Xiaoni Ai

Drug-induced liver injury (DILI) remains a major cause of clinical attrition and postmarketing withdrawal, but structure only DILI predictors are difficult to compare because public benchmarks are vulnerable to compound overlap, scaffold similarity and shared label provenance. We present Oakuloid ^TM , an open DILI prediction framework that pairs a leakage audited structure based model with an optional iBAC 3D primary human hepatocyte IC ₅₀ /C _max confirmation signal. The structural model integrates gradient boosted descriptor backbones, fingerprint random forests and LivTox proxy-DILI features through a logistic meta learner. Its evaluation is designed as part of the contribution: internal DILIrank, strict external TDC, scaffold disjoint TDC and independent Geci provenance checks are reported with released per compound predictions. Oakuloid reaches AUROC 0.811 on the strict external TDC benchmark and remains competitive under scaffold and fully clean TDC filtering. A channel attribution ablation shows that the external benchmark lead is driven by descriptor based gradient boosted trees rather than by DILIPredictor derived proxy features, reducing a potential circularity concern. The wet lab IC ₅₀ /C _max signal is largely orthogonal to structure and supports a confirmation mode that shifts the internal operating point toward higher specificity without claiming a universal AUROC gain. Oakuloid is released with code, model artifacts, calibration analysis, a 122 compound wet lab benchmark and a model card under the Apache License 2.0, supporting reproducible DILI screening and benchmark auditing.

The tumor microenvironment (TME) is a complex ecosystem composed of tumor cells, cancer-associated fibroblasts (CAFs), immune suppressive cells, and the extracellular matrix (ECM), playing a crucial role in tumor development and CAR-T cell therapy efficacy. CAR-T therapy has shown promise in hematological malignancies but faces challenges in solid tumors due to the TME’s ability to suppress CAR-T cell infiltration, proliferation, and cytotoxicity. Traditional drug evaluation models, such as 2D cell cultures and animal models, have significant limitations due to oversimplification of the in vivo environment or physiological differences between species. Organoid models offer a more biomimetic approach but often fail to fully recapitulate the TME’s complexity and heterogeneity. Our research developed a tumor organoid and CAF co-culture model using the IBAC co-culture chip, demonstrating that CAFs significantly impact CAR-T cell therapy efficacy by forming physical (e.g., fibronectin) and chemical (e.g., IL-10) barriers that prevent CAR-T cell infiltration and cytotoxicity. This model provides a high-biomimetic platform for investigating the TME’s effects on CAR-T therapy and highlights the importance of incorporating a comprehensive stromal component into in vitro models to enhance their predictive power for cancer treatment.

Idiosyncratic drug-induced liver injury (iDILI) poses significant challenges in both drug development and clinical practice due to its unpredictable nature and poorly understood mechanism. The current in vitro iDILI models are limited in their ability to replicate dynamic paracrine signaling in the inflammatory microenvironment. Here, we develop an iDILI model on a stacked array chip, allowing ease of assembly and disassembly for precise temporal manipulation of 3D liver microtissue and macrophages. First, the iDILI model is constructed and optimized on the chip to effectively distinguish drugs inducing idiosyncratic versus intrinsic liver injuries. Next, the iDILI mechanism is investigated using nimesulide (NIM) as a case study. Our organ-on-a-chip model successfully recapitulates iDILI, offering a platform to distinguish drugs between intrinsic and idiosyncratic liver injury. Our findings revealed that NIM-induced iDILI triggered inflammation-induced injury in the liver microtissues through activating the TNF pathways. Moreover, NIM-induced iDILI promotes the M1 polarization of macrophages through CCL5-mediated paracrine dynamics, influenced by the interactions between hepatocytes and macrophages. Leveraging the flexibility of the chip, we observe a dynamic equilibrium between preactivation of inflammation and the pretreatment of NIM during iDILI process. Therefore, our developed iDILI model on a stacked array chip provides a valuable tool for identifying iDILI drugs and understanding the importance of temporal specificity in intercellular signaling in iDILI.

The development of an in vitro model for predicting drug permeability through the human blood–brain barrier (BBB) will greatly accelerate the development of neural therapy. Previously reported platforms for BBB model construction cannot meet the requirements of constant-rate and high-throughput flow, as well as compatibility with the commercial meter for real-time transendothelial electrical resistance (TEER) measurement. Herein, a constant-rate perfused array chip (cPAC) was developed to establish a brain endothelium model for screening drug permeability. The cPAC consisted of 24 units with four layers. Three reservoirs on the top had a 0.5 mm center-to-center spacing, enabling real-time detection of the TEER with the commercial volt–ohm meter. With the optimized chip design, the constant-rate and high-throughput flow by gravity was achieved. Compared with the static culture of the Transwell, the brain endothelium model on the cPAC exhibited superior performance in barrier function, efflux functionality of the transporters, and reversible osmotic opening of the brain endothelium. More importantly, the permeability of model drugs on the cPAC matched the in vivo results with the correlation coefficient reaching 0.994. Finally, the brain endothelium model was cocultured with 3D tumor cells for simultaneous evaluation of drug permeability and brain tumor therapy. The drug efficacy at the target cells on the coculture model was also consistent with clinical findings. These results demonstrated that this platform provides a promising tool for brain endothelium model establishment to predict drug permeability and brain therapy. We anticipate the cPAC to be widely accepted for establishing various barrier models.

Two-dimensional (2D) tumor model has always poorly predicted drug response of animal model due to the lack of recapitulation of tumor microenvironment. Establishing a biomimetic, controllable, and cost-effective three-dimensional (3D) model and large-scale validation of its in vivo predictivity has shown promise in bridging the gap between the 2D tumor model and animal model. Here, we established a matrigel-based 3D micro-tumor model on an array chip for large-scale anticancer drug evaluation. Compared with the 2D tumor model, the 3D tumor model on the chip showed spheroid morphology, slower proliferation kinetics, and comparable reproducibility. Next, the results of the chemotherapeutic evaluation from 18 drugs against 27 cancer cell lines showed 17.6% of drug resistance on the 3D tumor model. Moreover, the evaluation results of targeted drugs showed expected sensitivity and higher specificity on the 3D tumor model compared with the 2D model. Finally, the evaluation results on the 3D tumor model were more consistent with the in vivo cell-derived xenograft model, and excluded 95% false-positive results from the 2D model. Overall, the matrigel-based 3D micro-tumor model on the array chip provides a promising tool to accelerate anticancer drug discovery.

An in vitro human renal proximal tubule model that represents the proper transporter expression and pronounced epithelial polarization is necessary for the accurate prediction of nephrotoxicity. Here, we constructed a high-throughput human renal proximal tubule model based on an integrated biomimetic array chip (iBAC). Primary human renal proximal tubule epithelial cells (hRPTECs) cultured on this microfluidic platform were able to form a tighter barrier, better transporter function and more sensitive nephrotoxicity prediction than those on the static Transwell. Compared with the human immortalized HK2 model, the hRPTECs model on the chip gained improved apical-basolateral polarization, barrier function and transporter expression. Polymyxin B could induce nephrotoxicity not only from the apical of the hRPTECs, but also from the basolateral side on the iBAC. However, other chemotherapeutic agents, such as doxorubicin and sunitinib, only induced nephrotoxicity from the apical surface of the hRPTECs on the iBAC. In summary, our renal proximal tubule model on the chip exhibits improved epithelial polarization and membrane transporter activity, and can be implemented as an effective nephrotoxicity-screening toolkit.

Drug-induced liver injury (DILI) is a leading cause of therapy failure in the clinic and also contributes much to acute liver failure cases. Investigations of predictive sensitivity in animal models have limitations due to interspecies differences. Previously reported in vitro models of liver injury based on primary human hepatocytes (PHHs) cannot meet the requirements of high physiological fidelity, low cost, simple operation and high throughput with improved sensitivity. Herein, we developed an integrated biomimetic array chip (iBAC) for establishing extracellular matrix (ECM)-based models. A collagen-based 3D PHH model was constructed on the iBAC as a case for prediction of clinical drug-induced liver injury at throughput. The iBAC has a three-layer structure with a core component of 3D implanting holes. At an initial cell seeding numbers of 5000-10000, the collagen-based 3D PHH model was optimized with improved and stabilized liver functionality, including cell viability, albumin and urea production. Moreover, basal activities of most metabolic enzymes on the iBAC were maintained for at least 12 days. Next, a small-scale hepatotoxicity screening indicated that the 3D PHH model on the iBAC was more sensitive for predicting hepatotoxicity than the 2D PHH model on the plate. Finally, a large-scale screening of liver toxicity using 122 clinical drugs further demonstrated that the collagen-based 3D PHH model on the iBAC had superior predictive sensitivity compared to all previously reported in vitro models. These results indicated the importance of the 3D collagen for liver physiological functionality and hepatotoxicity prediction. We anticipant it being a promising tool for risk assessment of drug-induced hepatotoxicity with a widespread acceptance in drug industry.