The DrugMatrix database contains systematically generated toxicogenomics data from short-term in vivo studies for over 600 chemicals. However, most potential endpoints are missing due to a lack of experimental measurements. Therefore, we leveraged matrix factorization and machine learning methods to predict the missing values, which includes gene expression across eight tissues on two expression platforms along with paired clinical chemistry, hematology, and histopathology. We propose a method, ToxCompl, that applies systematic hybrid sampling guided by Bayesian optimization in conjunction with low-rank matrix factorization to predict the missing values. In-depth validation of the ToxCompl predicted data from machine learning, biological, and toxicological perspectives shows that the predicted differential gene expression aligns well with what would be anticipated. This includes examining the connectivity pattern of predicted gene expression responses, characterizing molecular pathway-level responses from sets of differentially expressed genes, evaluating known transcriptional biomarkers of tissue toxicity, and characterizing predicted apical endpoints. For example, we identified kidney toxicants using the transcriptional biomarker Havcr1. All measured and predicted DrugMatrix data (i.e., gene expression, clinical chemistry, hematology, and histopathology) are available to the public (https://rstudio.niehs.nih.gov/toxcompl/). Notably, predicted clinical chemistry of subtle effects and histopathological prediction are two areas we will continue to improve. The main advantage of the ToxCompl approach is that it drastically extends the toxicogenomic landscape into many data-poor tissues in the absence of acquiring additional experimental data, thereby allowing researchers to formulate mechanistic hypotheses about effects in tissues that have been underrepresented in the literature.

BACKGROUND:Ventricular assist device (VAD) explantation following myocardial functional recovery (MFR) remains rare in both adults and children. We aim to study the characteristics and outcomes of durable VAD explantation following MFR in children from the Advanced Cardiac Therapies Improving Outcomes Network (ACTION) registry. METHODS:All explants of durable VAD, defined as intracorporeal continuous flow devices (ICF) and paracorporeal pulsatile flow devices (PPF), performed between March 2012 and August 2023 were identified within ACTION. Temporary devices were excluded. The primary outcome was VAD and heart transplant-free survival following explantation. RESULTS:A total of 31 out of 938 patients (3.3%) underwent durable VAD explantation following MFR after a median VAD support duration of 111 days (IQR 59-183). The median age was 1.17 years (IQR 0.32-10.1), and weight was 8.4 kg (IQR 6-23.8). The underlying diagnosis was dilated cardiomyopathy/myocarditis in 64% (20/31) and congenital heart disease in 10% (3/31); others (8/31) accounted for 26%. Following VAD explantation, 10% (3/29) underwent reimplantation of VAD, 14% (4/29) underwent heart transplantation, and 3.4% (1/29) died following explantation. At the last follow-up, 72% (21/29) are alive without reimplantation of a VAD or heart transplantation at a median duration post-explant of 2 years. CONCLUSIONS:Myocardial functional recovery resulting in VAD explantation remains a rare occurrence in children. However, the majority of children did not require advanced heart failure therapies after durable VAD explant, suggesting appropriate selection of candidates. Strategies to identify patients who may benefit from explantation and predictors of myocardial functional recovery after VAD are warranted.

Chemical toxicity assessment commonly includes in vivo rat exposure experiments, with transcriptomic measurements collected at various exposure times and chemical doses. The mechanisms underlying chemical-induced toxicity are then inferred by analyzing changes in gene expression. Recently, genome-scale metabolic models (GSMs), which represent the metabolic network of a cell/organism and contain metabolites, reactions, genes, and the relationship between the genes and reactions, have been used to provide a systems-level understanding of gene expression. However, most of the algorithms that integrate gene expression with GSMs require familiarity with MATLAB or Python programming, making them less accessible for users without computational experience. Here, we introduce ToxMet (https://toxmet.bhsai.org), an open-access, user-friendly web application that provides tabular and graph-based network views to visualize the latest rat GSM (iRno v4.2) and predicts chemical-induced metabolic perturbations in rat tissues by integrating toxicogenomic measurements with the rat GSM. ToxMet uses two well-validated computational algorithms, TIMBR and Pheflux, to predict metabolic perturbations and provides the prediction results as interactive and downloadable tables, scatter plots, and network visualizations. As such, the web tool can process a maximum of 10 conditions for a single job, and the results can be used for dose-response studies to monitor organ metabolism at the subsystem level. We evaluated ToxMet's ability to predict toxicity mechanisms by applying it to publicly available toxicogenomic data for two exemplar toxicants: Gentamicin and thioacetamide, which are known to induce kidney and liver injury, respectively. ToxMet predicted known toxicity mechanisms for both chemicals, thus demonstrating its ability to provide novel insights into the metabolic mechanisms of chemical-induced toxicity and aid in the discovery of biomarkers and therapeutics using gene expression data.

The liver and kidneys are the primary organs that clear chemicals from the body, yet they often exhibit different pathological responses to the same systemic exposure. While high-throughput transcriptomics (HTT) can map broad molecular perturbations, identifying the differential mechanisms that govern organ-specific injury remains a challenge. To address this knowledge gap, we investigated the dose-dependent effects of three peroxisome proliferator-activated receptor alpha agonists—coumarin, fenofibrate, and perfluorooctanoic acid (PFOA)—on liver and kidney metabolism to understand the organ-specific mechanisms of toxicity. We used HTT data from 5-day in vivo rat exposure studies and performed a comparative analysis on the paired liver and kidney data, using gene co-expression and pathway enrichment analyses together with genome-scale models, to investigate their differential sensitivity to chemical exposure. Our results revealed that all three chemicals caused larger gene perturbations in the liver compared to the kidney, in agreement with their well-known hepatotoxicity. Fenofibrate and PFOA triggered a profound upregulation in pathways related to fatty acid metabolism but suppressed several pathways related to amino acid metabolism in both organs. All three chemicals showed a strong upregulated antioxidant response in the liver, indicating an adaptive response to chemical stress, with fenofibrate and PFOA entering uncontrolled endoplasmic reticulum stress, which was also observed for PFOA in the kidney. Our comparative analysis revealed a striking divergence in cellular repair mechanisms in response to stress across both tissues: while the liver strongly upregulated apoptotic and cell-cycle repair pathways, signaling active hepatotoxicity, the kidney consistently downregulated these same vital processes in response to all three chemicals, indicating differential mechanisms that may lead to organ toxicity.

Background Exposure to environmental chemicals can influence fetal growth, and these alterations are associated with adverse health outcomes across the lifespan. Personal care products (PCPs) are frequently used by women of reproductive age, including many chemicals known to adversely impact fetal development. Given the high number of chemicals in PCPs, a class-based approach of grouping these chemicals by common characteristic(s) (e.g., chemical structure, mode of action) may support future hazard identification evaluations. Objectives To develop a systematic evidence map (SEM) identifying and characterizing the scientific literature on gestational exposure to PCPs or their chemical constituents and fetal growth to support potential chemical class-based assessments. Methods Following standardized systematic review methodology, three databases were searched for relevant human and experimental animal studies through June 2024. Study characteristics were extracted and summarized in interactive visualizations and text. Results Of the four main chemical classes assessed, phthalates and phenols were most frequently studied (40% and 30% of studies, respectively), followed by per- and polyfluoroalkyl substances (PFAS), and parabens. Few studies evaluated product use. Birthweight was the most frequently assessed outcome (approximately 99% of studies). Many human studies evaluated potential modifying factors of health (% of studies), such as infant sex (58%), race and ethnicity (7%), and socioeconomic status (1%). Discussion Given the availability of well-studied “anchor” chemicals within related chemical groups, this SEM supports the feasibility of class-based approaches to evaluate the association between phthalates, PFAS, phenols, and parabens and fetal growth. Further research on PCPs and fetal growth should address areas of uncertainty, including data gaps on potential effect modifiers, such as socioeconomic status.

Transcriptomic profiling technologies have advanced the analysis of biological and toxicological responses. However, substantial differences in probe design, dynamic range, gene coverage, and preprocessing pipelines across platforms introduce artifacts that limit cross-study integration and hinder the reuse of historical datasets. We aim to develop computational methods for accurate cross-platform translation to maximize the value of legacy resources. We present TransPlatformer a deep learning framework for translating gene expression profiles across heterogeneous toxicogenomics platforms. TransPlatformer employs a novel attention-based architecture to map high-dimensional fold-change vectors from legacy microarray technologies to current platforms. Models are trained and evaluated using DrugMatrix, spanning three technological generations. We investigate mixed-tissue, single-tissue, and cross-tissue training paradigms and benchmark performance against multilayer perceptron and matrix-completion baselines. In mixed-tissue training, TransPlatformer achieves a greater than 50

Aqueous film-forming foams (AFFFs) are complex product mixtures that often contain per- and polyfluorinated alkyl substances (PFAS) to enhance fire suppression and protect firefighters. However, PFAS have been associated with a range of adverse health effects (e.g., liver and thyroid disease and cancer), and innovative approach methods to better understand their toxicity potential and identify safer alternatives are needed. In this study, we investigated a set of 30 substances (e.g., AFFF, PFAS, and clinical drugs) using differentiated cultures of human hepatocytes (HepaRG, 2D), high-throughput transcriptomics, deep learning of cell morphology images, and liver enzyme leakage assays with benchmark dose analysis to (1) predict the potency ranges for human liver injury, (2) delineate gene- and pathway-level transcriptomic points-of-departure for molecular hazard characterization and prioritization, (3) characterize human hepatocellular response similarities to inform regulatory read-across efforts, and (4) introduce an innovative approach to translate mechanistic hepatocellular response data to predict the potency ranges for PFAS-induced hepatomegaly in vivo. Collectively, these data fill important mechanistic knowledge gaps with PFAS/AFFF and represent a scalable platform to address the thousands of PFAS in commerce for greener chemistries and next-generation risk assessments.

Per- and polyfluoroalkyl substances (PFASs) are widespread in the environment, bioaccumulate in humans, and lead to disease and organ injury, such as liver steatosis. However, we lack a clear understanding of how these chemicals cause organ-level toxicity. Here, we aimed to analyze PFAS-induced metabolic perturbations in male and female rat livers by combining a genome-scale metabolic model (GEM) and toxicogenomics. The combined approach overcomes the limitations of the individual methods by taking into account the interaction between multiple genes for metabolic reactions and using gene expression to constrain the predicted mechanistic possibilities. We obtained transcriptomic data from an acute exposure study, where male and female rats received a daily PFAS dose for five consecutive days, followed by liver transcriptome measurement. We integrated the transcriptome expression data with a rat GEM to computationally predict the metabolic activity in each rat’s liver, compare it between the control and PFAS-exposed rats, and predict the benchmark dose (BMD) at which each chemical induced metabolic changes. Overall, our results suggest that PFAS-induced metabolic changes occurred primarily within the lipid and amino acid pathways and were similar between the sexes but varied in the extent of change per dose based on sex and PFAS type. Specifically, we identified that PFASs affect fatty acid-related pathways (biosynthesis, oxidation, and sphingolipid metabolism), energy metabolism, protein metabolism, and inflammatory and inositol metabolite pools, which have been associated with fatty liver and/or insulin resistance. Based on these results, we hypothesize that PFAS exposure induces changes in liver metabolism and makes the organ sensitive to metabolic diseases in both sexes. Furthermore, we conclude that male rats are more sensitive to PFAS-induced metabolic aberrations in the liver than female rats. This combined approach using GEM-based predictions and BMD analysis can help develop mechanistic hypotheses regarding how toxicant exposure leads to metabolic disruptions and how these effects may differ between the sexes, thereby assisting in the metabolic risk assessment of toxicants.

Extracellular vesicles (EVs) are emitted from cells throughout the body and serve as signaling molecules that mediate disease development. Emerging evidence suggests that per- and polyfluoroalkyl substances (PFAS) impact EV release and content, influencing liver toxicity. Still, the upstream regulators of EV changes affected by PFAS exposure remain unclear. This study evaluated the hypothesis that PFAS exposures, individually and in a mixture, alter the expression of genes involved in EV regulation at concentrations comparable to genes involved in global biological response mechanisms. HepG2 liver cells were treated at multiple concentrations with individual PFOS, PFOA, or PFHxA, in addition to an equimolar PFAS mixture. Gene expression data were analyzed using three pipelines for concentration-response modeling, with results compared against empirically derived datasets. Final benchmark concentration (BMC) modeling was conducted via Laplace model averaging in BMDExpress (v3). BMCs were derived at an individual gene level and across different gene sets, including Gene Ontology (GO) annotations as well as a custom EV regulation gene set. To determine relative PFAS contributions to the evaluated mixture, relative potency factors were calculated across resulting BMCs using PFOS as a standard reference chemical. Results demonstrated that PFAS exposures altered the expression of genes involved in EV regulation, particularly for genes overlapping with endoplasmic reticulum stress. EV regulatory gene changes occurred at similar BMCs as global gene set alterations, supporting concurrent regulation and the role of EVs in PFAS toxicology. This application of transcriptomics-based BMC modeling further validates its utility in capturing both established and novel pathways of toxicity.

We developed the Reasoning Over Biomedical Objects linked in Knowledge Oriented Pathways (ROBOKOP) application as an open-source knowledge graph system to support evidence-based biomedical discovery and hypothesis generation. This study aimed to apply ROBOKOP to suggest biological mechanisms that might explain the hypothesized relationship between exposure to the herbicide and lipid-lowering drug clofibrate, an activator of peroxisome proliferator-activated receptor-α (PPARA), and hepatic fibrosis. We queried ROBOKOP to first establish that it could demonstrate a relationship between clofibrate and PPARA as a validation test and second to identify intermediary genes and biological processes or activities that might relate the activation of PPARA by clofibrate to hepatic fibrosis. Queries of ROBOKOP returned several paths relating clofibrate, PPARA, and hepatic fibrosis. One path suggested the following: clofibrate - affects / increases_ expression_ of / increases_ activity_ of / increases_ response_ to / decreases_ response_ to / is_ related_ to - PPARA - is_ actively_ involved_ in - cellular response to lipid - actively_ involves - CCL2 - is_ genetically_ associated_ with - hepatic fibrosis. This result established a relationship between clofibrate and PPARA and further suggested that PPARA is actively involved in the cellular response to lipids, which actively involves the chemokine ligand CCL2, a gene genetically associated with hepatic fibrosis; thus, we can infer that PPARA, upon activation by clofibrate, plays a role in hepatic fibrosis. We conclude that ROBOKOP can be used to derive insights into biological mechanisms that might explain relationships between environmental exposures and liver toxicity.

Gene expression biomarkers have the potential to identify genotoxic and non-genotoxic carcinogens, providing opportunities for integrated testing and reducing animal use. In August 2022, an International Workshops on Genotoxicity Testing (IWGT) workshop was held to critically review current methods to identify genotoxicants using transcriptomic profiling. Here, we summarize the findings of the workgroup on the state of the science regarding the use of transcriptomic biomarkers to identify genotoxic chemicals in vitro and in vivo. A total of 1341 papers were examined to identify the biomarkers that show the most promise for identifying genotoxicants. This analysis revealed two independently derived in vivo biomarkers and three in vitro biomarkers that, when used in conjunction with standard computational techniques, can identify genotoxic chemicals in vivo (rat or mouse liver) or in human cells in culture using different gene expression profiling platforms, with predictive accuracies of ≥92%. These biomarkers have been validated to differing degrees but typically show high reproducibility across transcriptomic platforms and model systems. They offer several advantages for applications in different contexts of use in genotoxicity testing including: early signal detection, moderate-to-high-throughput screening capacity, adaptability to different cell types and tissues, and insights on mechanistic information on DNA-damage response. Workshop participants agreed on consensus statements to advance the regulatory adoption of transcriptomic biomarkers for genotoxicity. The participants agreed that transcriptomic biomarkers have the potential to be used in conjunction with other biomarkers in integrated test strategies in vitro and using short-term rodent exposures to identify genotoxic and non-genotoxic chemicals that may cause cancer and heritable genetic effects. Following are the consensus statements from the workgroup. Transcriptomic biomarkers for genotoxicity can be used in Weight of Evidence (WoE) evaluation to: determine potential genotoxic mechanisms and hazards; identify misleading positives from in vitro genotoxicity assays; serve as new approach methodologies (NAMs) integrated into the standard battery of genotoxicity tests. Several transcriptomic biomarkers have been developed from sufficiently robust training data sets, validated with external test sets, and have demonstrated performance in multiple laboratories. These transcriptomic biomarkers can be used following established study designs and models designated through existing validation exercises in WoE evaluation. Bridging studies using a selection of training and test chemicals are needed to deviate from the established protocols to confirm performance when a transcriptomic biomarker is being applied in other: tissues, cell models, or gene expression platforms. Top dose selection and time of gene expression analysis are critical and should be established during transcriptomic biomarker development. These conditions are the only ones suited for transcriptomic biomarker use unless additional bridging or pharmacokinetic studies are conducted. Temporal effects for genotoxicants that operate via distinct mechanisms should be considered in data interpretation. Fixed transcriptomic biomarker gene sets and analytical processes do not need to be independently rederived in biomarker validation. Validation should focus on the performance of the gene set in external test sets. Robust external testing should ensure a minimum of additional chemicals spanning genotoxic and non-genotoxic modes of action. Genes in the transcriptomic biomarker do not need to be known to be mechanistically involved in genotoxicity responses. Existing frameworks described for NAMs could be applied for validation of transcriptomic biomarkers. Reproducibility of bioinformatic analysis is critical for the regulatory application of transcriptomic biomarkers. A bioinformatics expert should be involved with creating reproducible methods for the qualification and application of each transcriptomic biomarker.

The polyfluorinated alkyl substance (PFAS) Nafion BP2 (1,1,2,2-tetrafluoro-2-[1,1,1,2,3,3-hexafluoro-3-(1,1,2,2-tetrafluoroethoxy)propan-2-yl]oxyethane-1-sulfonic acid) has been detected in surface and ground water, as well as in the blood of people living near PFAS manufacturing facilities. Given that very little is known about the potential toxicity of Nafion BP2 and safe exposure levels have not yet been determined, we performed a benchmark dose analysis of phenotypic and genomic effects in mice. Male and female Balb-c mice were exposed daily to Nafion BP2 at multiple doses for 7 d by oral gavage. Full-genome transcript profiling showed that Nafion BP2 in both sexes activates a number of transcription factors linked to liver toxicity, including constitutive androstane receptor (CAR), pregnane X receptor, and NRF2, but unlike other long-chain PFAS, there was no activation of peroxisome proliferator-activated receptor α. Nafion BP2 caused hepatic steatosis in both sexes. Benchmark dose (BMD) estimates for 15 non-genomic effects were 0.25 mg/kg/d and above. BMDs for transcriptional effects were 0.04 mg/kg/d and above. The most sensitive gene sets in both males and females were related to effects on xenobiotic metabolism and the cell cycle, which are plausibly related at a mechanistic level to the activation of CAR. The xenobiotic metabolism and cell cycle findings were largely consistent when dose values based on internal dose were employed in the analysis. These values, along with the identification of molecular targets linked to hazards, may facilitate the determination of human health guidance for Nafion BP2.

Fenofibrate, a peroxisome proliferator-activated receptor α (PPARα) agonist, is widely prescribed to treat hyperlipidemia and has therapeutic potential in liver and kidney diseases. However, fenofibrate is also associated with adverse effects, including elevated creatinine and liver and kidney toxicity, although the underlying mechanisms remain unclear. In addition, how fenofibrate regulates lipid metabolism differently in the liver and kidney is not well understood. Therefore, in this study, we investigated the dose-dependent effects of fenofibrate on liver and kidney metabolism in rats, with a focus on PPARα activation and potential mechanisms contributing to organ-specific toxicity. We used high-throughput transcriptomic data from 5-day rat in vivo studies, where rats were exposed to fenofibrate, and performed pathway enrichment, injury module, and detailed individual gene comparison analyses to investigate how liver and kidney metabolism were differentially altered between the two organs. Fenofibrate exposure significantly increased liver but not kidney weights and caused larger perturbations in the liver compared to the kidney transcriptome, with the majority of the changes related to PPARα regulation. Interestingly, our study revealed that the PPARα and RXRα genes are differentially regulated between the liver and kidney. In addition, we identified several differences between them in cellular and mitochondrial fatty acid transport, lipoprotein metabolism, fatty acid oxidation, branched-chain amino acid degradation, and glucose metabolism pathways. Furthermore, we identified transcriptomic inflection points at which the changes in the PPARα-mediated regulation of lipid metabolism switched from beneficial to deleterious as the fenofibrate concentration increased leading to liver injury, providing potential mechanisms of toxicity.

Because the liver plays a vital role in the clearance of exogenous chemical compounds, it is susceptible to chemical-induced toxicity. Animal-based testing is routinely used to assess the hepatotoxic potential of chemicals. Although large-scale high-throughput sequencing data can indicate the genes affected by chemical exposures, we need system-level approaches to interpret these changes. To this end, we developed an updated rat genome-scale metabolic model to integrate large-scale transcriptomics data and utilized a chemical structure similarity-based ToxProfiler tool to identify chemicals that bind to specific toxicity targets to understand the mechanisms of toxicity. We used high-throughput transcriptomics data from a 5-day in vivo study where rats were exposed to different non-toxic and hepatotoxic chemicals at increasing concentrations and investigated how liver metabolism was differentially altered between the non-toxic and hepatotoxic chemical exposures. Our analysis indicated that the genes identified via toxicity target analysis and those mapped to the metabolic model showed a distinct gene expression pattern, with the majority showing upregulation for hepatotoxicants compared with non-toxic chemicals. Similarly, when we mapped the metabolic genes at the pathway level, we identified several pathways in carbohydrate, amino acid, and lipid metabolism that were significantly upregulated for hepatotoxic chemicals. Furthermore, using our system-level integration of gene expression data with the rat metabolic model, we could differentiate metabolites in these pathways that were systematically elevated or suppressed due to hepatotoxic versus non-toxic chemicals. Thus, using our combined approach, we were able to identify a set of potential gene signatures that clearly differentiated liver toxic responses from non-toxic chemicals, which helped us identify potential metabolic pathways and metabolites that are systematically associated with the toxicant exposure.

There is growing recognition across broad sectors of the toxicology community that gene expression biomarkers have the potential to identify genotoxic and nongenotoxic carcinogens through a weight-of-evidence approach, providing opportunities to reduce reliance on the 2-year bioassay to identify carcinogens. In August 2022, a workshop within the International Workshops on Genotoxicity Testing (IWGT) was held to critically review current methods to identify genotoxicants using various 'omics profiling methods. Here, we describe the findings of a workshop subgroup focused on the state of the science regarding the use of biomarkers to identify chemicals that act as genotoxicants in vivo. A total of 1341 papers were screened to identify those that were most relevant. While six published biomarkers with characterized accuracy were initially examined, four of the six were not considered further, because they had not been tested for classification accuracy using additional sets of chemicals or other transcript profiling platforms. Two independently derived biomarkers used in conjunction with standard computational techniques can identify genotoxic chemicals in vivo (rat liver or both rat and mouse liver) on different gene expression profiling platforms. The biomarkers have predictive accuracies of ≥92%. These biomarkers have the potential to be used in conjunction with other biomarkers in integrated test strategies using short-term rodent exposures to identify genotoxic and nongenotoxic chemicals that cause cancer.

Purpose: The TEAMMATE Trial is the first randomized clinical trial of immunosuppression in pediatric heart transplant (HT) recipients. We aimed to describe coronary angiography results from the trial (currently embargoed).

Benchmark dose (BMD) modeling estimates the dose of a chemical that causes a perturbation from baseline. Transcriptional BMDs have been shown to be relatively consistent with apical end point BMDs, opening the door to using molecular BMDs to derive human health-based guidance values for chemical exposure. Metabolomics measures the responses of small-molecule endogenous metabolites to chemical exposure, complementing transcriptomics by characterizing downstream molecular phenotypes that are more closely associated with apical end points. The aim of this study was to apply BMD modeling to in vivo metabolomics data, to compare metabolic BMDs to both transcriptional and apical end point BMDs. This builds upon our previous application of transcriptomics and BMD modeling to a 5-day rat study of triphenyl phosphate (TPhP), applying metabolomics to the same archived tissues. Specifically, liver from rats exposed to five doses of TPhP was investigated using liquid chromatography-mass spectrometry and 1H nuclear magnetic resonance spectroscopy-based metabolomics. Following the application of BMDExpress2 software, 2903 endogenous metabolic features yielded viable dose-response models, confirming a perturbation to the liver metabolome. Metabolic BMD estimates were similarly sensitive to transcriptional BMDs, and more sensitive than both clinical chemistry and apical end point BMDs. Pathway analysis of the multiomics data sets revealed a major effect of TPhP exposure on cholesterol (and downstream) pathways, consistent with clinical chemistry measurements. Additionally, the transcriptomics data indicated that TPhP activated xenobiotic metabolism pathways, which was confirmed by using the underexploited capability of metabolomics to detect xenobiotic-related compounds. Eleven biotransformation products of TPhP were discovered, and their levels were highly correlated with multiple xenobiotic metabolism genes. This work provides a case study showing how metabolomics and transcriptomics can estimate mechanistically anchored points-of-departure. Furthermore, the study demonstrates how metabolomics can also discover biotransformation products, which could be of value within a regulatory setting, for example, as an enhancement of OECD Test Guideline 417 (toxicokinetics).

In the original publication [...].

The DrugMatrix Database contains systematically generated toxicogenomics data from short-term in vivo studies for over 600 chemicals. However, most of the potential endpoints in the database are missing due to a lack of experimental measurements. We present our study on leveraging matrix factorization and machine learning methods to predict the missing values in the DrugMatrix, which includes gene expression across eight tissues on two expression platforms along with paired clinical chemistry, hematology, and histopathology measurements. One major challenge we encounter is the skewed distribution of the available measured data, in terms of both tissue sources and values. We propose a method, ToxiCompl, that applies systematic hybrid sampling guided by Bayesian optimization in conjunction with low-rank matrix factorization to recover the missing values. ToxiCompl achieves good training and validation performance from a machine learning perspective. We further conduct an in-depth validation of the predicted data from biological and toxicological perspectives with a series of analyses. These include examining the connectivity pattern of predicted gene expression responses, characterizing molecular pathway-level responses from sets of differentially expressed genes, evaluating known transcriptional biomarkers of tissue toxicity, and characterizing pre-dicted apical endpoints. Our analysis shows that the predicted differential gene expression, broadly speaking, aligns with what would be anticipated. For example, in most instances, our predicted differentially expressed gene lists offer a connectivity level comparable to that of measured data in connectivity analysis. Using Havcr1, a known transcriptional biomarker of kidney injury, we identify treatments that, based on the predicted expression data, manifest kidney toxicity in a manner that is mechanistically plausible and supported by the literature. Characterization of the predicted clinical chemistry data suggests that strong effects are relatively reliably predicted, while more subtle effects pose a greater challenge. In the case of histopathological prediction, we find a significant overprediction due to positivity bias in the measured data. Developing methods to deal with this bias is one of the areas we plan to target for future improvement. The main advantage of the ToxiCompl approach is that, in the absence of additional experimental data, it drastically extends the toxicogenomic landscape into a number of data-poor tissues, thereby allowing researchers to formulate mechanistic hypotheses about effects in tissues that have been underrepresented in the literature. All measured and predicted DrugMatrix data (i.e., gene expression, clinical chemistry, hematology, and histopathology) are available to the public through an intuitive GUI interface that allows for data retrieval, gene set analysis and high dimensional visualization of gene expression similarity (). ### Competing Interest Statement The authors have declared no competing interest.