Biology subjects

Cho, H.

Publications and source records attributed to Cho, H..

At least 19 recordsLinked to original sources

bioRxiv2026

A spatial and single-cell transcriptomic atlas of the embryonic synovial joint and its explant model

Synovial joint formation requires the coordinated differentiation of multiple cell types while balancing opposing migratory cues that recruit joint progenitors and exclude other populations. More specifically, a subset of limb bud mesenchyme is recruited into the interzone while neighboring chondrocytes and endothelial cells are excluded. Existing single-cell references sample this process sparsely, without position, or without the surrounding mesenchyme. Here, we present an atlas of the mouse hindlimb from E11.5 to E15.5 that pairs 228,703 single-cell transcriptomes with 47 spatial transcriptomic sections, annotates 56 cell types by their location, and extends the same profiling to limb explant culture. Explants preserve the lineages and axial patterning of the limb for three days, with delayed chondrocyte hypertrophy. Using the atlas, we found that Sema3c is co-expressed with Sox9 and Gdf5 during condensation and is required for chondrogenesis and joint formation in culture; that interzone cells carry a chondrogenic signature early and a tenogenic signature late; and that a core interzone signature includes eleven previously unreported markers, one of which, Igfbp5, is required for knee formation in explants.

developmental biology↗

bioRxiv2026

Geometry-dependent mRNA delivery with programmable RNA-DNA nanostructures

Efficient intracellular delivery of neoantigen-encoded mRNAs is central to the development of effective cancer vaccines and immunotherapies. However, how the nanoscale geometry of delivery carriers impacts mRNA delivery remains poorly understood, largely because conventional delivery platforms do not readily permit systematic variation of carrier architecture while maintaining comparable chemical composition. Here, we develop a modular RNA-DNA hybrid nanoplatform that enables controlled investigation of geometry-dependent mRNA delivery. Guided by a unified design principle, we construct a series of two- and three-dimensional nanostructures with programmable architectures and systematically compare their delivery performance using consistent functional readouts. We further investigate how mRNA embedding strategies and aptamer-mediated functionalization influence intracellular delivery efficiency. By providing precise control over carrier geometry, this platform establishes a framework for elucidating structure-function relationships in mRNA delivery and offers a versatile strategy for engineering targeted nucleic acid delivery systems.

bioengineering↗

bioRxiv2026

INFORME: coupling information-theoretic experimental design with nonlinear mixed-effects modeling for efficient observation scheduling

Mathematical models of treatment response can inform individualized therapy, but their calibration often requires longitudinal measurements that are costly, burdensome, and collected on fixed schedules. Such schedules may be inefficient, over-sampling patients whose response is already well characterized while delaying informative measurements for those whose model parameters remain uncertain. We present INFORME (INFORmation-theoretic design with Mixed Effects), a framework that combines Bayesian information-theoretic experimental design with nonlinear mixed-effects modeling to adaptively select each patients next measurement time. Population and response-subgroup parameter distributions learned from an existing cohort provide informative priors, allowing candidate measurement times to be ranked by their expected reduction in patient-specific parameter uncertainty. As observations accumulate, priors can be updated to reflect the response subgroup most consistent with the patients data. We evaluate INFORME in two radiotherapy datasets: 150 synthetic tumor volume trajectories from a hybrid cellular automaton model of prostate cancer spheroids (HD1) and longitudinal tumor volumes from 39 patients with head-and-neck cancer (HD2). In HD1, population priors allowed omission of both pretreatment scans, while adaptive scheduling reduced the protocol from nine scans to three or four, with the response group identified from a single post-treatment scan on day 27. In HD2, the adaptive schedule used three scans instead of six and improved prediction by delaying the first on-treatment scan from week 1 to week 2, avoiding transient dynamics that produced false-positive and false-negative response projections. Across both datasets, the adaptive schedules used a mean of 2.7 scans in stead of seven and advanced completion of the patient-specific prediction by a mean of 15.5 days (95% CI, 6.7-24.3) relative to the equidistant protocol, while treatment duration remained unchanged. INFORME therefore reduces measurement burden and accelerates patient-specific prediction by concentrating observations at times that are most informative for model calibration.

systems biology↗

bioRxiv2026

PhageTAILor leverages machine learning for phage tail-like elements detection and classification in plant-associated bacteria

Phage tail-like elements (PTEs) -- tailocins, bacterial type VI secretion systems (T6SS), and extracellular contractile injection systems (eCIS) -- are contractile nanomachines that bacteria use to kill their neighbors and compete within their micro-ecosystems. PTEs help shape microbial community composition. Most PTE detection tools only detect a single PTE class. Moreover, most tailocin detection methods are largely restricted to Pseudomonas, leaving a key part of tailocin diversity uncharacterized. In this work, we present PhageTAILor (https://github.com/hjcho-bio/PhageTAILor), an integrative and fully automated pipeline that detects and classifies prophages and 3 PTE classes from bacterial genomes. PhageTAILor combines a 6-detector homology-based candidate search (geNomad, tail-gene, PHROGs-tail, SecReT6, eCIStem, and a divergence-tolerant tail-HMM detector) with a LightGBM classifier comprising 1 multiclass and 3 binary heads, trained on 6,501 bacterial genomes carrying 13,082 prophages and PTEs. A phylogeny-free feature matrix used in our model keeps predictions reproducible between model construction and user inference. PhageTAILor performs strongly at the genome level and generalizes beyond its Pseudomonas-rich training set. On a 76-strain cross-clade benchmark, PhageTAILor detected tailocins at F1 = 0.955. Furthermore, it identified 12 of 13 experimentally validated tailocins spanning five genera versus 2 of 13 for a Pseudomonas-restricted tool TattleTail. PhageTAILor also demonstrated sensitivity equivalent to viral detection tool geNomad while avoiding its higher false-positive rate. Applied to 7,925 plant- and soil-associated bacterial isolates, PhageTAILor showed that prophages in the phyllosphere and tailocins in plant-associated bacteria, whereas eCIS are enriched in soil. PhageTAILor is distributed as an open-source, modular pipeline with a command-line interface.

microbiology↗

bioRxiv2026

A generalized resource-allocation growth law reveals mechanisms of differential responses to intermittent androgen deprivation therapy in prostate cancer

Resource-limited growth frameworks often developed independently. Here, we derived a growth law by assuming a cell balancing internal competing resource demands and generalizing the growth signal through a shape parameter h, the trade-off exponent. In Liebig's single-limiting-resource regime, it recovers Droop's cell quota at h=1 and approaches Prater's growth efficiency as h grows large. h determines how sharply growth declines when allocation departs from optimal, thereby shaping populations evolution. Cyclic environmental stress tests such growth laws. Hence, we built a model from this growth law and applied it to prostate cancer growth in patients undergoing cyclic intermittent androgen deprivation therapy. We showed the model recapitulated longitudinal prostate specific antigen (PSA) and serum androgen measurements from 71 prostate cancer patients using a nonlinear mixed-effects framework (pooled and median individual R^2 {approx} 0.9 for PSA and {approx} 0.8 for serum androgen). h is well-constrained by the data and separates treatment outcomes (success vs. failure, p = 0.003). The results predict that failing cases have cancer that can adapt to a wide range of growth conditions with minimal cost. A simplified model recapitulated PSA dynamics from 32 prostate cancer patients on cyclic adaptive therapy (R^2 {approx} 0.8). Estimates of h for castration resistant cancer in adaptive cohort were similar to failure cases in the intermittent cohort. We found no statistical differences in estimated parameters between the adaptive vs. standard of care arms, suggesting that adaptive schedule may drive the differential outcome. The results support this growth law, providing a foundation for resource-limited growth across biological systems.

ecology↗

bioRxiv2026

Convergent biology, divergent drivers: a cross-species comparison of human and canine invasive urothelial carcinoma

Traditional animal models are often inbred and genetically uniform. This makes them powerful for controlled experiments, but it limits how well they represent the patient-to-patient variation seen in real-world disease. Comparative oncology seeks to address this gap by studying naturally occurring cancers in outbred companion animals, especially dogs. Canine medicine offers two important advantages: first, prospective trials can often be completed faster than in humans and second, dogs are already part of the translational pipeline through pharmacokinetic and toxicology studies. Here, we assessed the transcriptional fidelity of human and canine invasive urothelial carcinoma in primary tumors and patient-derived organoids. We then used single-cell and spatial data to resolve the underlying cellular organization. Despite strong species and platform differences, human and canine tumors preserved the same major luminal-basal structure and a similar tumor microenvironment. The two species reached this shared biology through different recurrent mutations. These included FGFR3 alterations in humans and BRAF alterations in dogs, which converged on overlapping pathways and a luminal phenotype. Human and canine organoids also underwent a similar shift in culture. Both became more proliferative and metabolic while losing inflammatory programs. Thus, organoids preserved important tumor biology while introducing predictable platform effects. Single-cell and spatial analyses showed that the luminal-basal axis reflects a gradient of cell states organized around the tumor-stroma boundary, rather than two discrete tumor types. This helps explain why bulk RNA-sequencing subtypes are reproducible but coarse. Together, these findings define where canine and human bladder cancer agree, where they differ, and how dogs can support parallel therapeutic and diagnostic development.

bioinformatics↗

bioRxiv2026

Solanum americanum Bs2 and ZAR1 homologs recognize Xanthomonas euvesicatoria effectors essential for infection.

Multiple recognition events of pathogen-secreted effectors by immune receptors confer robust disease resistance in plants. Understanding the underlying mechanisms facilitates the discovery and deployment of valuable resistance genes for crop protection. Xanthomonas euvesicatoria causes devastating bacterial spot disease in solanaceous crops but cannot infect the wild relative Solanum americanum. Here, we identified X. euvesicatoria type III effectors (T3Es) that induce cell death in S. americanum when transiently expressed in leaf. By quantifying immune responses of the S. americanum SP2273 accession to X. euvesicatoria multiple-T3E knockout mutants, we demonstrated that at least nine T3Es (AvrBs2, XopAP, XopAU, XopE1, XopJ3, XopM, XopN, XopX, and XopZ1) collectively contribute to effector-triggered immunity (ETI). Among these, AvrBs2 and XopJ3 were the primary drivers of ETI, eliciting robust cell death and defense gene expression when naturally delivered into plant cells. We next generated S. americanum lines concomitantly edited at the corresponding immune receptor loci, SaBs2 and SaZAR1 (SP2273-bz). Genetic complementation of SP2273-bz confirmed that AvrBs2 and XopJ3 are specifically recognized by each of the four SaBs2 homologs and by SaZAR1, respectively. Moreover, enhanced growth of X. euvesicatoria on the characterized SP2273-bz line indicated that both effectors are required for bacterial multiplication in S. americanum. Together, our findings provide a framework for understanding the mechanisms of ETI-mediated resistance and establish a genetic foundation for resistance breeding in solanaceous crops.

plant biology↗

bioRxiv2026

A Label-Free Multi-Metric Pipeline for Benchmarking Single-Cell RNA-Sequencing Clustering and Testing the Reproducibility of Cell-Type Heterogeneity

A discovered sub-population from single-cell transcriptomic data is only meaningful if it is reproducible, yet clustering is usually done with one method on one embedding and rarely tested. We present a label- free, multi-metric pipeline that reframes clustering as an auditable, methods-blind decision and separates two notions of stability that are commonly conflated: reproducibility under cell resampling (bootstrap) and reproducibility under re-embedding (retraining the representation). The pipeline evaluates seven clustering configurations across cluster counts using five non-redundant quality metrics. As a whole- dataset control on a mouse retinal atlas, it recovers an eight-cell-type annotation at 96.3% accuracy (adjusted Rand index, ARI = 0.91) without labels. We then validate the discovery mode on two cell types with opposite ground truth. On bipolar cells, which have well-established subtypes, the pipeline accepts the sub-structure: across-embedding reproducibility rises with cluster number to a high plateau (mean pairwise ARI [~]0.93 near the [~]15 known bipolar subtypes), with quality metrics improving in parallel. On rod photoreceptors, treated as homogeneous, it rejects over-clustering: the metric-selected partition passes a bootstrap-stability check but is not reproducible when the embedding is retrained (mean pairwise ARI = 0.69), and the metrics do not improve with cluster number. On synthetic data, the test recovers real structure down to a 5% subpopulation while rejecting null data (high sensitivity and specificity). Bootstrap stability alone is therefore insufficient evidence for sub-population; the across-embedding test discriminates real sub-structure from over-clustering and applies to any cell type as a reproducible alternative to single-method, single-embedding clustering.

bioinformatics↗

bioRxiv2026

NNMT Loss Drives Cancer Progression by enhancing SAM availability for mTORC1 Signaling and Chromatin Methylation

Aberrant epigenetic reprogramming together with dysregulated mTOR signaling are hallmarks of cancer, where altered chromatin methylation and nutrient-sensing pathways cooperate to drive tumor progression. S-adenosylmethionine (SAM), the universal methyl donor, is essential for these processes, yet how tumors sustain elevated SAM availability to support oncogenic transmethylation reactions remains poorly defined. Here, using prostate cancer (PCa) as a model system, we identify nicotinamide N-methyltransferase (NNMT) as a critical metabolic-epigenetic regulator and tumor suppressor. Using a prostate-specific Nnmt knockout mouse model, we demonstrate that NNMT loss accelerates PCa progression, particularly in the context of Pten deletion, resulting in infiltrating carcinoma and reduced survival. Mechanistically, NNMT functions as a "SAM-sink," and its loss increases intracellular SAM abundance, thereby activating mTORC1 signaling through SAMTOR-dependent sensing and broadly enhancing chromatin methylation. In human PCa, recurrent genomic deletions of NNMT occur in up to 7% of cases, and NNMT protein expression is largely absent in primary tumors and metastases. NNMT-deficient PCa cells exhibit elevated SAM:SAH ratios, increased histone methylation, and heightened mTORC1 activity, enabling sustained tumor growth even under dietary methionine-restriction (MR). Notably, combined MR and pharmacologic mTORC1 inhibition synergistically suppresses the growth of NNMT-deficient tumors, revealing a previously unrecognized therapeutic vulnerability. Collectively, these findings establish NNMT as a key tumor suppressor that constrains SAM-driven epigenetic and signaling programs in PCa and suggest a rational, diet-based therapeutic strategy for advanced cancers with NNMT loss.

cancer biology↗

bioRxiv2026

Human meningiomas arise from developmental meningeal mosaicism

Meningioma is the most common primary intracranial tumour, yet its genetic origin and the temporal sequence of mutational events remain poorly defined. Here, we analysed 80 triple-matched tumour, histologically normal meninges, and blood samples from 22 patients. NF2 or TRAF7 driver mutations are detectable in phenotypically normal meninges in 81.8% of cases (95% CI: 59.7-94.8%; VAF [~]0.02%), corroborated by high-depth sequencing, single-cell cloning, and phylogenetic analysis. By distinguishing developmental mosaic mutations from postnatal tumour-private mutations, we revealed distinct mutational signatures and resolved the temporal sequence of meningioma evolution. The developmental origin was further underscored in patients with multiple meningiomas, where identical driver mutations were shared across genomically distinct tumours and meninges, and in intraventricular meningiomas (IVM), where driver mutations were detected in distant cranial dura. Reconstruction of mutational timing revealed lineage-specific trajectories, linking mosaicism to diverse disease presentations including solitary meningioma, IVM, meningiomatosis, and NF2-related schwannomatosis. Together, these findings reveal an early origin of human meningioma, in which developmental mosaicism establishes a pre-neoplastic field within the meninges, providing a developmental framework for adult tumourigenesis.

Cancer Biology↗

bioRxiv2026

A harmonized single-cell RNA-seq atlas of human localized and metastatic prostate cancers and benign tissues

Single-cell RNA sequencing (scRNA-seq) effectively captures the differences in transcriptomic landscape of cell types and cell states between benign and cancer tissues. Pooling publicly available datasets distributed across independent studies enables increased sample representation and cross-study comparisons. Here we present a harmonized scRNA-seq atlas of the human prostate constructed by integrating 17 available studies, comprising 163 samples from 106 donors. The dataset contains benign tissue, primary tumors, and metastatic disease profiles. Raw sequencing FASTQ data files were uniformly reprocessed to minimize technical variability. Study metadata were curated and standardized using a unified schema capturing donor identity, tissue site, disease context, and histologic grade. Post quality control, the integrated dataset contains 754,000 high-quality cells. Harmonized cell type annotations were generated using a pseudobulk correlation framework informed by multiple reference resources. The workflow identified 17 distinct cell types representing epithelial, mesenchymal, and immune compartments of the prostate. The processed expression matrices, standardized metadata, and analysis workflows are publicly available to support reproducible analysis and enable exploration of heterogeneity across prostate disease states.

cancer biology↗

bioRxiv2026

Glutamatergic neuron-tumor synapses shape human glioblastoma cell states through radial glia plasticity

Glioblastoma (GBM) is a devastating primary brain tumor with remarkable inter- and intra-tumoral heterogeneity. GBM cells assume a spectrum of neurodevelopmental-like phenotypes and co-opt normal neurophysiological processes, which include synaptic integration with their neuronal microenvironment. This is mediated by neuron-tumor synapses (NTS) that predominantly involve glutamatergic receptors, which drive calcium elevations that promote tumor proliferation and invasion. The exact relationship between synaptic signaling and tumor cell fate specification, however, remains largely unexplored. Here, we develop and leverage a synapse-optimized human organoid tumor transplantation (so-HOTT) model of GBM to decipher how glutamatergic signaling impacts GBM lineage trajectories. so-HOTT preserves patient tumor heterogeneity, features excitatory NTS, and enables clonal lineage tracing of tumor cells after NTS perturbations. Genetic and pharmacological inhibition of AMPA and kainate receptors in so-HOTT shifts tumor cell composition from neuronal fates toward progenitor-proximal astrocytic/mesenchymal states. This occurs through the attenuation of calcium signaling and reduced plasticity of malignant radial glia (RG)-like progenitors, a previously unrecognized target of NTS. Through the integration of inputs from the neuronal microenvironment into glutamatergic signaling, progenitor populations modulate their transcriptional programs and cell fate, ultimately shaping GBM tumor heterogeneity. Targeting synaptic input may thus constrain the heterogeneity that fuels GBM adaptation and therapeutic escape.

cancer biology↗

bioRxiv2026

Landscape-scale navigation unlocks antibody CDR structural logic for AI-guided rescue and therapeutic optimization

While AI offers transformative potential for therapeutic antibody design, the lack of ground-truth data fundamentally constrains our ability to model the epistatic topology of fitness landscapes. Here, we establish a high-throughput workflow to characterize tens of thousands of antibody variants per week with gold-standard biophysical precision. By combinatorially assembling functional variants from deep mutational scanning, we charted antibody fitness landscapes comprising over 17,000 data points, which revealed an extremely rugged, non-navigable epistatic topology. Yet, navigating at this unprecedented scale enabled the discovery of rare peak clusters exhibiting simultaneous enhancements in affinity and productivity. Strikingly, ProteinMPNN predicted the CDR-dependent productivity landscape with remarkable accuracy, suggesting that sequence-structure compatibility within CDRs gates cellular productivity. This insight enabled a structure-guided rescue strategy combining AlphaFold3 and ProteinMPNN, which successfully restored the cellular productivity of high-affinity, low-productivity clones via single amino acid substitutions. Two elite variants drawn directly from peak clusters further demonstrated 20- to 100-fold in vivo efficacy gains in a murine psoriasis model. Our findings establish CDR structural fitness as a fundamental determinant of antibody cellular productivity and validate landscape-scale navigation as a powerful framework for therapeutic antibody optimization.

bioengineering↗

bioRxiv2026

Systematic mapping of insertion-tolerant regions enables capsid engineering of the RNA phage PP7

RNA phages are attractive platforms for the design of programmable bioparticles, but their development has been constrained by limited knowledge of genomic sites that can tolerate sequence insertion. Here, we combined MuA transposase-mediated in vitro insertion mutagenesis with our established reverse genetics systems to systematically identify insertion-tolerant regions (ITRs) in the RNA phages MS2 and PP7. Screening of 4,555 MS2 and 2,228 PP7 random insertion clones identified 29 and 26 nonredundant ITRs, respectively. We further analyzed and compared these ITRs in the context of RNA genome organization and virion architecture. Both phages contained ITRs within the maturation protein (MP), whereas only PP7 tolerated insertions within the coat protein (CP). On the basis of structural location and plaque-forming capacity, an ITR situated between Gly74 and Glu75 (GGC^GAG) in the PP7 CP was selected for further study. Infectious phage particles generated from cDNA clones retained the 15-bp insertion at both the RNA and protein levels. Engineered PP7 phages carrying an RGD motif inserted into the CP at this ITR displayed enhanced in vivo clearance in a Drosophila model, despite having in vitro stability comparable to that of the wild type. These findings provide the first example of CP engineering in an RNA phage and establish a framework for engineering RNA phages for biological and biotechnological applications. IMPORTANCEA major obstacle to developing RNA phages as synthetic biology platforms is the lack of design principles for genomic insertion. Here, we address this limitation by establishing a mutagenesis-and-recovery workflow that systematically identifies insertion-tolerant regions (ITRs) in the RNA phages MS2 and PP7. The resulting maps reveal distinct structural constraints in the two phages and enable rational engineering of a peptide-display site in the PP7 capsid. Using this approach, we generated an engineered infectious phage with a modified capsid, thereby providing the first demonstration of capsid engineering in an RNA phage. This study lays the groundwork for the rational design of RNA phage virions as tractable and engineerable scaffolds for future biological and biotechnological applications.

microbiology↗

bioRxiv2026

Transcription Factor Subtype Governs Response and Resistance to DLL3-Directed T-Cell Engagement in Small Cell Lung Cancer

Although small cell lung cancer (SCLC) comprises transcription factor (TF)-defined molecular subtypes (ASCL1, NEUROD1, POU2F3), the extent to which these subtypes predict response to clinically effective therapy in patients--and whether therapy can select for subtype switching--remains unknown. The recent approval of the DLL3xCD3 bispecific T-cell engager tarlatamab represents one of the first meaningful advances in relapsed small cell lung cancer (SCLC) in decades, yet responses remain heterogeneous and resistance is inevitable. Here, we inferred SCLC gene expression from circulating chromatin in prospectively collected patient plasma (46 patients; 167 samples), enabling interrogation of response and acquired resistance to tarlatamab. Parallel development of the first immunocompetent syngeneic mouse model to study tarlatamab response and resistance enabled functional validation. Across species, findings converged on a central principle: TF subtype governs both initial response and acquired resistance. Therapeutic response was significantly associated with ASCL1-subtype tumors, whereas NEUROD1-subtype tumors exhibited inferior responses and POU2F3-subtype tumors were uniformly resistant, consistent with DLL3 being a direct ASCL1 transcriptional target and most highly expressed in ASCL1-positive tumors. Strikingly, one mode of acquired resistance revealed therapeutic selection for a NEUROD1-high state with concomitant DLL3 downregulation. Other resistant tumors exhibited enrichment of regulatory and exhausted T-cell programs, highlighting tarlatamabs dual-targeting mechanism of action. Together, these results reveal that tarlatamab exerts selective pressure against ASCL1-driven lineages, facilitating resistance through loss of an antigen intrinsically linked to that state. These findings underscore the clinical relevance of TF-defined molecular subtypes in human SCLC. More broadly, they highlight the power of integrating longitudinal in vivo plasma transcriptional profiling from patient plasma with functional mouse modeling to uncover clinical and biological mechanisms of response and resistance to cell-surface-targeted therapies.

cancer biology↗

bioRxiv2026

Computational aberration-corrected volumetric imaging of single retinal cells in the living eye

The eye offers a unique non-invasive window for accessing single-cell level structures and functions of the central nervous system (CNS) throughout the retina. However, strong and space-varying ocular aberrations, along with limited volume rates, challenge large-scale cellular imaging in living eyes and stymie the full potential of possible biological and pathological studies in retina. Here, we present plenoptic illumination scanning laser ophthalmoscopy (PI-SLO), a 3D fluorescent retinal imaging modality that enables high-speed, widefield, volumetric single-cell imaging with low phototoxicity. By capturing multiple angular images of fluorescence signals from the entire volume, PI-SLO enables digital aberration correction and 3D imaging across a >20{o} FOV with >23 Hz volume rate. We leverage this structural and functional imaging modality to investigate three key aspects of CNS physiology through the living mouse retina, including: microglial process dynamics, vascular perfusion, and light evoked calcium fluxes in inner retinal neurons. PI-SLO is a versatile non-invasive platform for in vivo investigation of retinal and CNS physiology at the cellular level.

bioengineering↗

bioRxiv2026

Accurate somatic variant detection from formalin fixed, paraffin embedded tissue (FFPE) derived WES and WGS by DeepOmicsFFPE-PLUS, a sequence context-based transformer

Formalin-fixed paraffin-embedded (FFPE) tissues represent a vast archival resource for genomic studies, yet their utility remains constrained by fixation-induced DNA damage and subsequent sequencing artifacts. To comprehensively characterize and address this challenge, we analyzed matched FFPE and fresh-frozen tumor samples from two institutions, spanning different storage durations, DNA qualities, sequencing platforms (WES and WGS), exome capture kits, and somatic variant callers. We found that FFPE-induced artifacts exhibit strong batch- and age-specific patterns, with a predominance of C:G>T:A substitutions, which particularly complicate the accurate identification of low allele frequency true variants. Enzymatic repair methods partially alleviated these artifacts but remained insufficient. To overcome these limitations, we developed DeepOmicsFFPE-PLUS(https://github.com/Theragen-Bio/DeepOmicsFFPE-PLUS), an advanced AI-based tool to accurately distinguish true somatic variants from FFPE-specific artifacts. DeepOmicsFFPE-PLUS demonstrated consistently superior performance across diverse conditions, achieving high sensitivity and specificity--even for low-frequency variants--outperforming existing tools. Application of our model to WGS data further enabled recovery of biologically relevant mutational signatures, including restoration of microsatellite instability (MSI)-associated signatures initially obscured by FFPE artifacts. Our findings underscore the necessity of artifact-aware variant calling in FFPE genomics and establish DeepOmicsFFPE-PLUS as a robust tool for artifact removal, enabling high-fidelity downstream analyses and personalized therapeutic target discovery.

cancer biology↗

bioRxiv2026

Model-Driven Hybrid AI Framework for End-to-End Autonomous Decision-Making in Drug Development

Decision-making in drug development spans heterogeneous stages from molecular design to clinical optimization, yet computer-aided workflows across stages remain fragmented, limiting traceability from evidence-derived clinical questions to simulation scenarios and decision endpoints. We present a model-driven hybrid AI framework for end-to-end decision support that treats PICO (Participants, Intervention, Comparison, Outcome) as a machine-actionable specification to define scenarios and align endpoints across modules. Given a PICO-defined clinical question and a drug SMILES, the framework infers PBPK-ready parameters from structure, generates exposure via mechanistic PBPK simulation, and links exposure outputs to clinic-facing therapeutic drug monitoring (TDM) decision support using an open-source NLME-based Clinical Pharmacokinetic Consultant Service (CPCS). We additionally implement an LLM-based PICO extraction layer to structure clinical abstracts into P/I/C/O elements for systematic evidence ingestion. Results include backbone verification for DTI-guided intrinsic clearance prediction under fold-error bounds and PBPK trajectory plausibility, and external validation of the hybrid AI-PBPK pipeline showing substantial dispersion for several endpoints. An input-controlled comparison between PhysioSim and PK-Sim on N = 9 drugs yields similarly degraded accuracy under identical AI-predicted inputs, suggesting upstream parameter quality as a dominant constraint. For evidence ingestion, prompting-only PICO extraction achieves F1 of 0.687 on EBM-NLP and 0.552 on TB-PICO. For clinical utility, CPCS improves TDM performance over a PKS baseline on phenobarbital and vancomycin, reducing MAPE by approximately 26-60% depending on configuration. Overall, the framework provides a modular, traceable blueprint linking evidence-defined questions to exposure simulation and TDM decision support while preserving mechanistic interpretability. Future work will integrate outcome-aligned time-to-event modeling and federated learning to close the PICO-to-outcome loop and improve parameter inference under multi-site governance.

pharmacology and toxicology↗