior, with unfolding onset (Tonset) and first melting (Tm1) temperatures falling within parental bounds and confirming that coformulation does not disrupt antibody conformational stability.
- Because all parental mAbs had isoelectric points above the formulation pH (pH 6.0) and carried net positive charges, cross-interaction behaviors were governed by localized, anisotropic surface patches rather than global charge effects. Exploratory DeepSP descriptor modeling revealed that localized charge and hydrophobic patches within complementarity-determining regions, notably positive charge in CDRL1, negative charge in CDRH3, and CDRH3 hydrophobicity, strongly correlated with viscosity behavior, providing key feature targets for early-stage computational screening.
2026-09-29
An interpretable open platform for sequence-based antibody developability prediction
- developability
- Set of open source models (DELPHI) for self-association and polyreactivity trained on a sizeable set of proprietary sequences.
- DELPHI benchmarks 25 protein language model (PLM) and classifier combinations to enable sequence-based prediction of antibody developability liabilities with residue-level interpretability.
- Evaluated using CDR H3-cluster-stratified cross-validation to prevent sequence-similarity leakage, DELPHI achieves high within-distribution performance for polyreactivity (mean AUC 0.959) and size-exclusion chromatography (SEC) monomer purity (mean AUC 0.933), while demonstrating strong zero-shot transfer to clinical cohorts and a 246,293-antibody public library (AUC up to 0.950).
- Benchmarks show that language model choice is a stronger performance determinant than downstream classifier architecture with paired-chain PLMs (AbLang2, IgBert) leading cross-library generalization and automated learning curves demonstrate that predictive performance plateaus around 5,000 CDR H3-diverse training sequences.
- Integrating SHAP, Integrated Gradients, and in silico CDR3 loop mutagenesis, DELPHI maps attributions to individual positions and uncovers a shared CDR H3 electrostatic risk signature (arginine/lysine enrichment, aspartate depletion, and net positive charge) that simultaneously drives polyreactivity and SEC non-monomeric failure
2026-09-21
Predicting non-specific binding of VHHs using machine learning models with cluster-aware validation
- developability
- nanobodies
- Prediction of non-specific binding using ML from a diverse dataset.
- Evaluated on a dataset of 1,563 VHH molecules measured via a high-throughput baculovirus particle (BVP) assay, the workflow uses a Leave-One-Group-Out cross-validation scheme to assess predictive performance in both high-similarity Lead Optimization (LO) and diverse Lead Isolation (LI) screening contexts.
- A soft-voting ensemble classifier combining Logistic Regression, Random Forest, and LightGBM using AlphaFold2-derived structural descriptors achieved an AUROC of 0.73 in cross-validation and 0.79 on an independent test set, identifying surface-exposed hydrophobicity and positive charge patches in the complementarity-determining regions (CDRs) as primary drivers of polyreactivity.
- Predictive performance sharply degrades for VHH sequences beyond a Levenshtein edit distance of 20 from the training dataset, establishing clear domain-of-applicability boundaries and demonstrating the necessity of sequence diversity for out-of-domain generalization.
2026-09-11
Review: application and opportunities for machine learning and artificial intelligence in preclinical immunogenicity risk assessment
- developability
- Review recounting insights from an EMBL-EBI workshop on how AI/ML are applied alongside in vitro assays and Quantitative Systems Pharmacology (QSP) to evaluate and mitigate preclinical immunogenicity risks for biotherapeutics.
- Advanced in silico tools (such as NetMHCIIpan 4.3, Graph-pMHC, and HLAIIPred) accurately predict HLA Class II peptide presentation using mass-spectrometry immunopeptidomics data, though predicting T-cell receptor (TCR) binding for unseen epitopes and B-cell epitopes remains a bottleneck.
- Pharmaceutical workflows integrate computational screening early in candidate selection to guide deimmunization, while QSP "middle-out" modeling combines computational predictions with empirical assay data to simulate clinical anti-drug antibody (ADA) formation and pharmacokinetic impacts.
- The risk assessment framework is expanding beyond standard monoclonal antibodies to address unique immunological challenges in complex modalities, including AAV viral vectors, CAR-T cell therapies, and CRISPR-Cas9 gene editing components.
- Future improvements in clinical ADA prediction rely on overcoming data fragmentation, assay non-standardization, and small dataset sizes, with collaborative initiatives like federated learning offering a path forward.
- Novel dataset of 160 VHH-Fc profiled across 10 biophysical assays.
- They demonstrated that tabular neural networks (TabICLv2, TabPFN v2.5) trained solely on 559 IgG heavy chains outperform intra-format VHH-Fc models in zero-shot predictions, showing that training data scale outweighs structural divergence between scaffolds.
- Surface-driven properties transfer with high accuracy, heparin binding (rho=.82), hydrophobicity (HIC, rho=.63), and self-association (AC-SINS, rho=0.62), whereas thermostability (Tm2, rho=0.16) remains scaffold-dependent and requires direct measurement.
- Augmenting models with simple surface property inputs (HIC and HAC) boosts prediction accuracy for complex liabilities like polyreactivity (PR-CHO, rho = 0.40 to 0.51).
2026-08-28
Aligning protein-generative models to experimental fitness with ProteinDPO
- developability
- language models
- Novel way (ProteinDPO) to apply pre-trained models to biophysical readouts.
- ProteinDPO is the first framework to apply Direct Preference Optimization (introducing a novel scalar-weighted DPO objective) to align protein generative models with experimental biophysical data, without overfitting like traditional supervised fine-tuning.
- Despite training exclusively on small monomeric stabilities, ProteinDPO generalizes zero-shot to accurately rank the thermal melting temperatures of multichain antibodies and score antibody-antigen binding affinities.
- Applied to H5N1 influenza hemagglutinin, it generated stabilized variants yielding up to a 32C improvement in thermal stability while retaining strong nanomolar binding affinity to broadly neutralizing anti-HA antibodies.
2026-08-21
A blinded, prospective benchmark of in silico antibody discovery anchored to experimental affinity and developability
- binding prediction
- developability
- ngs
- Results from the antibody discovery competition, where ML methods faced off with experimental baselines.
- Participating organizations were evaluated across three tasks on a SARS-CoV-2 receptor-binding domain (RBD) dataset: (1) in silico affinity maturation by modifying non-HCDR3 CDRs using sequencing outputs, (2) affinity ranking of candidates within three HCDR3-clustered sequence datasets, and (3) out-of-library CDR optimization/design to generate novel high-affinity binders not present in the sequencing data.
- For lead optimization, top ML methods matched experimental performance by producing sub-100 pM binders (top ML design: 95 pM vs. best experimental control: 113 pM), though overall ML submissions underperformed relative to experimental sorting libraries. Notably, a simple non-ML statistical consensus baseline also achieved a highly competitive 540 pM affinity without using machine learning.
- For sequence ranking within HCDR3 clusters, almost all ML algorithms performed worse than random clone picking; only 9.8%-13.8% of ML submissions beat the cluster baseline clone compared to 39% from random selection. Additionally, out-of-library ML designs yielded a high proportion of non-binders (30.4%) and top-ranked affinity hits frequently suffered from severe biophysical developability failures.
- ML is genuinely useful for timeline reduction during lead optimization and affinity maturation, replacing 2-3 weeks of physical combination library construction with an in silico step when provided with deep, biologically grounded experimental datasets.
- Pretrained protein language models (PLMs) paired with structure-aware modules (e.g., pairformers) showed utility in exploring novel out-of-library sequence spaces. However, current ML models lack cross-task generalization, struggle with local epistasis, and fail to reliably co-optimize affinity alongside multiparametric biophysical developability.
2026-07-13
Decoding Bispecific Antibody Developability: Design Rules and Predictive Models from a 160-Member Library
- developability
- Developability characterization of multispecific antibodies, for the purposes of training machine learning models
- To map out developer behavior, the researchers used advanced diversity-sampling algorithms to select 160 unique bispecific combinations from 65 parental antibodies, capturing a vast spectrum of biophysical profiles.
- Developability traits don’t transfer uniformly from parent to child; while hydrophobicity and charge are highly predictable, critical flaws like self-association often emerge unexpectedly as "outliers" when the two arms interact.
- The study revealed that thermostability is strictly format-driven, meaning it cannot be reliably predicted from parental data and must be measured directly on the final multi-specific construct.
- Very fast antibody modeling, FlashABB, with applications to developability screening.
- FlashABB accurately performs sota antibody antibody structure prediction in just 5 ms, making it faster than protein language models can even generate sequence embeddings, by completely discarding traditional pair representations in favor of an inner-product-based, linear-memory "Flashpoint Attention" (FPA) algorithm.
- Featuring roughly 7.2 million parameters trained on 8,395 SAbDab structures, the model is strictly forced to run on float32 precision to avoid the catastrophic numerical cancellation errors that heavily degrade 16-bit float implementations during distance calculations.
- Eliminating the structural compute bottleneck allows developers to scale downstream analysis to millions of sequences; using the variant tool FlashTAP, teams can filter out liability flags and predict therapeutic developability metrics for 100 antibodies per second.
- Having said that, CDR-H3 is still sota - so in the region of 2.5A, but it gets there way faster.