Biology subjects

Kim, D.

Publications and source records attributed to Kim, D..

At least 19 recordsLinked to original sources

bioRxiv2026

High-resolution HLA and MICA/MICB typing of the 1000 Genomes panel: a resource for precision medicine

High-resolution characterization of human leukocyte antigen (HLA) and MICA/MICB variation is important for transplantation, immunogenetics, and the interpretation of disease associations within the major histocompatibility complex (MHC). However, publicly available population reference datasets remain limited in their resolution and coverage of these loci. We performed high-resolution typing of the 1000 Genomes Phase 3 panel across 13 HLA and MIC loci, generating HLA genotypes for 2,486 individuals, extended HLA-A, -B, and -C allele assignments, and MICA/MICB genotypes for 2,393 individuals representing 25 populations and five continental super-populations. Concordance with previously reported HLA genotypes was 96.2-97.4% at two-field resolution, and independent retyping confirmed the present assignments for samples exhibiting reciprocal genotype discrepancies with an earlier dataset. Analysis of extended class I variation within defined assay boundaries revealed that most of the 78 common three-field HLA lineages were dominated by a single extended sequence group. However, eleven lineages exhibited substantial heterogeneity, with the predominant group accounting for less than 80% of observed copies. The distinguishing sequence differences were noncoding, involving intronic and 3' untranslated regions, and their distributions varied among ancestral populations. These findings identify recurrent, population-relevant variants for which functional consequences remain to be established. Population analyses demonstrated extensive allelic diversity at classical class I loci and DRB1, whereas the greatest between-population differentiation occurred at selected class II and DRB-linked loci. Most common alleles were shared among populations but differed substantially in frequency. Multilocus haplotypes revealed characteristic population-associated combinations of HLA and MIC alleles and provided the basis for predicting MIC genotypes from existing HLA typing. Using two-field HLA-B and HLA-C genotypes, five-fold cross-validation achieved exact genotype prediction accuracies of 77.3% for MICA and 53.6% for MICB, compared with 4.6% and 21.6%, respectively, without HLA information. Together, these results provide an extended-resolution HLA and MIC reference resource, demonstrate the limited routine utility of universal fourth-field typing while identifying specific variants for functional investigation, and establish the potential for recovering MIC variation from existing HLA-typed cohorts. The resource also supports future investigations of MHC variation underlying disease associations and transplantation outcomes.

genetics↗

bioRxiv2026

Predicting conformational changes induced by biomolecular interactions

Biomolecular interactions trigger structural changes of protein complexes essential for cellular signaling, regulation, and drug action. Modern co-folding models can now generate detailed structures of multi-component complexes directly from sequence. However, recent evaluations reveal that they frequently fail to capture specific conformational states induced by a ligand or partner. Here, we show that explicitly modeling the structural transition from an unbound to a bound state, rather than generating the bound complex alone, better resolves these interaction-induced states. Our generative learning framework, K-Fold, transforms independently predicted unbound component priors into an assembled complex through an ultrafast diffusion bridge, operating entirely without multiple sequence alignments. By eliminating co-evolutionary dependencies, K-Fold accurately recovers non-coevolved interfaces in antibody-antigen and targeted-degrader complexes, as well as ligand-directed regulatory states in GPCRs and kinases. Furthermore, by coupling this diffusion bridge with diverse monomeric equilibrium ensembles, K-Fold provides a computationally tractable foundation for sampling complex multi-state landscapes without retraining. These results demonstrate that formulating molecular assembly as an explicit structural transition moves structure prediction beyond static geometries toward capturing biomolecular dynamics.

molecular biology↗

bioRxiv2026

Integrating Multisource Target Evidence with Machine-Learning Potency Prediction for Alzheimer's Disease Drug Discovery

Alzheimer's disease (AD) target evidence is distributed across biological, chemical, structural, regulatory, and trial-registry sources, complicating reproducible target evaluation and modeling. We developed a traceable data-to-model platform that integrates AGORA-centered evidence through Target Explorer, Drug-Target Lookup, and Model Portfolio workspaces. The resource contains 967 targets, 452,249 compounds, and 1,430,260 measurements. On a curated PIK3CA dataset, target-specific fine-tuning of the Boltz-2 affinity head increased Pearson r from 0.248 to 0.560 and reduced mean absolute error from 0.900 to 0.706 on 469 held-out compounds. Separately, we screened 42 regression models over 50 random splits for ten moderate-sized target datasets; a random forest model for XDH achieved the highest mean R^2 (0.815). The platform connects curated evidence, model-ready datasets, reproducible evaluation, and public results while clearly separating contextual evidence from therapeutic claims and random-split performance from generalization to unseen chemical scaffolds.

bioinformatics↗

bioRxiv2026

Chirality-Driven Photonic PCR Enables Label-Free Genetic Amplification

Photonic polymerase chain reaction (PCR) offers a rapid alternative to conventional thermal cycling for genetic amplification, but current platforms are limited by low photothermal conversion efficiencies and a dependence on bulky fluorescent monitoring systems. Here we report a chirality-driven photonic amplification paradigm that achieves both accelerated nucleic acid replication and direct, label-free detection. By leveraging chiral plasmonic nanoparticles, we exploit their elevated absorption cross-sections to achieve exceptional light-to-heat conversion and rapid thermal relaxation, significantly accelerating heating and cooling rates during cycling. We show that localized chiral near-fields generated by these chiral plasmonic nanoparticles optimize DNA polymerase conformational dynamics, lowering the activation energy barrier for synthesis and boosting amplification efficiency. Furthermore, by tracking the reaction with circular dichroism spectroscopy, we directly monitor the intrinsic chiroptical signature of amplifying DNA, completely eliminating the need for external fluorescent dyes or labeled probes. We validate this system using SARS-CoV-2 E-gene plasmid DNA templates, demonstrating rapid, label-free amplification and single-copy detection within clinically relevant timelines. By bridging asymmetric photothermal dynamics with chiral-field-induced enzymatic activation, this work translates chiroplasmonic materials from fundamental physics into high-performance, point-of-care molecular diagnostics.

bioengineering↗

bioRxiv2026

Limits of marker information transfer]{Limits of marker information transfer in cross-panel cytometry imputation

Marker imputation can bridge cytometry datasets acquired with different panels and platforms. Its validity depends on whether common markers retain information about unmeasured target markers and whether those relationships transfer across assay contexts. We evaluated these limits in two paired flow cytometry-CyTOF cohorts comprising 134 patients and 257 specimens. A patient-held-out framework distinguished within-platform same-cell reconstruction from cross-platform distribution translation without cell correspondence. We compared conventional machine-learning and deep generative estimators with median imputation and used permutation controls to estimate the contributions of patient matching and source-exclusive markers. Evaluation covered marker values, specimen distributions, phenotype utility, and directional predictive relationships among markers. Within platforms, exact k-nearest neighbors reduced mean absolute error by 12.8-35.0% relative to median imputation. However, the median-based reconstructions were contracted, with a median ratio of predicted to observed interquartile range below 1 in every context and evaluable population. Patient matching improved cross-platform distribution recovery, whereas adding source-exclusive markers or higher-capacity estimators did not consistently produce net gains. In the independently annotated acute myeloid leukemia (AML) cohort, reconstructed markers largely preserved broad-lineage classification but did not recover fine-phenotype information available through direct measurement. In AML-CyTOF, the measured full panel outperformed the reconstructed representation by 10.25 percentage points for fine phenotypes. Predictive relationships among markers were context-specific. Most targets were better explained by several markers than by one. Marker selection based on these relationships showed a trade-off between panel size and pairwise coverage. Paired samples and imputation can therefore connect selected marker distributions across panels but should not be treated as substitutes for direct measurement. Cross-panel cytometry should be evaluated with patient-held-out tests matched to the resolution of available ground truth and the intended downstream use.

bioinformatics↗

bioRxiv2026

Decomposing scanner batch effects and their impact on pathology foundation models

Pathology foundation models (PFMs) are expected to represent the same tissue in the same way on any scanner, but their embeddings still carry scanner signals. Scanner-related image differences and the scanner sensitivity of PFMs have each been described. How tissue-dependent changes in color and spatial frequency relate to shifts in PFM representations, and what this means for correction, is less understood. We imaged 103 slides from 37 normal tissue types on six scanners (PanNormal; 95,780 matched locations) and used the real paired images as targets. We used the public PLISM dataset for external validation. First, we decomposed scanner differences into image properties. Color and spatial frequency changed together, by amounts that depended on tissue type. Second, corrections that matched these measured properties brought images closer to the target more reliably than representations. Pix2Pix reduced the image residual by 69% but increased UNI distance to the target by 27%. The same frequency matching helped one scanner and hurt another. Probes still distinguished corrected from real target embeddings after every correction. Third, changing one frequency band at a time showed that PFMs differ in how strongly each band moves them and whether it moves them toward the target. Across eight PFMs pretrained on whole-slide image tiles, the models that moved least under these small changes were the most robust to real scanner differences (Spearman {rho}=-0.90). Aligning images and aligning representations are therefore different goals. The gap between them depends on scanner, tissue, and model.

bioinformatics↗

bioRxiv2026

The porin MspA is a phage-specific determinant of mycobacteriophage infection

Mycobacteriophages are being developed as therapeutics for mycobacterial infections, but their highly variable host ranges limit this application and the bacterial factors that determine whether infection succeeds remain poorly defined. The determinants characterized so far are dominated by cell envelope lipids and glycolipids. Here, we performed a genome-wide CRISPRi screen in Mycobacterium smegmatis to identify host functions associated with susceptibility to BPs{Delta}33HTH HRM10. After prolonged phage challenge, the outer membrane porin genes mspA and mspC were among the most strongly enriched hits, whereas the closely related paralogs mspB and mspD were not enriched despite sharing high amino-acid identity across the family. Deletion of mspA or mspC reduced BPs{Delta}33HTH HRM10 plaque formation and produced smaller, more turbid plaques, with the stronger phenotype following loss of mspA, the major outer membrane porin of M. smegmatis. Across a panel of 25 mycobacteriophages, 11 showed either a more than tenfold reduction in titer or loss of discrete plaque formation on {Delta}mspA, while the remainder were largely unaffected. These findings identify outer membrane porins as phage-dependent determinants of mycobacterial susceptibility, revealing an additional layer of specificity in phage infection beyond previously characterized cell envelope lipids and glycolipids.

microbiology↗

bioRxiv2026

Netrin signaling shapes sexually dimorphic neurite branching of a sex-shared interneuron in C. elegans

Sexually dimorphic behaviors arise from underlying differences in neural circuit architecture between the sexes, yet the molecular mechanisms that establish these structural differences remain poorly understood. In Caenorhabditis elegans, the sex-shared interneuron AVG exhibits sexually dimorphic synaptic connectivity, offering a model for investigating the development of sexually dimorphic neuronal morphology. Here, we identified that AVG displays male-specific neurite branching and demonstrated that the conserved axon guidance cue UNC-6/Netrin and its receptor UNC-40/DCC are essential for this structural remodeling. Cell-specific manipulation of sexual identity revealed that male identity confers branching competence to AVG. Loss of UNC-6 or UNC-40 abolished branch formation, whereas loss of the alternate Netrin receptor UNC-5/Unc5 had no detectable effect. Cell-specific rescue experiments showed that UNC-40 functions in AVG through its UNC-6 binding domains. Furthermore, genetic ablation of the male-specific presynaptic neuron PCA, which expresses unc-6, diminished AVG branch length, establishing PCA as a candidate local source of UNC-6. Together, our findings support a model in which presynaptic UNC-6/Netrin cues from male-specific neurons promote dimorphic neurite branching in sex-shared neurons, illustrating how evolutionarily conserved signaling molecules are redeployed during sexual maturation to sculpt sex-specific neural circuits.

neuroscience↗

bioRxiv2026

Conserved mycobacteriophage protein uses structural mimicry to displace host initiation factors

Compact viral genomes encode sophisticated strategies to reprogram host gene expression, often by co-opting rather than replacing the host RNA polymerase. How bacteriophages achieve such control in Mycobacteria remains largely unknown. Here, time-resolved interaction proteomics of mycobacteriophage D29 infection identifies a conserved two-protein module, gp53-gp52, that binds the host RNA polymerase. Affinity purification confirmed association of both proteins with the polymerase, while native mass spectrometry demonstrated that gp53 displaces the host sigma factor. Structural modelling places gp53 at the sigma-binding surface of RNA polymerase, consistent with molecular mimicry of sigma-factor engagement. The module is conserved throughout Cluster A mycobacteriophages and occurs in fused or split genomic architectures. These findings suggest sigma-factor displacement as a mechanism of transcriptional takeover in viruses of mycobacteria and show how a compact viral module can repurpose an essential host molecular machine

microbiology↗

bioRxiv2026

The Lag3 KIEELE motif couples HPK1 to metabolic and functional regulation of Foxp3+ regulatory T cells

Lymphocyte activation gene-3 (Lag3) is an inhibitory receptor that limits effector T cell activation and function. We previously demonstrated that Lag3 supports regulatory T (Treg) cell function through Myc-dependent metabolic reprogramming. However, the intracellular mechanisms underlying this regulation remain incompletely understood. Here, we generated mice with Treg-specific deletion of the Lag3 KIEELE motif and demonstrated that this motif is required for Treg-mediated suppression of autoimmune inflammation. Loss of the Lag3 KIEELE motif led to enhanced TCR/PI3K-Akt/MEK-ERK signaling and Myc expression, accompanied by increased glycolytic activity and acquisition of effector T cell-like features. We also identified that hematopoietic progenitor kinase 1 (HPK1) and Cbl-interacting protein of 85 kDa (CIN85) associate with Lag3 through the KIEELE motif. HPK1 depletion partially recapitulated the phenotypic changes and impaired Treg suppressive function in vivo. Together, these findings identify the KIEELE motif as a critical link between Lag3-associated signaling machinery and metabolic regulation that sustains Treg suppressive function during autoimmune inflammation.

immunology↗

bioRxiv2026

P4HA1 sustains endocrine resistance in luminal breast cancer via regulation of cellular stress

The hormone receptor-positive luminal subtype accounts for approximately 70% of all diagnosed breast cancer cases, with endocrine regimens serving as the first-line therapy. However, 30% to 50% of luminal breast cancer patients acquire therapeutic resistance with five to ten years of treatment. Understanding the mechanisms of endocrine resistance and identifying new therapeutic targets remain urgent priorities. Here, we demonstrate that expression of the alpha subunit of prolyl 4-hydroxylase (P4HA1), a key collagen-modifying enzyme, is significantly upregulated in endocrine therapy-resistant luminal breast cancer cells compared to parental endocrine therapy-sensitive cells. Clinically, elevated P4HA1 expression correlates with shortened relapse-free survival in patients with luminal breast cancer. Genetically knocking down P4HA1 or pharmacologically inhibiting its activity successfully restores the sensitivity of endocrine-resistant luminal breast cancer cells to hormone therapy. Mechanistically, we reveal that combining P4HA1 inactivation with endocrine treatment dramatically enhances endoplasmic reticulum (ER) stress-related gene expression. Consistently, treating resistant cells with an exogenous ER stress inducer similarly enhances sensitivity to hormone therapy. Furthermore, we demonstrate that ER stress induced by P4HA1 inactivation is molecularly mediated by an accumulation of reactive oxygen species (ROS). Together, these findings uncover a novel, non-canonical role for P4HA1 as a critical mediator linking endocrine resistance to cellular stress adaptations, positioning P4HA1 as a promising therapeutic vulnerability in endocrine-resistant breast cancer. P4HA1 serves as a necessary role in maintenance resistant state, making it a viable target for combination therapies to overcome endocrine resistance.

cancer biology↗

bioRxiv2026

fastACCORD enables ultrahigh-dimensional partial correlation modeling for multi-omic data integration

Gene co-expression networks reflect transcription factor-associated regulation together with epigenetic influences such as DNA methylation, chromatin states, and histone modifications. To distinguish gene-gene dependencies that persist after accounting for methylation covariation, large-scale statistical inference conditioning on hundreds of thousands of molecular features is necessary, but the task remains computationally intractable for conventional Gaussian graphical modelling approaches. Here we present fastACCORD, a scalable computational framework for ultrahigh-dimensional partial correlation modeling. fastACCORD combines row-separable optimization, {ell}2 stabilization, and a semismooth Newton solver in a PyTorch implementation for CPU and CUDA-enabled GPU hardware. We applied it to matched transcriptomic and methylomic profiles from 16 TCGA cancer types and obtained joint networks containing >300,000 molecular features per cancer. The resulting multi-omic networks revealed cancer-specific methylation-expression dependencies, including recurrent methylation-associated expression repression of metabolic genes. Methylation-adjusted gene co-expression networks were sparser than networks estimated from mRNA data alone, but more enriched for ChIP-seq-supported TF-target relationships and curated co-regulon annotations. TF-target subnetworks further revealed cancer-specific architectures consistent with lineage identity, oncofetal reactivation, and tumor microenvironment-associated programs. Together, these analyses establish fastACCORD as a practical framework for ultrahigh-dimensional multi-omic partial correlation modeling and demonstrate how joint modeling of transcriptomic and epigenomic measurements can refine the interpretation of gene co-expression networks.

bioinformatics↗

bioRxiv2026

Cascaded Nano-ring Dialyzer for Continuous Regenerative Peritoneal Dialysis

End-stage kidney disease (ESKD) requires lifelong kidney replacement therapy, yet conventional hemodialysis and peritoneal dialysis remain limited by intermittent treatment, restricted mobility, and repeated dialysate exchange. Here, we developed a continuous regenerative peritoneal dialysis platform based on a nano-ring-assisted nanoelectrokinetic dialyzer for portable artificial kidney applications. We first identified key geometric and electrokinetic parameters governing the removal performance of nano-ring meshes, including mesh dimension, nano-ring coating thickness, the number of cascaded nano-rings, and the relative orientation between fluid flow and the applied electric field. We then developed a single nano-ring dialyzer operating at 1.33 mL/min and scaled the treatment capacity to approximately 10 mL min/min through parallel integration of multiple modules. In a canine model, the closed-loop system achieved approximately 10 % reduction in circulating uremic toxin levels during continuous treatment. We further improved the biocompatibility of the regenerated dialysate by integrating plate-type electrodes, activated carbon, bicarbonate buffering, and UV-C treatment. In an unanesthetized canine model, the improved system sustained continuous toxin removal for 4 h while major hepatic and inflammatory markers remained within recoverable ranges. These results demonstrate that cascaded nano-ring enables continuous dialysate regeneration with scalable throughput, reduced dependence on fresh dialysate, and in vivo control of uremic toxins, providing a technological foundation for portable or wearable regenerative peritoneal dialysis systems.

bioengineering↗

bioRxiv2026

PREpiBind: Protein Representation-integrated Epitope--MHC Class II Binding Prediction

Accurate prediction of peptide-MHC class II (pMHC-II) binding remains challenging because of extensive MHC polymorphism and context-dependent peptide recognition, and it is unclear which protein representation best captures these determinants when the downstream model is held fixed. Here, we present PREpiBind, a pMHC-II prediction method based on a dual-stream, joint-attention framework that integrates MHC and epitope representations. We integrated ten protein representations spanning substitution-matrix, structure-prediction-derived, and protein language model (PLM) families and evaluated them under a common downstream architecture and identical splits across Qualitative, mass spectrometry (MS), and thresholded IC50 datasets. Overall, PREpiBind with PLM representations yielded higher performance than the evaluated reference methods in the Qualitative and MS datasets, whereas NetMHCIIpan-4.3 was higher on the thresholded IC50 datasets using its binding-affinity head. In pooled evaluations, PLM representations yielded the highest ROC-AUC values among the tested representations. ESM3 Large led the Qualitative dataset with an AUC value of 0.927 +/- 0.002, and PLMs also led on the MS and thresholded IC50 datasets. This advantage remained, but narrowed in allele-wise and leave-one-molecule-out evaluations, where structure-prediction-derived Chai-1 was competitive with the leading PLM representations. Under cross-species H2-out transfer, PLMs led when all H2 rows were pooled, whereas Chai-1 led when the eight H2 molecules were weighted equally. The small H2 panel did not support a stable ordering among these leading representations. The results indicate that protein representation choice should depend on the intended prediction scenario rather than on a single global ranking. PREpiBind is an openly available pMHC-II prediction framework with modular and flexible protein representations.

bioinformatics↗

bioRxiv2026

A methylome-guided toolkit for genetic engineering of the fast-growing cyanobacterium UTEX 3222

Cyanobacterium aponinum UTEX 3222 is a fast-growing marine cyanobacterium that reaches high cell densities, making it an attractive chassis strain for low-cost biomanufacturing. However, past efforts to genetically engineer UTEX 3222 have been unsuccessful. We hypothesized that the complex restriction modification systems in the UTEX 3222 genome constitute a major barrier to genetic transformation. Using a combined methylation avoidance and methylation mimicking approach, we overcame this barrier and successfully transformed UTEX 3222 by natural transformation, generating a large panel of fully segregated axenic strains. Transformation method was also established for a novel shuttle vector that replicates in UTEX 3222. A large-scale promoter screen identified promoters which exhibited strong expression at high cell density. These advances establish the foundation for developing UTEX 3222 as a scalable platform for photosynthetic biomanufacturing.

synthetic biology↗

bioRxiv2026

Folic acid polyglutamylation triggers MAP2K5 signaling to induce adipocyte lipid catabolism

Adipocyte signaling adaptively responds to macronutrients, but how essential micronutrients impact lipid homeostasis remains poorly understood. Here, we demonstrate that folate polyglutamylation, the sequential conjugation of glutamate residues to folate, serves as a dynamic biochemical process regulating metabolic signaling. Using our folate metabolomics platform, we show that polyglutamylated folic acid accumulates in healthy adipose tissue, but is depleted in obesity across mice and humans, independent of circulating folate levels. Genetic ablation of the polyglutamylation enzyme folylpolyglutamate synthase (Fpgs) in adipocytes suppresses lipid catabolism to induce cell-autonomous lipid accumulation, operating independently of canonical adipogenesis or downstream one-carbon flux. Target-engagement proteomics identifies monoglutamylated folic acid as a Map2k5-interacting molecule that inhibits Map2k5 activity, whereas Fpgs-mediated folic acid polyglutamylation acts as a chemical switch that disrupts this interaction to promote lipolysis. In vivo, whole-body inhibition of Map2k5 or adipose-targeted genetic depletion of Fpgs increases body fat mass in the absence of dietary obesogenic triggers. Furthermore, single-cell and single-nuclear transcriptomic analyses establish the Fpgs-Map2k5 pathway as a core transcriptional signature of mouse and human adipose tissues. These findings uncover a non-canonical signaling role for folate that regulates adipocyte lipid homeostasis.

biochemistry↗

bioRxiv2026

High frame rate in vivo two-photon microscopy to quantify murine cerebrospinal fluid flow heterogeneity

Cerebrospinal fluid (CSF) flows through perivascular spaces (PVSs) surrounding brain vasculature, and impaired flow has been linked to neurodegenerative diseases such as Alzheimer's. However, the mechanisms driving CSF flow and its oscillatory dynamics remain poorly understood. Using high frame rate two-photon imaging (up to 113 Hz), we show that CSF flow is spatially heterogeneous and highly pulsatile, quantified using a regional CSF pulsatility index (PI). Reduced-order simulations with realistic domain length and Windkessel boundary conditions show that arterial pulsations generate peak CSF velocities but contribute negligibly to net transport. PI varies non-monotonically in space and substantially with downstream hydraulic resistance and compliance, indicating that CSF pulsatility is an emergent property of the hydraulic network rather than simply a consequence of local arterial wall motion. Measurements and simulations of the phase between arterial wall motion and CSF velocity further constrain proposed CSF driving mechanisms. Finally, we perform mouse-specific simulations to estimate average and peak wall shear stresses which support the potential role of pulsatile CSF flow in perivascular mechanotransduction. Together, these results establish CSF pulsatility as a measurable signature of network-scale CSF dynamics.

neuroscience↗

bioRxiv2026

TAPAS: Learned integration of AlphaFold3 confidence and geometric features for TCR-pMHC binding prediction

Motivation Recent advances in biomolecular structure prediction, exemplified by AlphaFold3, have opened new opportunities for the prediction of TCR-pMHC binding specificity. Although individual AlphaFold3 confidence metrics provide informative binding signals, their predictive performance varies across datasets, highlighting the need to combine complementary signals rather than rely on any single metric. Results We present TAPAS, a tabular learning framework that integrates AlphaFold3-derived interface confidence and structural geometry with sequence embeddings. Although no single zero-shot metric was the strongest across all benchmarks, TAPAS was consistently the top-ranked method on VDJdb and two external benchmarks, matching or exceeding the strongest zero-shot AlphaFold3 metric. Feature group ablation showed that the contributions of sequence, confidence, and geometric features varied across evaluation settings, and that combining them resulted in the best overall performance. These results highlight the value of integrating complementary structural and sequence features within a unified framework for robust TCR-pMHC binding prediction.

bioinformatics↗