ImmunoInformatics

T cell receptor (TCR) specificity prediction is critical for understanding adaptive immunity and for enabling the development of therapeutic interventions. Current machine learning models for predicting TCR-peptide-MHC interactions perform poorly, despite large efforts by many groups and advances in protein language models. This work identifies and addresses the fundamental issues that might be hindering progress in this field, focusing on current approaches for constructing training and validation sets, particularly on the strategies used by commonly used methods, which generate negative data either by shuffling TCRs between epitopes or by pairing epitopes with TCRs from background datasets of healthy donors. Here, we visualize and quantify the overlap between positive and generated negative TCR sets produced by both strategies using a range of sequence embedding approaches, including simple encodings (one-hot and BLOSUM62), general protein language models (ESM-2 and ProtGPT2), and a TCR-specific embedding model (TCR-BERT). Using paired {beta}TCR data from VDJdb for two well-studied epitopes (GILGFVFTL and KLGGALQAK), we train Random Forest classifiers on each embedding representation and evaluate their ability to distinguish positive from negative data examples. We find that general sequence embedding methods struggle to differentiate between positive and negative TCRs. We further show that the choice of negative data generation method significantly impacts classifier performance. Our findings highlight fundamental limitations in current TCR specificity training data and general protein language models, and underscore the need for TCR-specific embeddings, experimentally validated negative datasets, transparent reporting of negative data construction, and epitope-specific modeling approaches.

Accurate prediction of T cell receptor (TCR) specificity can potentially accelerate development of novel immunotherapies, yet public databases used to train these models contain substantial label noise. This uncertainty complicates model evaluation and limits the reliability of predicted TCR-pMHC interactions. In this work, we present a simple supervised machine learning denoising scheme to identify low-quality labels in TCR specificity data. We evaluated the framework using the TCRvdb validation dataset and further corroborated it via structural modeling with AlphaFold 3. Here, our denoising approach successfully identified non-validated TCRs, achieving an average MCC of 0.47 for filtering non-validated examples while retaining high sensitivity for validated examples. Applying our denoising approach to the complete NetTCR training dataset, we observed that data associated with known noisy attributes were significantly enriched in the filtered subset. Furthermore, TCR-pMHC complexes identified as noise consistently produced lower AlphaFold 3 confidence scores, providing independent evidence of misannotation. Experimental validation of 30 selected TCRs further supported the approach, with 80% of TCRs predicted as noise failing to demonstrate functional reactivity. Our findings demonstrate that current performance metrics may be obscured by dataset inaccuracies and provide a method for refining training data in the absence of gold-standard benchmarks.

Antibody developability is determined by multiple physicochemical properties, and the relevance of individual properties varies among experimental assays. Existing computational approaches, however, often provide fixed assessments or rely on computationally intensive descriptors. We present kitAb, an open-source framework for scalable physicochemical characterization of antibody Fv regions and assay-specific developability modeling. kitAb calculates 59 interpretable sequence- and structure-based descriptors and integrates them with automated feature selection and regression. We evaluated kitAb across ten public datasets comprising 46 experimental endpoints, using sequence-aware or predefined splits where feasible. Predictive performance was strongly assay-dependent. In nested evaluations with sequence-aware or predefined data splits, models using kitAb descriptors achieved predictive performance comparable on average to those using PROPERMAB descriptors, while outperforming models based on TAP or sequence-only features. In a single-structure CPU timing benchmark, kitAb descriptor calculation was approximately 11-fold faster than PROPERMAB. Predictive performance was similar across the three structure-generation backends, and broader AutoML searches did not substantially improve mean performance. We further applied the same descriptor representation to approximately 3.4 million paired natural antibodies and 706 Phase II or later therapeutic antibodies. Repertoire-scale analysis revealed systematic differences in surface charge, hydrophobic and aromatic organization, CDR length and germline-associated physicochemical profiles. The kitAb code and analysis workflows are openly available at https://github.com/katyachemistry/kitAb/.

Antidiabetic peptides (ADPs) are short bioactive sequences of therapeutic interest, and sequence-based classifiers prioritise experimental candidates. A classifier is useful only if its accuracy transfers to unseen sequences, which depends on how the benchmark was as sembled and partitioned. We term the distance between what such a classifier is scored on and the function it is meant to predict the provenance-to-function gap, and we quantify it. We re-evaluate the two-layer ADP benchmark of Basith et al. under a protocol that groups homologous peptides rather than splitting them at random. Of the 877 ADPs, 218 attribute to a precursor protein by exact substring containment; within that subset the second-layer label coincides exactly with precursor identity, all 140 human-insulin fragments carrying the type-1 label and all 76 bovine milk-protein fragments the type-2 label, without exception. Accuracy tracks identity to the training set, rising from a Matthews correlation coefficient (MCC) of 0.39-0.43 below 50% identity to 0.92-0.96 between 70% and 90%, and peptide length alone reaches MCC 0.619 on held-out data. Auditing sixteen further peptide bench marks shows the coupling is not confined to this resource: fragment families share a label more often than chance in eleven of fourteen testable datasets, and not in three, so the prop erty is common rather than universal. Rebuilding the published architecture on identical folds shows its advantage over a single classifier is a function of the split: present under random partitioning, absent once homologues are separated. ADP-Hybrid, one tree-ensemble classifier per layer, reaches MCC 0.843 on layer 1 against a published 0.841 at 21-38 times the inference throughput, and 0.801 against 0.858 on layer 2. We release the protocol and the controls that expose these properties.

Structure-guided antibody CDR design constrains residue selection using backbone geometry, but structural evidence alone may leave multiple amino acids plausible. We investigate whether the fixed antibody framework provides complementary sequence information for resolving this ambiguity. We introduce AbRefine, an information-aware fusion framework that combines backbone-conditioned structural predictions with framework-conditioned sequence distributions derived from a pretrained antibody language model with all designed CDRs masked. Its Information-Aware Fusion (IAF) module learns residue-specific fusion weights from the uncertainty, disagreement, and information content of the two predictive distributions, while keeping both pretrained experts frozen. Controlled interventions show that the signal is target specific: matched frameworks outperform shuffled frameworks by 7.37 percentage points, and the largest gain occurs at positions with high structural uncertainty and informative framework evidence (+5.77$ points). Compared with static fusion, IAF further improves AbMPNN recovery by +1.53 points (95\% CI +0.93$ to +2.14$) and significantly reduces native negative log-likelihood for both AbMPNN and AntiFold, without detectable degradation in fold-back structural compatibility. These results show that framework context provides information beyond backbone geometry and that learning its residue-specific utility enables effective adaptive fusion for antibody CDR design.

Accurate modelling of biomolecular interactions is fundamental to drug discovery, yet current artificial intelligence (AI) workflows remain fragmented across structure prediction, affinity estimation, molecular design, and experimental decision-making. We introduce AnewDDE, an agentic Drug Discovery Engine that connects these capabilities into a closed-loop system for biomolecular interaction modelling and design. We demonstrate several components of this system: AnewFold delivers superior performance on challenging targets, including antibody-antigen complexes and molecular glues, providing more reliable structural hypotheses together with pocket identification and conformational analysis; AnewDesign achieves a 10.7% success rate in identifying binders with single-digit-nanomolar affinities measured by surface plasmon resonance (SPR) in a representative nanobody discovery campaign, substantially reducing the experimental search burden; AnewAffinity combines accuracy approaching that of free-energy perturbation (FEP) methods with the speed required for high-throughput screening and prioritizes promising transformations; and AnewMind, a large language model (LLM) at the hundred-billion-parameter scale that underwent in-house full-parameter post-training, provides scientific reasoning and absorption, distribution, metabolism, excretion, and toxicity (ADMET) prediction for pharmaceutical research and development (R&D). To systematically evaluate AnewMind, we introduce PharmBench, an internal benchmark designed by pharmaceutical R&D experts to evaluate long-horizon decision-making in drug discovery. Evaluations on ADMET tasks and PharmBench show that AnewMind performs competitively with leading frontier models. Together, these capabilities establish AnewDDE as an integrated engine for addressing challenging therapeutic targets and translating molecular insights into experimentally validated candidates at scale.

Designed binders are routinely filtered on predicted interface confidence, which reports how self-consistently a model places an interface, not where it lies. We designed 200 nanobody loops against the survivin BIR-domain groove, co-folded 196, and checked epitope contact in every predicted complex. Confidence showed no detectable association with epitope engagement ({rho} = +0.100, 95% CI -0.04 to +0.24, n = 196); its highest and lowest quartiles failed alike (odds ratio 0.97, 95% CI 0.41 to 2.26). It tracked instead the run's other confidence estimates ({rho} = +0.61). Twelve-seed replication separated two candidates the criteria had not: one reached the complete epitope in 10 of 12 runs, the other never. Across 155 predictions the top-ranked model was no better than one drawn at random from its five (P = 0.610), though only 3 of 31 failed on every model. Nothing was assayed; nb_cand_057 is an in silico candidate only.

This work presents DSSNA (Define Spatial Structure of Nucleic Acids) v2026, a free and open-source standalone GROMACS module for the automated analysis of nucleic acid structure based on atomic coordinates. DSSNA reproduces the core functionality of the X3DNA and DSSR approach while extending it to the analysis of molecular dynamics trajectories through integration with the GROMACS software package. The algorithm identifies and classifies structural features, including canonical and noncanonical base pairs, hydrogen-bond networks, stacking interactions, helices, and many other elements of nucleic acid secondary and tertiary structure. Its modular output system enables users to select the structural metrics required for a particular analysis, while the use of neighbor-search algorithms and data-processing capabilities facilitates the analysis of large trajectory datasets. The performance and applicability of DSSNA were demonstrated using a molecular dynamics trajectory of a GUAA tetraloop mutant of the sarcin-ricin domain from Escherichia coli 23S ribosomal RNA (PDB ID: 1MSY). The analysis showed that only a subset of the observed base pairs and stacking interactions remains stable throughout the trajectory. Persistent canonical base pairs formed the structural core responsible for maintaining the secondary-structure profile, whereas most noncanonical base pairs and stacking interactions were transient. The results demonstrate that DSSNA can be used to characterize both static nucleic acid structures and the temporal evolution of structural interactions in molecular dynamics simulations. The work also provides a glossary of terms relevant to the analysis and interpretation of nucleic acid structures.

Advances in machine learning (ML) have enabled the de novo design of synthetic proteins and peptides tailored for specific structural or functional properties. As ML becomes a central tool in forward design, we suggest that its value extends beyond optimizing known objectives. It can also lead to what we term here in this work as silicon exaptation, revealing unexpected functional properties that were never part of the original design goals. This reflects a principle common in biology, that repurposes traits where features evolved for one role can later enable unforeseen functions. In this context, these accidental findings are not constrained by preconceptions but arise from latent function emergence and the algorithm's gradual drift, a new process that can co-evolves with human curiosity, intuition, and interpretive reasoning, which can be used to recognize unexpected patterns and transform them into new discoveries. In this study, we discovered a set of peptides through silicon exaptation, in which the ML pipeline was originally optimized for structural characteristics unrelated to any antimicrobial functionalities. The peptides were originally designed to adopt defined secondary structures in response to environmental stimuli, with no antimicrobial properties intended or included in the training objectives. We systematically evaluated the antimicrobial potential of these peptides using broth microdilution assays and membrane integrity tests, identifying several candidates with potent and selective antibacterial activity without detectable cytotoxicity toward mammalian cells.

Peptides containing noncanonical amino acids (ncAAs) are becoming increasingly important in drug discovery, but their molecular dynamics (MD) simulation remains limited by labor-intensive residue parameterization and topology preparation. Here, we present TopFor (Topology and Residue Force-Field Parameter Preparation), an open-source, peptide-oriented workflow that automates the generation of ncAA residue templates and force-field parameter files for MD simulations with AMBER using established charge-derivation methods. TopFor supports single-residue, whole-peptide, and batch parameterization workflows. The workflow was validated using the MDM2-p53 complex by comparing native AMBER residue definitions with residues processed by TopFor using three charge-derivation methods, each evaluated in three independent 100 ns MD simulations. TopFor-generated residue definitions showed structural and conformational behavior comparable to the native AMBER representation. Applicability to more complex peptide architectures was demonstrated using the head-to-tail cyclic peptide cyclosporin A and the cyclic depsipeptide cyclohexylgriselimycin. At library scale, 1,138 of 1,214 cyclohexylgriselimycin derivatives (93.7%) successfully completed TopFor parameterization and the subsequent initial AMBER-based computational workflow across two derivatization campaigns. TopFor was further integrated into a structure-based peptide discovery workflow based on the DnaN-griselimycin complex. Six GM-derived peptides were computationally prioritized and experimentally evaluated, of which four produced SPR responses against Mycobacterium smegmatis (M. smegmatis) DnaN and three showed concentration-dependent responses. TopFor reduces the manual effort required to prepare ncAA-containing peptides for MD simulations and enables their integration into both library-scale computational screening and structure-based peptide discovery workflows. Keywords: noncanonical amino acids; peptide parameterization; molecular dynamics; AMBER; AmberTools; peptide design; high-throughput virtual screening

Relating the drivers of binding affinity to structural features remains difficult because affinity emerges from many weak, context-dependent interactions, while experimental affinity and mutation data are limited. Here, we ask whether a machine-learning model can learn energetically relevant interaction preferences directly from native structures, without affinity or mutation labels. We decompose interfaces into local structural motifs and learn the compatibility of each motif with its surrounding molecular environment. Applied across a protein surface, these models yield target preference maps (TPMs), spatial fields of local interaction compatibility. Trained exclusively on native antibody-antigen structures without affinity or mutation labels, TPMs recover charged, aromatic and backbone-mediated recognition across peptide-protein and other non-antibody interfaces. On an out-of-domain SKEMPI benchmark, wild-type TPM scores discriminated mutation-sensitive positions (AUC 0.643, n=1323), while TPM-based substitution scores ranked alternative amino acids comparably to FoldX and Rosetta Flex {Delta}{Delta}G. These results show that native interface structures can supervise learning of transferable, energetically relevant molecular preferences without direct thermodynamic labels.

Directed evolution enables the discovery of protein mutants with improved fitness through iterative rounds of selection and has been widely applied to adeno-associated virus (AAV) capsid engineering. Machine learning (ML) can augment this process by prioritising mutants for experimental validation, but whether its benefits outweigh the added cost of ML-designed library construction remains unclear. Here, we address this question by considering improvements in manufacturing efficiency of AAV capsids in the context of exosomal encapsulation, which is a technique for improving AAV immunogenicity. We generated a directed evolution dataset comprising 53,974 mutants across three rounds of selection and an independent assessment dataset of 472 ML-designed mutants with detailed profiling. Directed evolution alone yielded a 3-fold improvement, while ML-assisted directed evolution yielded a 41-fold improvement in manufacturing efficiency, demonstrating that ML-assisted directed evolution outperforms directed evolution alone. We show that data quality - specifically the number of selection rounds and the statistical power of mutant counts - has a greater impact on model performance than model architecture. Finally, we provide practical guidance on experimental design and implementation of ML tuning strategies that can augment model performance in protein engineering applications.

Protein complex structure prediction depends on both the breadth of candidate coverage and the ability to select accurate models within a limited output budget. Most existing approaches emphasize only one stage. End-to-end models and molecular docking workflows generate candidate structures, whereas quality-assessment models primarily rerank a predefined candidate pool. A complete strategy must account for complementary sampling across generators, differences among scoring scales and the allocation of final candidate quotas. Here, EnsPlex is presented as a multi-source framework that couples structural sampling with candidate selection for protein-protein interaction complex prediction. EnsPlex combines conformation-expanded docking, AlphaFold-Multimer, AlphaFold3 and Boltz-1 to expand the sampled conformational space. Conformation-expanded docking comprises monomer conformational expansion followed by flexible HADDOCK docking. FACET is trained to predict candidate quality using DockQ and its component metrics as supervision. Across 102 antigen-antibody systems, EnsPlex achieved up to a 27.8% relative improvement in target success rate over AlphaFold3 with its built-in ranking under matched output budgets. FACET also improved within-source ranking in the internal candidate pools and in homology-filtered external data. These findings support the utility of EnsPlex for structure prediction in the evaluated antigen-antibody systems.

Accurate and fast prediction of protein binding sites remains essential for understanding molecular interactions and facilitating protein engineering. Here, we present SurfGraphPro, a geometric deep learning approach that bridges protein language models with surface-based structural representations for a rapid binding interface prediction. Our method operates on coarse-graphed triangulated protein surfaces, utilizing solvent-excluded surface meshes downsampled into amino acid residue centered patches. Unlike existing approaches that rely on extensive physicochemical feature engineering, we leverage evolutionary information directly from protein language model embeddings, eliminating the need for computationally expensive multiple sequence alignments and hand-crafted features. This integration achieves on average $\sim18-28 \times$ speedup for proteins of $\sim$100 to a few 1000s of amino acids over current state-of-the-art surface-based model while maintaining comparable accuracy on diverse binding interfaces, including challenging antibody-antigen complexes. To our knowledge, this represents the first approach to integrate protein language model embeddings with coarse geometric surface representations for binding site prediction, demonstrating that learned evolutionary features coupled with geometric transformers can replace traditional feature engineering without sacrificing performance.

Natural killer (NK) cells display striking variability in their cytotoxic responses, but whether this variation reflects random events, stable differences between cells, or effects of previous encounters remains unclear. We developed a Bayesian framework that uses single-cell interaction histories to disentangle these sources of variability. The framework models target encounters and killing decisions to quantify population heterogeneity, determine the sample sizes needed to distinguish competing mechanisms, and separate stable cell-to-cell differences from history-dependent behaviour. Synthetic data established when these mechanisms can be reliably distinguished in practice. We tested the framework using time-lapse imaging of NK cells exposed to rituximab or CC-96673, an antibody co-targeting CD20 and CD47. Despite similar overall killing, the framework identified distinct underlying responses: rituximab mainly increased mean killing rate, whereas CC-96673 reduced cell-to-cell variation. Event-count analysis supported continuous population heterogeneity, while ordered interaction histories revealed that both stable differences in killing propensity and previous encounters shape cytotoxic decisions. To make the framework directly usable, we introduce BARRACUDA, an open-source web platform and Python package implementing these analyses, including donor-aware extensions, for single-cell cytotoxicity datasets.

Simplified spiking neuron models such as the Adaptive Exponential Integrate-and-Fire (AdEx) model reproduce the firing behaviour of diverse neuron types at a fraction of the cost of biophysically detailed models, making them attractive for large-scale brain simulation. Fitting their parameters to recordings, however, still relies on derivative-free search such as grid search or evolutionary algorithms, because the discrete spike-and-reset mechanism renders these models non-differentiable. Surrogate gradients, used in deep spiking networks, replace the spike derivative with a smooth approximation on the backward pass. Whether they also enable efficient gradient-based parameter estimation for single simplified neurons is untested. We implement a differentiable AdEx model with a surrogate-gradient variant in the Jaxley framework. The implementation is on par with target-compiled C++ Brian2 code while remaining end-to-end differentiable, vmap-batchable and GPU/TPU-portable. It is 200 times faster than Brian2's default runtime mode. Across a recovery-radius benchmark of 2,700 paired runs, gradient-based fitting never matched a gradient-free Nelder-Mead baseline. At sub- threshold levels, gradient descent instead beats Nelder-Mead. The obstacle is therefore the spike mechanism and the loss geometry it induces, not gradient descent itself.

Genomic variant calling is a central computational task in disease research and diagnostics, where somatic mutations such as JAK2 V617F serve as diagnostic markers in myeloid neoplasms. VarDict, originally written in Perl, and its reimplementation VarDictJava are widely used variant callers in the bioinformatics community. The Perl implementation requires run times of up to several hours; VarDictJava reduces this to a few minutes but can consume large amounts of main memory, which limits the number of jobs that can be run in parallel on a given machine. We present vardictcpp, a functionally identical reimplementation of VarDictJava in C++17. vardictcpp is approximately four times faster in single-core execution and 9-11 times faster using eight cores, while requiring up to 17 times less main memory. Variant calls are identical to those of VarDictJava on the evaluated whole-exome samples. The reduced memory footprint allows more concurrent jobs on existing hardware and thereby increases sample throughput, which is particularly relevant for clinical sequencing laboratories with fixed compute resources. vardictcpp is implemented in C++17 and freely available at https://github.com/MHH-Bioinformatics-Hematology/vardictcpp

Membrane proteins play crucial roles in cellular signaling, transport, and pharmacological regulation, yet their structural and functional characterization remains challenging due to experimental limitations. Molecular dynamics (MD) simulations have become indispensable for probing membrane protein behavior at atomic resolution; however, executing these simulations at scale remains a significant bottleneck. Existing tools excel at system preparation for individual cases, but none offer a complete, high-throughput pipeline spanning embedding, simulation, and analysis across multiple membrane compositions, and rarely integrate modern structure prediction and molecular docking frameworks. Here, we present GalaxyMemMD, a fully automated, open-source Galaxy framework that provides a direct, scalable alternative to existing preparation-only tools. GalaxyMemMD integrates every stage membrane embedding, system preparation, MD execution, and downstream analysis into a single reproducible pipeline. Leveraging Galaxy's HPC-ready job scheduling, GalaxyMemMD natively supports high-throughput parallel simulations across multiple membrane compositions and protein variants, including AlphaFold-derived structural inputs and downstream molecular docking workflows, without repeated manual setup. We validate GalaxyMemMD across six structurally diverse membrane proteins four GPCRs, one ion channel, and one beta-barrel demonstrating consistent system preparation, while the benchmark analysis shows stable embeddings, and reproducible trajectory-level production workflow. Across these benchmarks, GalaxyMemMD matches the setup quality of established tools, therefore serves as both a practical alternative and a capability extension for the membrane protein simulation community.

De novo antibody design promises to transform therapeutic antibody and nanobody discovery. Yet, success remains inconsistent across targets, with generative models often yielding no experimentally validated binders. Here we introduce MIMOSA (MIMic-Oriented Structural Antibody generation), a model-agnostic, inference-time framework that constrains diffusion-based antibody design pipelines to reproduce the interaction geometry and chemistry of a target's known cognate binder interface. Rather than searching for a productive interface from scratch, MIMOSA constrains generation around residue identities and geometries already known to support binding, while allowing the generative model to complete the surrounding antibody interface. We test MIMOSA on different interface topologies: contiguous motifs that fit within a single CDR loop and spatially dispersed hotspots distributed across the paratope. Applied without retraining to two architecturally distinct generators (RFAntibody and BoltzGen), this framework delivers experimentally validated binders-achieving sub-micromolar to single-digit nanomolar affinities-at 20-26% hit rates on three targets on which unconstrained methods produce zero or near-zero binders: KEAP1, uPA, and IL-8. By turning any target with a structurally characterised cognate binder into an accessible de novo design problem, this framework broadens the practical reach of generative antibody design to targets where unconstrained approaches currently fail.

Scientific design requires reasoning in scientific spaces that represent the objects being designed. Building on Scinetics' Chain of Science Tokens, we present chain-of-structures (COS) as a concrete protein-design case study that connects reasoning and generation through explicit structural states. In this technical report, we apply COS to enzyme design, where catalytic function requires precise local geometry within a compatible global fold, making it a demanding test of reasoning from chemical requirements to protein structure. Starting from a reaction specification and target length, COS couples a language reasoning expert with an all-atom generation expert across four rounds: catalytic core, active site, global backbone, and complete sequence and structure. Each reasoning state directly conditions generation, and the resulting structure becomes context for the next decision. Discrete residue constraints and continuous latents connect the two experts, enabling structural supervision of the reasoning interface. On the evaluated benchmark, COS leads the compared baselines on five of six metrics, achieving 57.0% EC3 agreement and a TM-score of 0.729.