Biology subjects

Jung, S.

Publications and source records attributed to Jung, S..

At least 19 recordsLinked to original sources

bioRxiv2026

Benchmarked sequencing methods reveal hidden mosaic variation in the human brain

Mosaic variants (MVs) record cellular lineages, but their recovery is strongly method-dependent. Here, we benchmarked four duplex sequencing methods alongside bulk and single-neuron genome sequencing in matched human brain tissue. The methods differed in genome representation, MV yield and mutational profiles, with limited caller-level concordance. However, read-level interrogation revealed substantially greater cross-method molecular support. To systematically recover this evidence, we developed Read-level Integration for Variant Evidence Recovery (RIVER), expanding the pool of cross-method-supported candidates by approximately two orders of magnitude relative to caller-level overlap. Orthogonal sequencing at ~10,000x across brain and peripheral tissues validated approximately half of the tested RIVER-rescued candidates. MVs recovered from a single brain region resolved broadly distributed and regionally restricted clones and provided lineage information among individual neurons. These findings show that RIVER substantially expands the pool of endogenous lineage markers available to investigate human development, aging, disease-associated mosaicism and clonal evolution.

genomics↗

bioRxiv2026

Torque-optimized body morphing enables rapid surface to underwater diving in whirligig beetles

Known for their rapid water surface locomotion, whirligig beetles (Gyrinidae) are also fast divers. They can transition from the surface to underwater swimming in under 100 ms. Diving requires them to generate sufficient hydrodynamic torque to break the water surface and reorient the body. However, the mechanism by which whirligigs create the necessary pitch-down torque is unknown. Here, we show that their diving is achieved by bending the abdomen. The inertial hydrodynamic pressure on the flexed abdomen generates the torque necessary for surface breaking and body rotation. The torque is maximized at an abdominal angle of 45{degrees}, whereas for effective diving, whirligigs display a preferential abdominal angle of around 28{degrees}, closely aligning with the angle that maximizes the torque-to-drag ratio. This abdominal angle is optimized by balancing torque generation with the drag penalty. Our findings inspire a mechanism for surface-to-underwater transition, offering design principles for multimodal robots capable of locomotion across the air-water interface.

biophysics↗

bioRxiv2026

Dysregulation of FMR1 Splicing in Human Fragile X Syndrome

Fragile X Syndrome (FXS) is a neuro-developmental disorder caused by a CGG expansion in FMR1, leading to transcriptional silencing and loss of the encoded protein FMRP. Surprisingly, ~70% of FXS individuals express FMR1, but the RNA is mis-spliced to isoform FMR1-217, composed of exon 1 spliced to a pseudo-exon in intron 1 and cannot produce FMRP. Splice-switching ASOs rescue proper FMR1 splicing and restore FMRP. FMR1-217 mis-splicing increases with CGG repeat length and is negatively correlated with patient IQ. FMR1-217 is associated with ribosome footprints, indicating it is translated into a polypeptide that may impair cognition. R-loops form at the FMR1 locus and extend into the pseudo-exon, but splice-switching ASOs reduce FMR1-217 and elevate FMRP independently of R-loop formation. DRB-based transcription analysis identified impaired Pol II elongation at the 5 prime region of FMR1 in FXS cells, indicated by accumulation of hypophosphorylated Pol II at the transcription start site. Consistent with this, camptothecin-induced Pol II stalling increased FMR1-217 pseudo-exon inclusion. The splicing factors PTBP1 and PTBP2 regulate FMR1-217 splicing in a differentiation stage-dependent manner. Together, these findings indicate that FMR1-217 mis-splicing in FXS is associated with CGG repeat expansion, R-loop formation, impaired co-transcriptional Pol II elongation and context-dependent regulation by PTBP1/PTBP2.

molecular biology↗

bioRxiv2026

Evaluation of the Translational Readiness of Pluripotent Stem Cell-Derived Vascular Cell Therapy for Limb Ischemia: A Systematic Review and Meta-analysis

Background: For the first time, pluripotent stem cell (PSC)-derived endothelial cells were administered to a human for peripheral artery disease (PAD), in August 2026. This translation is based on a large preclinical literature claiming that donor cells build blood vessels in animal models of limb ischemia, but whether it supports that claim has not been assessed systematically. Methods: In this systematic review and meta-analysis, we searched MEDLINE, Web of Science, Europe PMC, and Embase to August 2026 for studies of PSC-derived vascular cells in animal models of limb ischemia with an acellular control. Limb perfusion at each experiment's primary timepoint was pooled as Hedges' g in a three-level random-effects model with cluster-robust variance. Donor-cell fate and a seven-level ordinal of vascular contribution were graded for every study. This meta-analysis was registered on PROSPERO (CRD420261465308). Results: Sixty-eight studies in 69 reports met the criteria; 51 contributed 114 comparisons at primary timepoints. Cell therapy increased perfusion (Hedges' g 2.26, 95% CI 1.72-2.79; P=4.1x10-11; I2=87.5%; 95% prediction interval -1.47 to 5.98). No prespecified moderator survived Holm-Bonferroni adjustment, including the donor-cell fate grade (P=0.30). A donor cell was resolved in a vessel wall in 35 studies, and flow through such a vessel in 7. Small-study effects were strong (Egger t=6.25, P=9x10-8) but were not reproduced by a sample-size-based test (P=0.80); trim-and-fill moved the estimate to 1.15. On the natural scale, treated limbs recovered about twice the control perfusion (ratio of means 1.96, 95% CI 1.66-2.33). Limb preservation favored cell therapy in 11 studies with countable events (risk ratio 4.88, 95% CI 2.28-10.45), but 43 of 68 studies reported no functional outcome, and only 22 comparisons from 8 studies were both randomized and blindly assessed. In a post hoc analysis that did not survive multiplicity adjustment, effects were largest in studies whose images could not resolve their incorporation claims (g 3.23) versus those that resolved incorporation (g 2.33). Conclusions: PSC-derived vascular cell therapy improved limb perfusion in animal models, consistently across every sensitivity analysis, although the magnitude is uncertain given heterogeneity and small-study effects. Benefit was not associated with the degree of donor-cell incorporation across studies and was largest where incorporation was least resolvable, a pattern more consistent with reporting and measurement bias than with a true mechanistic effect. The first rigorous, blinded test of that benefit may come not from animals but from the patients now receiving these cells.

cell biology↗

bioRxiv2026

Phenotype-associated spatial biomarker discovery in spatial transcriptomics with spHOT

Spatial transcriptomics now profiles patient cohorts at single-cell resolution, enabling analysis of disease-associated cell organization in situ. However, discovering such spatial biomarkers remains challenging because relevant structures occur at unknown scales and cell-or niche-level annotations are rarely available. We present spHOT, a deep learning framework that localizes phenotype-associated spatial biomarkers from sample-level labels. spHOT combines spatial foundation model embeddings, a hierarchical domain tree for multi-resolution tissue representation, and a teacher-student multiple instance learning architecture that converts sample labels into cell-level biomarker scores. In controlled simulations and real-tissue benchmarks, spHOT outperformed existing spatial and single-cell methods in localizing ground-truth biomarkers. Across fibrotic, metabolic, and autoimmune disease datasets, spHOT recovered disease-relevant niches and tissue states reported by supervised analyses in the original studies. Cross-disease application of spHOT transferred biomarkers across chronic lung diseases without retraining. spHOT enables scalable, annotation-efficient spatial biomarker discovery in cohort-scale spatial transcriptomics.

bioinformatics↗

bioRxiv2026

Spatial Metabolomics by Desthiobiotin Ligase (DESTNI) in Live Cells

Proximity labeling has transformed spatial proteomics by enabling compartment-resolved mapping of protein environments in living cells, yet its extension to small-molecule metabolites has not been demonstrated, probably due to limitations in labeling chemistry and labeled metabolites identification. Here, we introduce DESTNI, an engineered desthiobiotin (DTB) ligase derived from TurboID through directed evolution, and establish a platform for spatially resolved profiling of amine-containing metabolites. A directed evolution strategy based on a yeast display system yielded DESTNI with efficient DTB-dependent reactivity, enabling robust and compartment-specific proximity labeling across diverse subcellular environments. To identify the DTB-modified amino metabolome, we developed an integrated analytical framework combining DTB-modified amino metabolite standards, in vitro DESTNI profiling, and in silico MS/MS prediction, enabling systematic annotation of DTB-modified amino metabolites. To extend this chemistry to metabolites, we combined synthetic DTB-conjugated metabolite reference standards, in vitro DESTNI-reactive metabolite discovery, and machine-learning prediction of DTB-derivatized metabolites and oligopeptides. Organelle-targeted DESTNI recovered reproducible compartment-enriched amino metabolite signatures, including mitochondrial matrix-enriched glycine, 5-aminolevulinic acid, ornithine and spermidine adducts, as well as nuclear-enriched {gamma}-aminobutyric acid and 5-aminovaleric acid adducts. Together, this work establishes DESTNI as a proximity labeling platform that bridges spatial proteomics and metabolomics and provides a general strategy for mapping subcellular biochemical environments in living cells.

biochemistry↗

bioRxiv2026

Structural Basis for Inhibition of the HKU5-CoV Main Protease by Clinical SARS-CoV-2 Protease Inhibitors

The identification of Pipistrellus bat coronavirus HKU5 lineage 2 (HKU5-CoV-2) as a human ACE2-adapted virus highlights the need for antiviral strategies to control emerging HKU5-related merbecoviruses. However, despite the importance of the main protease (Mpro) as a key antiviral target, structural and biochemical characterization of HKU5-CoV Mpro in the context of clinical inhibitors has remained limited. Here, we obtained high-resolution crystal structures of HKU5-CoV-1 Mpro in its apo state (1.75 [A]) and in complex with the approved covalent inhibitor nirmatrelvir (1.91 [A]) and the non-covalent inhibitor ensitrelvir (1.55 [A]). These structures reveal an induced-fit mechanism, in which the S2 loop containing Met49 adopts an open conformation in the apo state and undergoes inhibitor-specific conformational changes upon nirmatrelvir and ensitrelvir binding. These structures served as a foundation for the characterization of HKU5-CoV-2 Mpro via modeling and molecular dynamics simulations. Biochemical assays revealed that HKU5-CoV-1 and HKU5-CoV-2 Mpro exhibited nearly identical kinetic profiles, with turnover rates approximately two-fold higher than SARS-CoV-2 Mpro. Structural analysis revealed a highly conserved S1 subsite but distinct local environments in the S1', S2, and S4 substrate-binding sites relevant to inhibitor recognition. Despite these variations, nirmatrelvir and ensitrelvir showed potent inhibitory activity, with comparable double-digit nanomolar IC50 values across all three Mpro proteins. Computational analyses indicated that HKU5-CoV-2 Mpro engages nirmatrelvir and ensitrelvir in binding modes comparable to those observed for HKU5-CoV-1, supporting a conserved mechanism of inhibitor recognition and providing a structural basis for developing pan-coronavirus antivirals targeting emerging merbecoviruses.

biochemistry↗

bioRxiv2026

FBXL21 regulates diurnal proteostasis and stress response by targeting DNAJB6 and client proteins

Circadian regulation of proteostasis, a key determinant of muscle health, remains poorly understood. Here, we identified DNAJB6, an Hsp40 (DnaJ) co-chaperone, as a substrate of the circadian E3 ligase FBXL21. FBXL21 mediated the ubiquitination-dependent proteasomal degradation of both DNAJB6 and its client proteins including Desmin; causative mutations of DNAJB6 in myopathies, however, rendered resistance to FBXL21-directed degradation. Fbxl21 KO C2C12 cells displayed aberrant accumulation of Desmin, and showed aggravated cytoplasmic accumulation of TDP-43, another DNAJB6 client protein, in heat shock response. Under timed exercise as a physiological stressor, WT mice displayed robust diurnal rhythms in the levels of stress granule markers (G3BP1 and FUS) and TDP-43 as a function of exercise timing. In contrast, the Fbxl21 hypomorph Psttm mutant mice showed elevated expression of these proteins without exercise, which was exacerbated under exercise-induced stress conditions; importantly, these abnormalities were rescued by skeletal muscle-specific FBXL21 expression. Our study elucidates a novel diurnal regulatory mechanism of skeletal muscle proteostasis via FBXL21 as a chaperone-linked E3 ligase, highlighting the FBXL21-DNAJB6 axis as a potential therapeutic target for myopathies.

cell biology↗

bioRxiv2026

Secreted phospholipase A2α generates a pathogen-derived lysophospholipid to signal local immunity

In plants, effector-triggered immunity (ETI) is initiated when NLR receptors recognize pathogen effectors, yet the molecular signals linking this recognition to downstream defense activation remain poorly defined, in contrast to the well-characterized immunogenic signals of pattern-triggered immunity. Here, we show that a lysophospholipid is a previously unidentified ETI-mediating immune signal generated through host enzymatic conversion of pathogen membrane lipids. Upon recognition of avirulent Pseudomonas syringae, secretory phospholipase A2 (PLA2) is rapidly induced and secreted into the apoplast, where it hydrolyzes bacterial phosphatidylethanolamine to produce lysophosphatidylethanolamines (LPEs), predominantly LPE18:1. Genetic ablation of PLA2 compromises local immunity, hypersensitive response confinement, and defense gene activation, all of which are largely rescued by exogenous LPE18:1. Mechanistically, LPE18:1 promotes ICS1-dependent salicylic acid biosynthesis and NPR1-mediated transcriptional reprogramming. Together, these findings support pathogen-derived LPE18:1 as a newly identified lipid-based immune signal that links NLR activation to spatially confined defense responses. The evolutionary conservation of secretory PLA2 enzymes and lysophospholipid signaling from plants to mammals suggests that host-directed enzymatic remodeling of pathogen membranes into immune-activating lipid signals may constitute a conserved strategy of innate immunity.

plant biology↗

bioRxiv2026

Fungal Hyphae as Distributed Vapor Sinks

Hygroscopic surfaces act as local vapor sinks that reshape the condensation field around them, but whether distributed biological structures do the same has not been investigated. We have established that hyphae of fungal colonies functionally behave as vapor sinks, creating a dry region of width{delta} around themselves when placed on a cooled substrate. In addition, the radial distribution of droplet sizes steepens during condensation, and the rate at which droplets evaporate locally after chamber drying increases. In order to quantify this behavior, we employed a combination of time-resolved imaging and survival analysis to determine how long individual droplets persist on the surface surrounding the colony. These data were used to derive three quantitative measures of the vapor-sink effect. Each measure was found to be directly proportional to the vapor-sink strength of the substrate, as calibrated against NaCl-agar hydrogels of known water activity (LOOCV RMSE = 0.031 for recovered aw). These findings were consistent across three fungal genera (35 experiments), and all species fell along calibration lines defined by the hydrogel standards. This result is consistent with a diffusion-limited vapor-depletion framework. The measured genus-level{delta} ratios agreed to within 6% of predictions from structural absorbing capacity, and field measurements on Gymnosporangium-infected apple leaves were consistent with the same signatures under natural conditions. These results establish a non-contact method for inferring the material properties of thin hygroscopic biological surfaces from their condensation patterns.

biophysics↗

bioRxiv2026

Nanoscale Organization of Membrane Tension during Neutrophil Extracellular Trap Formation Revealed by Fluorescence Lifetime Imaging

Cells continuously generate and respond to mechanical forces across compartments, with the plasma membrane acting as a nanoscale interface for sensing and transmitting tension. How intracellular forces are translated into membrane tension during dynamic processes such as neutrophil extracellular trap (NET) formation remains unclear. Here, we combine the mechanosensitive fluorescent probe Flipper-TR with fluorescence lifetime imaging microscopy (FLIM) to map spatiotemporal changes in plasma membrane tension in living cells. After validation in HeLa and dHL-60 cells under osmotic perturbation, we apply this approach to primary human neutrophils undergoing NETosis. Membrane tension transiently increases during chromatin decondensation and nuclear swelling within 90 min and is followed by a marked decrease 30 min after membrane rupture. Prior to rupture, tension is spatially heterogeneous, indicating localized nanoscale mechanical regulation. Cholesterol depletion abolishes the transient increase and reduces heterogeneity without affecting overall NETosis kinetics. Together, these findings establish the plasma membrane as a dynamic nanoscale reporter of intracellular mechanical stress during NETosis.

immunology↗

bioRxiv2026

Temperature-induced shift in the rare microbiota of the sponge Haliclona

Ocean warming is currently leading to distributional shifts of species and an alternation of coastal communities. Vulnerable species that are most sensitive to ocean warming are able to use several acclimation mechanisms, with one of the fastest being a shift in and shuffling of their partnerships with symbiotic microorganisms. Assessing symbiosis-focused mechanisms of acclimation and adaptation in response to ocean warming is a technical challenge due to the difficulty of accurately simulating the de novo formation of coastal communities. Here, we use the Kiel Outdoor Benthocosm facility to assess which sponges species are experimentally recruited and whether they exhibit symbiosis-focused mechanisms of acclimation following selection to ocean warming. We observed one sponge species (Haliclona sp.) and found that this sponge exhibited significant shifts in the membership and composition of its associated microbiome in response to ocean warming, with much of this being attributed to the rare microbiota. Moreover, Haliclona sp. maintained the diversity and dominance of its microbiome members. Four bacteria taxa were differentially abundant at elevated temperatures, with two being a Francisella sp. that is a suspected pathogen and an uncultured Francisellaceae that is most closely related to sulfur-oxidizing endosymbionts. Changes to the Haliclona sp. microbiome are largely consistent with a limited acclimation response, which could indicate that this sponge may use microbial symbionts as part of a mechanism to acclimate and adapt to a warmer future ocean.

ecology↗

bioRxiv2026

Wearable Dual-Modality Plethysmography for Arterial Modulation and Blood Pressure Dip

Continuous, non-invasive cardiovascular monitoring is limited by the superficial sensing depth of Photoplethysmography (PPG), which is susceptible to peripheral artifacts. This study evaluates a wearable dual-modality prototype integrating dryelectrode Impedance Plethysmography (IPG) and PPG within a smartwatch form factor. Results from a pilot study (N=2) demonstrate that IPG signals exhibit a temporal lead over PPG across ventral and dorsal sites, supporting its greater penetration depth. During brachial artery modulation, IPG showed superior sensitivity to arterial recovery on the ventral forearm. Furthermore, 60-minute napping sessions revealed that while PPG remained morphologically stable, IPG signals underwent significant evolution, capturing distinct pulsewave archetypes. These findings suggest that wearable IPG provides a high-fidelity window into deep systemic hemodynamics typically reserved for clinical instrumentation.

physiology↗

bioRxiv2026

The long noncoding RNA Dory is required for female but not male spatial learning and memory

Large numbers of long noncoding RNAs (lncRNAs) exhibit region- or cell-specific expression and subcellular locations in the mammalian brain. We analyzed the expression and function of the mouse lncRNA 2700046G09Rik (also called Sgms1os1), which we have named Dory due to sex-specific disruption of spatial memory in rodent gene knockout and RNA knockdown. We show that Dory is predominantly expressed in a punctate pattern in nuclei of excitatory neurons in the hippocampus, and in both neuronal and non-neuronal cells in the cerebellum. We show by high resolution RNA sequencing that Dory brain transcripts are composed of 1-3 exons with multiple isoforms. Deletion of the constitutive first exon of Dory by CRISPR-Cas9 genome editing resulted in the impairment of spatial memory and some aspects of balance in female but not male mice. Dory-null mice presented without overt changes in brain morphology. Knockdown of the rat homolog of Dory in dorsal hippocampus using antisense oligonucleotides confirmed inhibition of spatial memory in females only. RNA sequencing and mass spectrometry revealed differential hippocampal gene and protein expression profiles, notably of prolactin, growth hormone and pro-opiomelanocortin, between male and female Dory knockout mice.

neuroscience↗

bioRxiv2026

Circulating miR-10b-5p drives Kawasaki vasculitis through endothelial reprogramming and nominates CXCL8 as an early diagnostic biomarker

BackgroundKawasaki disease (KD) is an acute systemic vasculitis of unknown etiology and the leading cause of acquired heart disease in children. Coronary artery aneurysm formation represents its most serious complication, yet no specific biomarker exists for early diagnosis at emergency department (ED) presentation. We sought to define the molecular mechanisms driving KD vasculitis and to identify a clinically actionable early diagnostic marker. MethodsWhole microRNA sequencing was performed on buffy coat specimens from KD patients and febrile controls presenting to the ED. Disease-associated miRNA candidates were validated in two independent murine KD-like vasculitis models: Lactobacillus casei cell wall extract (LCWE; n=12-20) and Candida albicans water-soluble fraction (CAWS; n=12). The functional role of miR-10b-5p was assessed by antagomir-mediated inhibition in the LCWE model. Transcriptomic and chromatin-level reprogramming were characterized by single-cell RNA sequencing (scRNA-seq), bulk RNA-seq, and assay for transposase-accessible chromatin with sequencing (ATAC-seq) in human coronary artery endothelial cells (HCAECs). Transcription factor binding was validated by in vitro binding assays and chromatin immunoprecipitation. Serum chemokine (C-X-C motif) ligand 8 (CXCL8) was quantified by enzyme-linked immunosorbent assay. ResultsForty miRNAs were upregulated in KD patients and fourteen miRNAs were prioritized for further analysis. miR-10b-5p was consistently elevated across both vasculitis models, and its antagomir-mediated inhibition attenuated aortic root inflammation in vivo. In HCAECs, miR-10b-5p suppressed its direct targets marker of proliferation Ki-67 (MKI67)/Chromobox5 (CBX5), driving a cell-state transition from proliferative/metabolic programs toward a pro-inflammatory phenotype. Consistently, analyses of bulk RNA-seq supported cell-cycle arrest and altered chromatin remodeling. ATAC-seq revealed increased chromatin accessibility at the CXCL8 and Matrix metalloproteinases 10 (MMP10) promoters, and motif analysis identified CCAAT/enhancer-binding protein alpha (CEBPA) as the key transcriptional activator; CEBPA knockdown abrogated miR-10b-5p-induced CXCL8 and MMP10 upregulation. Serum CXCL8 was markedly elevated in KD patients relative to febrile controls at first ED presentation, prior to definitive diagnosis, and declined significantly at 8-week follow-up. ConclusionsWe define a miR-10b-5p-MKI67/CBX5-CEBPA-CXCL8 axis as a mechanistic driver of endothelial reprogramming and neutrophil-recruiting inflammation in KD. Serum CXCL8 emerges as a candidate early diagnostic biomarker that may facilitate timely KD recognition at ED presentation.

molecular biology↗

bioRxiv2026

Benzoxaboroles are structurally unique binders of eukaryotic translation initiation factor 4E

Benzoxaboroles offer unusual reactivity and protein recognition for the development of small molecule drugs. Despite this potential, they are uncommon in drug discovery or in fragment screening libraries. We synthesized a series of structurally related benzoxaboroles containing a diazirine/alkyne tag to enable in-cell photoaffinity labeling experiments. A subset of this library was found to have high selectivity for eukaryotic translation initiation factor 4E (eIF4E). The benzoxaborole-eIF4E interaction was found to be stereoselective in nature and competitive with the 7-methylguanosine cap of mRNA. Site of labeling experiments revealed that the benzoxaborole fragment interacts with the cap binding pocket of eIF4E. In silico modeling of the modified protein suggests that H-bonding interactions between the main chain of Trp102 and the side chain of Asn155 to the amide carbonyl and anionic boronate of the benzoxaborole, respectively, drive affinity for this challenging to drug pocket.

cancer biology↗

bioRxiv2026

Redundant roles of YES and SRC tyrosine kinases in driving malignant peripheral nerve sheath tumors

Malignant peripheral nerve sheath tumors (MPNST) are highly aggressive soft tissue sarcomas that are largely incurable with no clinically effective systemic therapies or immunotherapies for advanced disease. Here, we identify the SRC-family kinases (SFKs) YES and SRC as redundant, essential drivers of MPNST growth. Dual inhibition of YES/SRC activity by genetic silencing or pharmacological SFK inhibitors markedly suppressed the proliferation of multiple NF1-mutant MPNST cell lines. In vivo, conditional genetic depletion of YES/SRC in MPNST cells abrogated tumor growth in subcutaneous and orthotopic models, and dasatinib treatment delayed tumor progression and improved overall survival. Integrated transcriptomic and phosphotyrosine proteomic analyses revealed that YES/SRC inactivation extensively rewires MPNST signaling, coordinately repressing multiple oncogenic signaling pathways and downstream cell cycle transcriptional programs. Unexpectedly, YES/SRC inhibition also upregulated interferon and antigen processing and presentation pathways and increased cell-surface MHC class I expression, consistent with tumor-intrinsic immune reactivation. Clinically, analysis of a large sarcoma cohort demonstrated that YES1 is significantly overexpressed in MPNST compared to benign soft tissue tumors. Collectively, our findings establish YES/SRC as non-oncogene vulnerabilities in MPNST.

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↗