Drug interaction with Udp-Glucuronosyltransferase (UGT) enzymes is a predictor of drug-induced liver injury.

DILI frequently contributes to the attrition of new drug candidates and is a common cause for the withdrawal of approved drugs from the market. Although some noncytochrome P450 (non-CYP) metabolism enzymes have been implicated in DILI development, their association with DILI outcomes has not been systematically evaluated. In this study, we analyzed a large data set comprising 317 drugs and their interactions in vitro with 42 non-CYP enzymes as substrates, inducers, and/or inhibitors retrieved from historical regulatory documents. We examined how these in vitro drug-enzyme interactions are correlated with the drugs' potential for DILI concern, as classified in the Liver Toxicity Knowledge Base database. Our study revealed that drugs that inhibit non-CYP enzymes are significantly associated with high DILI concern. Particularly, interaction with UDP-glucuronosyltransferases (UGT) enzymes is an important predictor of DILI outcomes. Further analysis indicated that only pure UGT inhibitors and dual substrate inhibitors, but not pure UGT substrates, are significantly associated with high DILI concern. Notably, drug interactions with UGT enzymes may independently predict DILI, and their combined use with the rule-of-two model further improves overall predictive performance. These findings could expand the currently available tools for assessing the potential for DILI in humans.

Machine Learning to Predict Drug-Induced Liver Injury and Its Validation on Failed Drug Candidates in Development.

Drug-induced liver injury (DILI) poses a significant challenge for the pharmaceutical industry and regulatory bodies. Despite extensive toxicological research aimed at mitigating DILI risk, the effectiveness of these techniques in predicting DILI in humans remains limited. Consequently, researchers have explored novel approaches and procedures to enhance the accuracy of DILI risk prediction for drug candidates under development. In this study, we leveraged a large human dataset to develop machine learning models for assessing DILI risk. The performance of these prediction models was rigorously evaluated using a 10-fold cross-validation approach and an external test set. Notably, the random forest (RF) and multilayer perceptron (MLP) models emerged as the most effective in predicting DILI. During cross-validation, RF achieved an average prediction accuracy of 0.631, while MLP achieved the highest Matthews Correlation Coefficient (MCC) of 0.245. To validate the models externally, we applied them to a set of drug candidates that had failed in clinical development due to hepatotoxicity. Both RF and MLP accurately predicted the toxic drug candidates in this external validation. Our findings suggest that in silico machine learning approaches hold promise for identifying DILI liabilities associated with drug candidates during development.

A novel function of claudin-5 in maintaining the structural integrity of the heart and its implications in cardiac pathology.

This study aims to investigate the role of claudin-5 (Cldn5) in cardiac structural integrity. Proteomic analysis was performed to screen the protein profiles in enlarged left atrium from atrial fibrillation (AF) patients. Cldn5 shRNA adeno-associated virus (AAV) or siRNA was injected into the mouse left ventricle or added into HL1 cells respectively to knockdown Cldn5 in cardiomyocytes to observe whether the change of Cldn5 influences cardiac morphology and function, and affects those protein expressions stem from the proteomic analysis. Mitochondrial density and membrane potential were also measured by Mitotracker staining and JC-1 staining under the confocal microscope in HL1 cells. Cldn5 was reduced in cardiomyocytes from the left atrial appendage of AF patients compared to non-AF donors. Proteomic analysis showed 83 proteins were less abundant and 102 proteins were more abundant in AF patients. KEGG pathway analysis showed less abundant CACNA2D2, CACNB2, MYL2 and MAP6 were highly associated with dilated cardiomyopathy. Cldn5 shRNA AAV injection caused severe cardiac atrophy, dilation and myocardial dysfunction in mice. The decreases in mitochondrial numbers and mitochondrial membrane potentials in HL1 cells were observed after Cldn5 knockdown. We demonstrated for the first time the mechanism of Cldn5 downregulation-induced myocyte atrophy and myocardial dysfunction might be associated with the downregulation of CACNA2D2, CACNB2, MYL2 and MAP6, and mitochondrial dysfunction in cardiomyocytes.

Automatic text classification of drug-induced liver injury using document-term matrix and XGBoost.

Introduction: Regulatory agencies generate a vast amount of textual data in the review process. For example, drug labeling serves as a valuable resource for regulatory agencies, such as U.S. Food and Drug Administration (FDA) and Europe Medical Agency (EMA), to communicate drug safety and effectiveness information to healthcare professionals and patients. Drug labeling also serves as a resource for pharmacovigilance and drug safety research. Automated text classification would significantly improve the analysis of drug labeling documents and conserve reviewer resources. Methods: We utilized artificial intelligence in this study to classify drug-induced liver injury (DILI)-related content from drug labeling documents based on FDA's DILIrank dataset. We employed text mining and XGBoost models and utilized the Preferred Terms of Medical queries for adverse event standards to simplify the elimination of common words and phrases while retaining medical standard terms for FDA and EMA drug label datasets. Then, we constructed a document term matrix using weights computed by Term Frequency-Inverse Document Frequency (TF-IDF) for each included word/term/token. Results: The automatic text classification model exhibited robust performance in predicting DILI, achieving cross-validation AUC scores exceeding 0.90 for both drug labels from FDA and EMA and literature abstracts from the Critical Assessment of Massive Data Analysis (CAMDA). Discussion: Moreover, the text mining and XGBoost functions demonstrated in this study can be applied to other text processing and classification tasks.

A Single-Arm Multi-Center Phase II Clinical Trial of Cadonilimab (anti-PD-1/CTLA-4) in Combination with or without Conventional Second-Line Treatment for Patients with Extensive Stage Small Cell Lung Cancer.

BACKGROUND: Cadonilimab (AK104) is a bispecific IgG-single-chain Fv fragment (ScFv) antibody that binds to PD-1 and CTLA-4. Cadonilimab has shown encouraging anti-tumour activity and a favourable safety profile in several tumour types. In second-line treatment, there is no defined standard of care for patients with extensive-stage small-cell lung cancer (ES-SCLC). Cadonilimab is expected to show substantial clinical efficacy. OBJECTIVE: To assess the antitumor activity and safety of cadonilimab monotherapy or combination with conventional therapy in ES-SCLC patients who failed first-line treatment. METHODS: In this multicenter, open-label, phase II study, ES-SCLC patients who had failed first-line treatment, also aged 18 years to 70 years with histologically or cytologically confirmed ES-SCLC, and an Eastern Cooperative Oncology Group performance status (ECOG-PS) of 0-2 were eligible. Patients will receive cadonilimab 10 mg/kg every three weeks (Q3 W) among 24 months until progressive disease (PD) or adverse events (AE) discovery. The primary endpoint is progression-free survival (PFS). TRIAL REGISTRATION: NCT05901584.

Composite Plot for Visualizing Aminotransferase and Bilirubin Changes in Clinical Trials of Subjects with Abnormal Baseline Values.

INTRODUCTION: On-treatment excursions of liver laboratory test values in clinical trials involving subjects with underlying liver disease are relevant for the efficacy and safety assessment of drug products and biologics. Existing visualization and analysis tools do not efficiently provide an integrated view of these excursions when baseline liver tests are abnormal. OBJECTIVE: The aim of this study was to develop a composite plot that enables visualization of on-treatment changes in liver test results both as multiples of the upper limit of normal defined by each laboratory's reference population (×ULN) and multiples of the subjects' baseline (×BLN) values. METHODS: The composite plot approach combines biochemical evaluation for drug-induced severe hepatotoxicity (eDISH) plots sequentially applied to subjects' baseline and peak on-treatment liver test results normalized by ULN and integrates them into a four-panel shift plot of peak on-treatment values normalized by BLN. RESULTS: The composite plot enabled efficient assessment of improvement in liver test values during treatment compared with pretreatment in subjects treated with the investigational drug (or the natural history of placebo-treated subjects) and identified outlier subjects for potential drug-induced liver injury. CONCLUSION: For studies in subjects with abnormal baseline values, the composite plot has potential application in the assessment of beneficial and concerning on-treatment modifications in liver test values in reference to the individual subject's baseline and population threshold values.

Medical device report analyses from MAUDE: Device and patient outcomes, adverse events, and sex-based differential effects.

Post-market medical device-associated failures and patient problems are reported in Medical Device Reports (MDRs) to the US Food and Drug Administration. Reports are accessible through Manufacturer and User Facility Device Experience (MAUDE), a database including both required and voluntary submissions. We present an overview of >10 million MDRs received from 2011 to 2021. Approximately 92% of reporting issues represent medical device physical or functional failures, categorized from 1704 codes related to medical device integrity or function. ∼8% were coded adverse events (AEs). Patient outcomes are reported via 998 patient codes in 19 medical specialties (cardiovascular, orthopedic, etc.). ∼40% of patient reports indicated "no health consequences"; however, a small number of devices had consistently high AE reports. While overall reports did not exhibit a sex-based dichotomy, ∼9% of the reported AEs occurred more frequently in females, many of which were related to immune effects. The analyses are subject to uncertainties and potential bias based on data available and data selected for analysis. However, such an overview of post-market MDR data, not previously published, fills a gap in understanding medical device issues and patient-based outcomes related to medical device use. Trends identified may be subjects of additional hypotheses, analysis, and research.

TCCIA: a comprehensive resource for exploring CircRNA in cancer immunotherapy.

BACKGROUND: Immunotherapies targeting immune checkpoints have gained increasing attention in cancer treatment, emphasizing the need for predictive biomarkers. Circular RNAs (circRNAs) have emerged as critical regulators of tumor immunity, particularly in the PD-1/PD-L1 pathway, and have shown potential in predicting immunotherapy efficacy. Yet, the detailed roles of circRNAs in cancer immunotherapy are not fully understood. While existing databases focus on either circRNA profiles or immunotherapy cohorts, there is currently no platform that enables the exploration of the intricate interplay between circRNAs and anti-tumor immunotherapy. A comprehensive resource combining circRNA profiles, immunotherapy responses, and clinical outcomes is essential to advance our understanding of circRNA-mediated tumor-immune interactions and to develop effective biomarkers. METHODS: To address these gaps, we constructed The Cancer CircRNA Immunome Atlas (TCCIA), the first database that combines circRNA profiles, immunotherapy response data, and clinical outcomes across multicancer types. The construction of TCCIA involved applying standardized preprocessing to the raw sequencing FASTQ files, characterizing circRNA profiles using an ensemble approach based on four established circRNA detection tools, analyzing tumor immunophenotypes, and compiling immunotherapy response data from diverse cohorts treated with immune checkpoint blockades (ICBs). RESULTS: TCCIA encompasses over 4,000 clinical samples obtained from 25 cohorts treated with ICBs along with other treatment modalities. The database provides researchers and clinicians with a cloud-based platform that enables interactive exploration of circRNA data in the context of ICB. The platform offers a range of analytical tools, including browse of identified circRNAs, visualization of circRNA abundance and correlation, association analysis between circRNAs and clinical variables, assessment of the tumor immune microenvironment, exploration of tumor molecular signatures, evaluation of treatment response or prognosis, and identification of altered circRNAs in immunotherapy-sensitive and resistant tumors. To illustrate the utility of TCCIA, we showcase two examples, including circTMTC3 and circMGA, by employing analysis of large-scale melanoma and bladder cancer cohorts, which unveil distinct impacts and clinical implications of different circRNA expression in cancer immunotherapy. CONCLUSIONS: TCCIA represents a significant advancement over existing resources, providing a comprehensive platform to investigate the role of circRNAs in immuno-oncology.

Microfluidic Fabrication of Monodisperse and Recyclable TiO2-Poly(ethylene glycol) Diacrylate Hybrid Microgels for Removal of Methylene Blue from Aqueous Medium.

Nearly monodisperse titanium oxide-polyethylene glycol diacrylate [TiO2-P(EGDA)] hybrid microbeads containing 0.5 wt % TiO2 nanoparticles entrapped within a P(EGDA) cross-linked polymeric network were synthesized using a modular Lego-inspired glass capillary microfluidic device. TiO2-P(EGDA) hybrid microgels were characterized by optical microscopy, scanning electron microscopy, X-ray diffraction, energy dispersive X-ray spectroscopy, and thermogravimetric analysis. The fabricated TiO2-P(EGDA) hybrid microgel system showed 100% removal efficiency of methylene blue (MB) from its 1-3 ppm aqueous solutions after 4 h of UV light irradiation at 0.2 mW/cm2 at the loading of 25 g/L photocatalyst beads in the reaction mixture, corresponding to the loading of naked TiO2 of just 0.025 g/L. No decrease in photocatalytic efficiency was observed in 10 repeated runs with recycled photocatalyst using a fresh 1 ppm MB solution in each cycle. The rate of photocatalytic degradation was controlled by the UV light irradiance, catalyst loading, and the initial dye concentration. Physical adsorption of MB onto the surface of composite microgel was also observed. The adsorption data was best fitted with the Langmuir adsorption isotherm and the Elovich kinetic model. TiO2-P(EGDA) microgel beads are biocompatible, can be prepared with a tunable size in the microfluidic device, and can easily be separated from the reaction mixture by gravity settling. The TiO2-P(EGDA) system can be used for the removal of other toxic dyes and micropollutants from industrial wastewater.

Selective Adsorption of Ionic Species Using Macroporous Monodispersed Polyethylene Glycol Diacrylate/Acrylic Acid Microgels with Tunable Negative Charge.

Monodispersed polyethylene glycol diacrylate (PEGDA)/acrylic acid (AA) microgels with a tuneable negative charge and macroporous internal structure have been produced using a Lego-inspired droplet microfluidic device. The surface charge of microgels was controlled by changing the content of AA in the monomer mixture from zero (for noncharged PEGDA beads) to 4 wt%. The macroporosity of the polymer matrix was introduced by adding 20 wt% of 600-MW polyethylene glycol (PEG) as a porogen material into the monomer mixture. The porogen was successfully leached out with acetone after UV-crosslinking, which resulted in micron-sized cylindrical pores with crater-like morphology, uniformly arranged on the microgel surface. Negatively charged PEGDA/AA beads showed improved adsorption capacity towards positively charged organic dyes (methylene blue and rhodamine B) compared to neutral PEGDA beads and high repulsion of negatively charged dye molecules (methyl orange and congo red). Macroporous microgels showed better adsorption properties than nonporous beads, with a maximum adsorption capacity towards methylene blue of 45 mg/g for macroporous PEGDA/AA microgels at pH 8.6, as compared to 23 mg/g for nonporous PEGDA/AA microgels at the same pH. More than 98% of Cu(II) ions were removed from 50 ppm solution at pH 6.7 using 2.7 mg/mL of macroporous PEGDA/AA microgel. The adsorption of cationic species was significantly improved when pH was increased from 3 to 9 due to a higher degree of ionization of AA monomeric units in the polymer network. The synthesized copolymer beads can be used in drug delivery to achieve improved loading capacity of positively charged therapeutic agents and in tissue engineering, where a negative charge of scaffolds coupled with porous structure can help to achieve improved permeability of high-molecular-weight metabolites and nutrients, and anti-fouling activity against negatively charged species.

In silico modeling-based new alternative methods to predict drug and herb-induced liver injury: A review.

New approach methods (NAMs) have been developed to predict a wide range of toxicities through innovative technologies. Liver injury is one of the most extensively studied endpoints due to its severity and frequency, occurring among populations that consume drugs or dietary supplements. In this review, we focus on recent developments of in silico modeling for liver injury prediction using deep learning and in vitro data based on adverse outcome pathways (AOPs). Despite these models being mainly developed using datasets generated from drug-like molecules, they were also applied to the prediction of hepatotoxicity caused by herbal products. As deep learning has achieved great success in many different fields, advanced machine learning algorithms have been actively applied to improve the accuracy of in silico models. Additionally, the development of liver AOPs, combined with big data in toxicology, has been valuable in developing in silico models with enhanced predictive performance and interpretability. Specifically, one approach involves developing structure-based models for predicting molecular initiating events of liver AOPs, while others use in vitro data with structure information as model inputs for making predictions. Even though liver injury remains a difficult endpoint to predict, advancements in machine learning algorithms and the expansion of in vitro databases with relevant biological knowledge have made a huge impact on improving in silico modeling for drug-induced liver injury prediction.

Microfluidic Fabrication and Applications of Microgels and Hybrid Microgels.

Smart microgels have gained much attention because of their wide range of applications in the field of biomedical, environmental, nanotechnological and catalysis sciences. Most of the applications of microgels are strongly affected by their morphology, size and size distribution. Various methodologies have been adopted to obtain polymer microgel particles. Droplet microfluidic techniques have been widely reported for the fabrication of highly monodisperse microgel particles to be used for various applications. Monodisperse microgel particles of required size and morphology can be achieved via droplet microfluidic techniques by simple polymerization of monomers in the presence of suitable crosslinker or by gelation of high molecular weight polymers. This report gives recent research progress in fabrication, characterization, properties and applications of microgel particles synthesized by microfluidic methods.

Computational models for predicting liver toxicity in the deep learning era.

Drug-induced liver injury (DILI) is a severe adverse reaction caused by drugs and may result in acute liver failure and even death. Many efforts have centered on mitigating risks associated with potential DILI in humans. Among these, quantitative structure-activity relationship (QSAR) was proven to be a valuable tool for early-stage hepatotoxicity screening. Its advantages include no requirement for physical substances and rapid delivery of results. Deep learning (DL) made rapid advancements recently and has been used for developing QSAR models. This review discusses the use of DL in predicting DILI, focusing on the development of QSAR models employing extensive chemical structure datasets alongside their corresponding DILI outcomes. We undertake a comprehensive evaluation of various DL methods, comparing with those of traditional machine learning (ML) approaches, and explore the strengths and limitations of DL techniques regarding their interpretability, scalability, and generalization. Overall, our review underscores the potential of DL methodologies to enhance DILI prediction and provides insights into future avenues for developing predictive models to mitigate DILI risk in humans.

Facile Microfluidic Fabrication of Biocompatible Hydrogel Microspheres in a Novel Microfluidic Device.

Poly(ethylene glycol) diacrylate (PEGDA) microgels with tuneable size and porosity find applications as extracellular matrix mimics for tissue-engineering scaffolds, biosensors, and drug carriers. Monodispersed PEGDA microgels were produced by modular droplet microfluidics using the dispersed phase with 49-99 wt% PEGDA, 1 wt% Darocur 2959, and 0-50 wt% water, while the continuous phase was 3.5 wt% silicone-based surfactant dissolved in silicone oil. Pure PEGDA droplets were fully cured within 60 s at the UV light intensity of 75 mW/cm2. The droplets with higher water content required more time for curing. Due to oxygen inhibition, the polymerisation started in the droplet centre and advanced towards the edge, leading to a temporary solid core/liquid shell morphology, confirmed by tracking the Brownian motion of fluorescent latex nanoparticles within a droplet. A volumetric shrinkage during polymerisation was 1-4% for pure PEGDA droplets and 20-32% for the droplets containing 10-40 wt% water. The particle volume increased by 36-50% after swelling in deionised water. The surface smoothness and sphericity of the particles decreased with increasing water content in the dispersed phase. The porosity of swollen particles was controlled from 29.7% to 41.6% by changing the water content in the dispersed phase from 10 wt% to 40 wt%.

Whole Exome Sequencing Reveals Genetic Variants in HLA Class II Genes Associated With Transplant-free Survival of Indeterminate Acute Liver Failure.

INTRODUCTION: Indeterminate acute liver failure (IND-ALF) is a rare clinical syndrome with a high mortality rate. Lacking a known etiology makes rapid evaluation and treatment difficult, with liver transplantation often considered as the only therapeutic option. Our aim was to identify genetic variants from whole exome sequencing data that might be associated with IND-ALF clinical outcomes. METHODS: Bioinformatics analysis was performed on whole exome sequencing data for 22 patients with IND-ALF. A 2-tier approach was used to identify significant single-nucleotide polymorphisms (SNPs) associated with IND-ALF clinical outcomes. Tier 1 identified the SNPs with a higher relative risk in the IND-ALF population compared with those identified in control populations. Tier 2 determined the SNPs connected to transplant-free survival and associated with model for end-stage liver disease serum sodium and Acute Liver Failure Study Group prognostic scores. RESULTS: Thirty-one SNPs were found associated with a higher relative risk in the IND-ALF population compared with those in controls, of which 11 belong to the human leukocyte antigen (HLA) class II genes but none for the class I. Further analysis showed that 5 SNPs: rs796202376, rs139189937, and rs113473719 of HLA-DRB5; rs9272712 of HLA-DQA1; and rs747397929 of IDO1 were associated with a higher probability of IND-ALF transplant-free survival. Using 3 selected SNPs, a model for the polygenic risk score was developed to predict IND-ALF prognoses, which are comparable with those by model for end-stage liver disease serum sodium and Acute Liver Failure Study Group prognostic scores. DISCUSSION: Certain gene variants in HLA-DRB5, HLA-DQA1, and IDO1 were found associated with IND-ALF transplant-free survival. Once validated, these identified SNPs may help elucidate the mechanism of IND-ALF and assist in its diagnosis and management.

Redefinition of Fatty Liver Disease from NAFLD to MAFLD through the Lens of Drug Development and Regulatory Science.

Metabolic (dysfunction)-associated fatty liver disease (MAFLD) affects a third of the population and is a leading cause of liver-related death. Since no effective treatments exist, novel approaches to drug development are required. Unfortunately, outdated terminology and definitions of the disease are hampering efforts to develop new drugs and treatments. An international consensus panel has put forth an influential proposal for the disease to be renamed from nonalcoholic fatty liver disease (NAFLD) to MAFLD, including a proposal for how the disease should be diagnosed. As allies with the many stakeholders in MAFLD care-including patients, patients' advocates, clinicians, researchers, nurse and allied health groups, regional societies, and others-we are aware of the negative consequences of the NAFLD term and definition. We share the sense of urgency for change and will act in new ways to achieve our goals. Although there is much work to be done to overcome clinical inertia and reverse worrisome recent trends, the MAFLD initiative provides a firm foundation to build on. It provides a roadmap for moving forward toward more efficient care and affordable, sustainable drug and device innovation in MAFLD care. We hope it will bring promising new opportunities for a brighter future for MAFLD care and improve care and outcomes for patients of one of the globe's largest and costliest public health burdens. From this viewpoint, we have revisited this initiative through the perspectives of drug development and regulatory science.

Integrative approaches for studying the role of noncoding RNAs in influencing drug efficacy and toxicity.

INTRODUCTION: Drug efficacy and toxicity are important factors for evaluation in drug development. Drug metabolizing enzymes and transporters (DMETs) play an essential role in drug efficacy and toxicity. Noncoding RNAs (ncRNAs) have been implicated to influence inter-individual variations in drug efficacy and safety by regulating DMETs. An efficient strategy is urgently needed to identify and functionally characterize ncRNAs that mediate drug efficacy and toxicity through regulating DMETs. AREAS COVERED: We outline an integrative strategy to identify ncRNAs that modulate DMETs. We include reliable tools and databases for computational prediction of ncRNA targets with regard to their advantages and limitations. Various biochemical, molecular, and cellular assays are discussed for in vitro experimental verification of the regulatory function of ncRNAs. In vivo approaches for association of ncRNAs with drug treatment and toxicity are also reviewed. EXPERT OPINION: A streamlined integration of computational prediction and wet-lab validation is important to elucidate mechanisms of ncRNAs in the regulation of DMETs related to drug efficacy and safety. Bioinformatic analyses using open-access tools and databases serve as a powerful booster for ncRNA Research in toxicology. Further refinement of computational algorithms and experimental technologies is needed to improve accuracy and efficiency in ncRNA target identification and characterization.

Machine Learning Models for Predicting Liver Toxicity.

Liver toxicity is a major adverse drug reaction that accounts for drug failure in clinical trials and withdrawal from the market. Therefore, predicting potential liver toxicity at an early stage in drug discovery is crucial to reduce costs and the potential for drug failure. However, current in vivo animal toxicity testing is very expensive and time consuming. As an alternative approach, various machine learning models have been developed to predict potential liver toxicity in humans. This chapter reviews current advances in the development and application of machine learning models for prediction of potential liver toxicity in humans and discusses possible improvements to liver toxicity prediction.

A systematic comparison of hepatobiliary adverse drug reactions in FDA and EMA drug labeling reveals discrepancies.

Drug labeling informs physicians and patients on the safe and effective use of medication. However, recent studies suggested discrepancies in labeling of the same drug between different regulatory agencies. Here, we evaluated the hepatic safety information in labeling for 549 medications approved by the US Food and Drug Administration (FDA) and the European Medicines Agency (EMA). Limited discrepancies were found regarding risk for hepatic adverse drug reactions (ADRs) (8.7% in hepatic ADR warnings and 21.3% in contraindication for liver disease), while caution should be exercised over drugs with inconsistencies in contraindications for liver disease and evidence for hepatotoxicity (4.9%). Most discrepancies were attributable to less-severe hepatic events and low-frequency hepatic ADR reports and had limited implication on clinical outcomes.

BERT-Based Natural Language Processing of Drug Labeling Documents: A Case Study for Classifying Drug-Induced Liver Injury Risk.

Background & Aims: The United States Food and Drug Administration (FDA) regulates a broad range of consumer products, which account for about 25% of the United States market. The FDA regulatory activities often involve producing and reading of a large number of documents, which is time consuming and labor intensive. To support regulatory science at FDA, we evaluated artificial intelligence (AI)-based natural language processing (NLP) of regulatory documents for text classification and compared deep learning-based models with a conventional keywords-based model. Methods: FDA drug labeling documents were used as a representative regulatory data source to classify drug-induced liver injury (DILI) risk by employing the state-of-the-art language model BERT. The resulting NLP-DILI classification model was statistically validated with both internal and external validation procedures and applied to the labeling data from the European Medicines Agency (EMA) for cross-agency application. Results: The NLP-DILI model developed using FDA labeling documents and evaluated by cross-validations in this study showed remarkable performance in DILI classification with a recall of 1 and a precision of 0.78. When cross-agency data were used to validate the model, the performance remained comparable, demonstrating that the model was portable across agencies. Results also suggested that the model was able to capture the semantic meanings of sentences in drug labeling. Conclusion: Deep learning-based NLP models performed well in DILI classification of drug labeling documents and learned the meanings of complex text in drug labeling. This proof-of-concept work demonstrated that using AI technologies to assist regulatory activities is a promising approach to modernize and advance regulatory science.