Biology subjects

Xie, L.

Publications and source records attributed to Xie, L..

At least 19 recordsLinked to original sources

bioRxiv2026

Harnessing the PRDM1-PGC1A axis to enhance CAR T cell efficacy

Despite its promise in treating B cell malignancies, chimeric antigen receptor (CAR) T cell therapy remains limited by resistance mechanisms that impair CAR T cell stemness, persistence, and effector function. Here, we identified PRDM1, which encodes the transcriptional repressor, BLIMP-1, as a crucial regulator of CAR T cell terminal differentiation and exhaustion. PRDM1 was highly enriched in exhausted-like tumor-infiltrating CAR T cells from lymphoma patients and in pre-infusion products from clinical non-responders. These findings link increased PRDM1 expression to CAR T cell dysfunction and poor clinical outcomes. Genetic deletion of PRDM1 significantly promoted CAR T cell proliferative capacity, memory differentiation and antitumor efficacy in both hematologic and solid tumor models. Mechanistically, PRDM1 deletion increased the activity of PGC1a;, a transcriptional coactivator central to mitochondrial biogenesis and antioxidant capacity, thereby improving mitochondrial function, oxidative stress resistance, and metabolic reserve. Depletion of PGC1a; attenuated the improved efficacy of PRDM1-deficient CAR T cells, indicating an essential role for this pathway downstream of PRDM1 loss. Pharmacologic activation of the PPAR-PGC1a; program with a clinically approved agonist, bezafibrate, enhanced CAR T cell antitumor activity without direct tumoricidal effects. Notably, transient exposure to bezafibrate during CAR T cell manufacturing was sufficient to confer durable treatment benefits. Together, our study defines a PRDM1-PGC1a; axis that controls CAR T cell differentiation, mitochondrial fitness and therapeutic response, which provides a mechanistic and translational framework for advancing CAR T cell therapy across distinct tumor contexts.

immunology↗

bioRxiv2026

TRIP6 condensates constrain Hippo pathway hyperactivation under osmotic stress

The Hippo pathway integrates mechanical and chemical cues to regulate cell proliferation and tissue homeostasis, whereas its dysregulation contributes to impaired regeneration and tumorigenesis. Liquid-liquid phase separation has emerged as an organizing principle for Hippo signal transduction, but how cells constrain excessive pathway activation under stress remain poorly defined. Here we demonstrate that TRIP6, a LIM-domain scaffold and tension-sensitive Hippo inhibitor, undergoes intrinsic phase separation both in vitro and in living cells. Osmotic stress triggers rapid redistribution of TRIP6 condensates from large cortical assemblies to dispersed cytoplasmic puncta. Domain mapping shows that the N-terminal IDR2 drives TRIP6 phase separation, whereas the C-terminal LIM domains negatively regulate condensate fluidity. Mechanistically, TRIP6 condensates selectively recruit the LATS1 inhibitor AJUBA and the MST1 scaffold SAV1 while excluding MST1 itself. The phase-separation-deficient {Delta}IDR2 mutant fails to undergo stress-induced condensate redistribution, exhibits weakened LATS1 binding, and shows increased YAP phosphorylation and cytoplasmic retention under osmotic stress, with reduced cell proliferation and stress resilience. Finally, TRIP6 condensates disassemble in a mouse model of elevated epidermal tissue tension, identifying them as mechanosensitive platforms that link phase transitions to Hippo-dependent growth control. Together, these findings establish TRIP6 condensates as a negative regulatory node that buffers excessive Hippo activation to preserve cellular homeostasis under hyperosmotic conditions.

cell biology↗

bioRxiv2026

Hybrid Mechanistic-Neural Modeling of Concentration-Time Dynamics Generalizes Human Pharmacokinetics Prediction Across Unseen Chemical Space

The pharmacokinetics concentration-time profile encodes ADME dynamics. Its optimization in drug discovery curbs late-stage attrition, yet forecasting human pharmacokinetics from chemical structure remains difficult. Machine learning models ignore continuous ADME dynamics, whereas physiologically based pharmacokinetics (PBPK) models demand unavailable parameters and impose rigid, misspecified compartmental structures. We present PK-MUSE (PharmacoKinetic hybrid Mechanistic Universal System Equation), augmenting a two-compartment model with bounded, state- and time-dependent neural corrections to elimination and intercompartmental distribution. Molecular structure parameterizes compound-specific behavior, while learned corrections refine dynamic equations. Across scaffold and temporal out-of-distribution evaluations, PK-MUSE achieved the lowest mean exposure-endpoint MdFE, reducing MdFE by 7.2%, 45.2%, and 26.2% versus state-of-the-art machine learning (HML- RF), deep learning compartmental (DeepCt), and PBPK (PK-Sim) models. Interrogating the corrections localized misspecification to distribution rather than elimination. By extending mechanistic simulators with data-driven constrained neural dynamics that preserve interpretability and generalize to unseen chemotypes, PK-MUSE advances digital twins for drug discovery.

pharmacology and toxicology↗

bioRxiv2026

Ultrafast and Accurate Selection of High-Quality Protein Complex Models from Large-Scale Prediction

In the post-AlphaFold era, advances in protein complex structure prediction have enabled the generation of thousands of candidate models per target, making the accurate and efficient selection of high-quality models from large candidate pools a critical challenge. Existing model-selection approaches commonly employ estimated model accuracy (EMA) methods, which include single-model and consensus-based methods. The former lack cross-model comparative information, whereas the latter often incur high computational costs from pairwise structural alignments and are influenced by candidate-pool quality. Here, we present DeepUMQA-Selection, a two-stage framework combining single-model accuracy estimators for preliminary selection with Ultrafast Shape Recognition (USR)-based structural alignment for consensus selection. Preliminary selection enriches the candidate pool with high-quality models and reduces subsequent structural comparisons, while USR-based structural alignment enables efficient consensus selection without explicit structural superposition. On the retrospective CASP16 QMODE3 benchmark dataset, DeepUMQA-Selection outperformed all participating methods and predictor-derived confidence metrics, as measured by mean Top-5 weighted penalty. The advantage was particularly pronounced for heteromeric targets, with a 13.9% improvement in model-selection performance relative to the best-performing CASP16 participating method. Furthermore, a confidence-assisted strategy enabled the framework to accommodate larger protein complexes while maintaining superior model-selection performance over competing methods, supporting its scalability. Runtime benchmarking showed that the proposed USR-based structural alignment achieved a substantial 317-fold speedup over US-align for candidate pools containing 1,000 models, demonstrating its computational efficiency for large-scale model selection. Our results demonstrate that DeepUMQA-Selection enables accurate and computationally efficient selection of high-quality protein complex models from large-scale candidate pools.

bioinformatics↗

bioRxiv2026

Cardiomyocyte-derived Wnt5a drives doxorubicin-induced cardiomyopathy by amplifying cellular senescence

Doxorubicin (DOX) is an effective anthracycline chemotherapeutic agent, but its use is limited by cardiotoxicity that can progress to cardiomyopathy and heart failure. Cellular senescence contributes to DOX-induced cardiac injury, yet the upstream signals that initiate and propagate senescence in the injured heart remain unclear. Here, we identify Wnt5a, a non-canonical Wnt ligand, as a mediator of anthracycline cardiomyopathy. WNT5A was increased in serum from cancer patients receiving anthracycline therapy and in a pathologic human cardiomyocyte population in the context of DOX-induced cardiomyopathy. In mouse hearts, DOX induced early cardiomyocyte-enriched Wnt5a expression before overt cardiac dysfunction. Cardiomyocyte-specific Wnt5a deletion attenuated DOX-induced cardiac dysfunction, fibrosis and senescence marker induction, whereas recombinant Wnt5a and cardiomyocyte-targeted Wnt5a overexpression were sufficient to promote cardiomyocyte senescence and cardiac dysfunction. Mechanistically, DOX activated a Wnt5a-Fzd2 feed-forward axis that amplified Wnt5a expression in cardiomyocytes and propagated senescence to neighboring fibroblasts. Genetic disruption of this pathway in cardiomyocytes, fibroblasts or senescent cells reduced DOX-induced cardiomyopathy. Pharmacological inhibition of Wnt5a signaling with secreted frizzled-related protein 5 suppressed DOX-induced cardiac injury without compromising the anticancer efficacy of DOX. These findings identify Wnt5a-Fzd2 signaling as a senescence-amplifying mechanism in anthracycline cardiomyopathy and suggest a therapeutic strategy to mitigate DOX cardiotoxicity.

cell biology↗

bioRxiv2026

MEGA-ODE: Learning Biologically Structured and Navigable Continuous Perturbation Dynamics from Sparse Omics

Perturbation-omics experiments usually measure only a subset of molecular feature, intervention and time space, leaving many response trajectories, perturbation effects and disease- or differentiation-associated transitions unobserved. Here we present MEGA-ODE, a graph-constrained continuous-time framework for reconstructing sparse dynamic omics landscapes, predicting unmeasured molecular states and prioritizing virtual perturbations toward defined biological endpoints. MEGA-ODE integrates molecular-network priors, graph neural ordinary differential equations and context-adaptive mixture-of-experts routing. In L1000 transcriptomic perturbations and CPPA proteomic drug-response data, MEGA-ODE improved held-out-feature and unseen-perturbation prediction over baseline methods, and in SARS-CoV-2 infection time-series data it remained competitive for future-time-point forecasting. In a COVID-19 patient cohort, predicted intermediate profiles improved retrospective disease-stage stratification relative to observed profiles alone, while expert programs highlighted immune and inflammatory signals associated with severity. Across the MAPK drug-response and stem-cell differentiation case studies, graph- and expert-level attributions prioritized perturbation-associated MAPK edges, developmental regulators and TF-target relationships supported by independent promoter-proximal ChIP-seq overlap. In hESC-to-definitive-endoderm differentiation, MEGA-ODE prioritized candidate transcription-factor perturbations predicted to shift 12-36 h profiles toward 96 h definitive-endoderm marker signatures, framing trajectory navigation as a concrete hypothesis-generation task. Together, these results support biologically structured continuous-time modeling for prediction, interpretation and virtual-perturbation prioritization from sparse temporal omics data.

bioinformatics↗

bioRxiv2026

GEM-GPT Enables Personalized Cell Type-Resolved Therapeutic Design for Systems Pharmacology

Generative AI is transforming drug discovery, yet most approaches follow one-drug-one-target paradigms ill-suited to the heterogeneity of chronic, systemic diseases. Systems pharmacology offers an alternative, but generative tools designed for it remain scarce. We introduce GEM-GPT, a transcriptomics-guided framework that generates personalized therapeutic candidate molecules intended to shift cell type-specific disease states toward healthy phenotypes. GEM-GPT uses a biology-inspired deep fusion architecture that couples a single-cell RNA-sequencing foundation model with a molecular GPT, modeling cell type-specific chemical-gene interactions throughout molecule generation rather than through fixed conditioning. Across bulk and single-cell chemical perturbations and CRISPR knock-out signatures, GEM-GPT outperforms state-of-the-art baselines, produces cell type-resolved molecules, and generalizes to unseen cellular contexts. In a case study on opioid use disorder (OUD), it generates novel candidates, recovers FDA-approved OUD-related drugs absent from training, and yields predicted binders to OUD-related targets. GEM-GPT bridges single-cell omics and molecular generation for personalized, cell-type-resolved, systems-aware therapeutic design.

pharmacology and toxicology↗

bioRxiv2026

mirCCC: Repression-aware graph learning for miRNA-mediated cell-cell communication inference

Cell-cell communication analyses usually focus on protein ligands and receptors and therefore miss extracellular vesicle transfer of microRNAs, an important route of signalling in cancer. Here we show that microRNA-mediated communication can be inferred from standard single-cell RNA sequencing by detecting coordinated decreases in the expression of validated miRNA target genes. We developed mirCCC, a computational framework that estimates cell-specific microRNA activity, models cellular sending and receiving capacity for extracellular vesicle transfer, and learns microRNA-resolved communication graphs from transcriptomic data. In synthetic benchmarks with strong confounding signals, mirCCC improved whereas all comparison methods declined. Applied to a human colorectal cancer atlas, mirCCC recovered known colorectal cancer-associated microRNAs and identified stromal- and myeloid-to-epithelial communication converging on a plasticity program linked to TGF-{beta} and Wnt/{beta}-catenin signalling. These results provide a practical route for studying extracellular vesicle-mediated communication in existing single-cell atlases. Author summaryCell-cell communication plays important roles across a wide range of biological processes. Most studies of cell-cell communication focus on interactions between ligands and receptors. However, cells can also use extracellular vesicles to deliver microRNAs (miRNAs), which regulate gene activity in receiving cells and contribute to cancer progression, immune regulation, and changes in the tissue environment. Since standard single-cell sequencing does not directly capture miRNAs, this mode of communication remains largely unexplored. Here we developed mirCCC, a computational method for inferring miRNA-mediated communication from standard single-cell transcriptomic data. Instead of detecting miRNAs directly, mirCCC examines the effects they leave in receiving cells. Active miRNAs often cause coordinated suppression of their target genes. mirCCC uses these coordinated reductions in the expression of known target genes to estimate miRNA activity and combines them with the ability of sending cells to package and release miRNAs and of receiving cells to process them. It can therefore reconstruct directional communication networks for individual miRNAs. Our evaluations show that mirCCC remains reliable in the presence of misleading signals and reveals biologically meaningful regulatory patterns in human colorectal cancer data. Overall, mirCCC provides a practical way to study miRNA-mediated cell-cell communication using existing single-cell transcriptomics datasets.

bioinformatics↗

bioRxiv2026

False discovery rate control for trustworthy AI-based de novo peptide sequencing

AI-based de novo peptide sequencing predicts peptide sequences from tandem mass spectra, enabling identification beyond predefined databases but leaving prediction reliability difficult to assess, particularly with respect to false discovery rate (FDR). In database search, FDR control is provided by target-decoy competition over a finite search space, whereas de novo predictions are generated in open sequence space and lack naturally matched sequence-level decoys. Here we introduce Counterpart Calibration Theory (CCT), a theory-guided framework that reframes de novo FDR control as a four-group score-ranking and threshold-selection problem over target-side predictions and matched counterpart-side comparators. Implemented in {pi}-NovoQC, CCT provides dual-level FDR control at the peptide-spectrum match and peptide levels. Across models, datasets, instruments and acquisition modes, {pi}-NovoQC achieves stable FDR control while preserving identification yield. In large-scale proteomic applications, {pi}-NovoQC recovers low-abundance in-database peptides missed by database search and provides de novo-supported protein-group and variant evidence.

bioinformatics↗

bioRxiv2026

Real Science Is Harder Than Benchmarks: Evaluating Advanced AI Frameworks on Published Studies. I. Uncertainty Quantification, ML on Therapeutic Data Commons, and Agent-Based Modeling

Artificial Intelligence (AI) frameworks for automating scientific research have shown strong performance on benchmarks, but their capacity to routinely reproduce results from multiple real-life published studies remains largely untested. We evaluated five advanced AI research frameworks (Kosmos, K-Dense, ToolUniverse, BioAgents from bio.xyz, and the AI Scientist-v2 from Sakana AI) on three real-life tasks (including two recently published papers) spanning uncertainty quantification for molecular property predictions, machine learning on Therapeutic Data Commons benchmarks, and agent-based modeling. AI frameworks demonstrated genuine strengths: generating original hypotheses, competently executing routine data acquisition and coding tasks, providing statistical measures of confidence often absent from the original papers, and producing well-formatted final reports. At the same time, our experiments revealed that real-world scientific tasks remain considerably harder than current benchmarks suggest. No AI framework matched the scope or depth of the original studies, results varied across multiple runs of the same framework with the same prompt, and we documented cases of severe hallucinations in final reports, gaps in literature coverage, and overconfident conclusions. Verification of AI outputs required substantial domain expertise. While these three tasks are only partially representative of the broader scientific landscape, they offer a starting point for developing a more rigorous methodology for evaluation of AI performance than what is currently practiced. We conclude that AI frameworks are already valuable for prototyping research directions and stress-testing completed studies, and some of the limitations documented here appear largely tractable through infrastructure improvements and continued development.

bioinformatics↗

bioRxiv2026

ROS Impair Mitophagy via PARylation of PINK1

Mitophagy is the process of selective autophagic clearance of damaged mitochondria and is closely implicated in neurodegenerative disease. PTEN-induced kinase 1 (PINK1) and a RBR E3 ubiquitin-protein ligase (Parkin) constitute a positive feedback loop in mitophagy initiation. It is known that reactive oxygen species (ROS) modulate mitophagy, while the exact regulatory mechanism remains largely elusive. Here, we found that exogenously applied ROS effectively block mitophagy induced by acute mitochondrial damage agents, which could be reversed by antioxidants. Mechanistically, ROS activate poly(ADP-ribose) polymerase 1 (PARP1), and suppression of PARP1 eliminates the inhibitory effect of ROS on mitophagy. Notably, PARP1 directly interacts with PINK1 and mediates its PARylation at residue E417, thereby negatively regulating PINK1 function. Collectively, our study identifies PARylation as a new form of post-translational modification of PINK1 and reveals a novel mechanism underlying the regulatory role of ROS in mitophagy by PARP1 activation and PARylation of PINK1. In briefGao et al. demonstrate that exogenous ROS inhibit mitophagy. Mechanistically, ROS activate PARP1, which mediates PARylation of PINK1, a central regulator of mitophagy, leading to its functional impairment. This study reveals a novel regulatory mechanism of ROS on mitophagy through PARP1 activation and identifies PARylation as a novel form of post-translational modification of PINK1. HighlightsO_LIROS block PINK1-Parkin-mediated mitophagy. C_LIO_LIROS activate PARP1. C_LIO_LIPARP1 suppression eliminates the inhibitory effect of ROS on mitophagy. C_LIO_LIPARylation of PINK1 by PARP1 impairs its activity and mitophagy. C_LI

Cell Biology↗

bioRxiv2026

PertDiffBench: Benchmarking Diffusion Models for Single-Cell Perturbation Response Prediction

Diffusion models are increasingly used to predict transcriptional responses to perturbations, but whether they improve on simpler generative and representation-based baselines remains unclear. Existing evaluations often do not separate the effects of model architecture, input representation, biological context and metric choice, making it difficult to determine where diffusion-based methods are useful. Here we introduce PertDiffBench, a standardized benchmark for diffusion-based transcriptomic perturbation prediction across single-cell and bulk RNA-seq datasets. PertDiffBench evaluates diffusion-based models across three complementary evaluation settings: standard prediction in known single-cell contexts and bulk perturbation conditions, generalization to unseen cell types, species, drugs and intermediate time points, and stress tests of feature dimensionality, input representation, noise type and gene ordering. Across these settings, diffusion models did not show a consistent advantage. scGen remained a strong baseline in common prediction tasks, whereas scDiffusion was the most competitive diffusion-based method in several generalization settings. Temporal imputation showed a different pattern, with a simple DDPM operating directly in expression space outperforming more specialized models. Stress tests showed that performance was model dependent and sensitive to feature dimensionality, encoder choice, noise type and gene ordering. Pretrained encoders did not consistently improve performance, with the classical scVI representation slightly exceeding STATE in seen-condition and unseen-cell-type settings. These results indicate that diffusion-model performance in perturbation response prediction depends strongly on task design and representation choice. PertDiffBench provides a practical framework for evaluating these models under biologically varied and stress-tested conditions.

bioinformatics↗

bioRxiv2026

SciCore-Omics: a tri-modal foundation model unifying histology, spatial transcriptomics and language for spatial biology

Histomorphology and spatial transcriptomics capture complementary aspects of tissue biology, but their relationships remain difficult to extract, align, and interpret at scale. Existing foundation models typically connect histology, omics, or language only pairwise, which limits their capacity to jointly infer molecular states, decode spatial tissue organization, and generate biologically grounded explanations. Here, we show SciCore-Omics, the first tri-modal foundation model linking histology images, spatial transcriptomics, and biological language. We constructed a spatially paired image-gene-text dataset comprising 151,182 spots across multiple tissues and performed a three-stage progressive training of SciCore-Omics on this dataset. Across gene expression prediction and spatial domain recognition, SciCore-Omics achieved 23.6-80.9% relative gains in task-specific metrics over the strongest external baselines. It further showed robust zero-shot generalization in histopathology classification, outperforming GPT-5 by 6.16 percentage points in mean accuracy across four benchmarks. Expert evaluation in 10 breast cancer cases confirmed its H&E-only case-level molecular reasoning capability. Together, our method demonstrates that a tri-modal framework can effectively bridge histomorphology and molecular state, providing a more general and interpretable foundation model for computational pathology and omics analysis.

bioinformatics↗

bioRxiv2026

Supercharged binding modules can modulate engineered poly(ethylene terephthalate) hydrolase thermostability and functional persistence

Poly(ethylene terephthalate) (PET) is a highly recalcitrant polyester plastic whose resistance to degradation has contributed to widespread environmental accumulation. Enzymatic PET depolymerization has emerged as a promising bioremediation strategy, but PET hydrolysis remains challenging due to the insoluble and semi-crystalline nature of PET and the poor thermostability of many PET hydrolases at elevated temperatures. Here, several electrostatically supercharged PET binding modules (PBM) were fused to a PET-hydrolyzing Cutinase Catalytic Domain (CD) from the thermophilic microbe Thermobifida fusca to investigate how engineered PBM surface charge influences PET hydrolysis behavior. All PBM designs were derived from a native T. fusca family-2a carbohydrate binding module (CBM) as starting template. Since PET exhibited a substantially negative zeta potential, and accordingly, all positively supercharged PBMs displayed the strongest PET binding interactions in pull-down binding assays. However, stronger PET binding did not translate to improved hydrolysis activity for the fusion constructs. Instead, a slightly negatively charged PBM-CD fusion (D2 construct) exhibited activity comparable to the Cutinase CD on finely milled PET powder while showing substantially improved activity on intact PET discs, suggesting potential advantages for depolymerization of minimally processed PET feedstocks. Thermostability analysis identified an approximately 10 {degrees}C increase in melting temperature for the D2 fusion construct, corresponding to enhanced catalytic persistence and a shifted optimal hydrolysis temperature. Consequently, this construct exhibited an approximately 2-fold increase in long-term hydrolysis activity on milled PET and up to a 10-fold increase on intact PET discs, even at high solids loadings, compared to the native Cutinase CD. Collectively, these findings demonstrate that thermostability, rather than adsorption to PET alone, is a dominant factor governing functional persistence of PET hydrolases.

bioengineering↗

bioRxiv2026

Causally validated phase-amplitude coupling enables high-fidelity motor decoding for next-generation brain-computer interfaces

Modern Brain-Computer Interfaces (BCIs) face a fundamental performance plateau due to their reliance on broad spectral power reduction (Event-Related Desynchronization, ERD) as the primary decoding feature. ERD acts as a coarse metabolic proxy for cortical activation, discarding the high-frequency temporal syntax necessary for high-dimensional motor control. Here, we demonstrate that Phase-Amplitude Coupling (PAC)--specifically, the phase-locking of high-gamma (70-150 Hz) firing to residual Beta oscillations--provides a high-fidelity temporal feature for decoding motor intent. Using high-density electrocorticography (HD-ECoG) during motor imagery, we show that incorporating Beta-PAC into machine-learning classifiers (LDA/SVM) significantly outperforms traditional power-based features. To definitively validate the causal robustness of this feature against mathematically-derived artifacts, we leveraged a rare in vivo human structural lesion model. In a patient with focal tumor infiltration of the motor tract, the metabolic gate (ERD) was preserved, yet structural uncoupling completely abolished Beta-PAC, collapsing network topology and reducing single-trial decoding accuracy to chance levels. Our findings causally validate Beta-PAC as a robust, independent control signal, establishing a physiological foundation for next-generation, phase-dependent neuroprostheses.

neuroscience↗

bioRxiv2026

Dual-view Guided Context-aware Network for Automated Bone Lesion Segmentation and Quantification in Whole-body SPECT

Whole-body SPECT bone scintigraphy reflects skeletal metabolic activity throughout the body and plays an indispensable role in the screening, treatment evaluation, and prognostic assessment of bone metastases in tumors. However, the automatic detection and segmentation of hypermetabolic bone lesions remain challenging due to low contrast, limited spatial resolution, and complex lesion distributions. In this study, we proposed Bone-Segnet, a dual-view guided automatic segmentation network for hypermetabolic bone lesions that integrated multi-scale feature modeling, global context modeling, and view-conditioned modulation. Pixel-level annotated anterior and posterior whole-body bone scintigraphy images were used for model training and prediction. The proposed network enhanced the recognition of low-contrast and small-scale lesions through small-lesion enhancement and multi-scale contextual modeling. A Transformer module was further introduced to strengthen global feature representation, while cross-view collaborative modeling was achieved by incorporating the complementary characteristics of anterior and posterior imaging. Experimental results demonstrated that the proposed method outperformed existing approaches across multiple evaluation metrics, with the Dice score improving from 0.7440 to 0.8750, indicating a substantial improvement in segmentation performance. Further quantitative analysis based on the segmentation results revealed significant differences among disease types in lesion count, pixel burden, and spatial distribution patterns, reflecting the heterogeneity of disease-related skeletal metabolic activity. Overall, the proposed method improved automatic lesion segmentation performance and enabled quantitative analysis of lesion burden and spatial distribution patterns, providing objective data support for the assessment of related diseases. Index Terms--Whole-body SPECT, bone lesion segmentation, dual-view modeling, quantitative analysis.

bioinformatics↗

bioRxiv2026

A Scalable Sign-Aware Multi-Omics Knowledge Graph Foundation Model for Mechanistic Drug Action and Clinical Response Predictions

Mechanistically predicting drug action requires distinguishing activating from inhibitory interactions across broad chemical space, yet most biomedical knowledge graphs and graph neural networks (GNNs) rely on unsigned associations that obscure regulatory logic and have a limited chemical coverage. Here we present SIGMA-KG (SIGned Multi-omics Atlas Knowledge Graph) and FLASH (Fast Lightweight Architecture for Signed Heterogeneous GNN), a graph foundation model pretrained through self-supervised learning on SIGMA-KG. SIGMA-KG integrates chemogenomic perturbation, transcriptomic, proteomic, and clinical data while explicitly encoding biological polarity and directionality. FLASH preserves polarity composition across multi-hop pathways through structural balance principles, enabling scalable mechanistic reasoning. Without task-specific fine-tuning, FLASH consistently outperforms or matches nine state-of-the-art unsigned, relational, and signed graph baselines across drug mode-of-action prediction, clinical response modeling, and drug-drug interaction prediction, while substantially improving computational efficiency. FLASH further enables explainable inductive drug repurposing, achieving a 69.6% external clinical validation success rate across four complex diseases.

systems biology↗

bioRxiv2026

DrugPTM-Bench: A Large-Scale Dataset for Predictive Modeling of Drug-Induced Cell Type-Specific Protein Post-Translational Modifications

Protein post-translational modifications (PTMs), particularly phosphorylation, serve as the primary "molecular switches" that orchestrate cellular signaling and drug response. While PTM dysregulation is a hallmark of cancer and neurodegeneration, the lack of standardized, drug-perturbed datasets has hindered the development of predictive models capable of capturing context-dependent PTM responses. Effective predictive modeling must therefore integrate multidimensional data, including the specific drug, dosage, treatment duration, cellular background, and the modified site. However, existing PTM resources remain largely static and fail to capture drug-induced regulation across these critical dimensions. To address this gap, we present DrugPTM-Bench, a curated, large-scale benchmark derived from decryptM-derived dose-dependent PTM measurements, standardizing site-level drug response across 7 cancer cell lines, 27 drugs, and 11,167 proteins. Comprising 99.5% phosphorylation events, the dataset includes six time points, 16 dosage levels, and pEC50 potency values (half-maximal effective concentration). We formulate a classification task to identify upregulated, downregulated, or unchanged PTM sites (following a drug treatment), a critical step in deciphering drug Mechanism of Action (MoA) and target engagement. Our evaluation reveals that in protein-disjoint out-of-distribution (OOD) setting, baseline machine learning and deep learning models struggle to recover minority regulation classes, while standard rebalancing strategies improve recall only at the cost of precision and overall F1-score. These results indicate that current methods do not learn robust decision boundaries between regulated and unchanged PTM events. DrugPTM-Bench provides a phosphoproteomics benchmark for modeling drug-induced PTM regulation in imbalanced biological settings. Beyond classification, DrugPTM-Benchs retention of pEC50 values, drug perturbation profiles, and site-level sequence context enables additional predictive tasks including drug potency regression, mechanism-of-action prediction from PTM fingerprints, and drug-specific PTM site sensitivity ranking, establishing a multi-task benchmark for PTM-centric drug discovery. Ultimately, DrugPTM-Bench establishes a rigorous framework for developing robust, context-aware models to elucidate drug MoA and signaling dynamics.

systems biology↗