Quantifying the Benefits of Imputation over QSAR Methods in Toxicology Data Modeling

Imputation machine learning (ML) surpasses traditional approaches in modeling toxicity data. The method was tested on an open-source data set comprising approximately 2500 ingredients with limited in vitro and in vivo data obtained from the OECD QSAR Toolbox. By leveraging the relationships between different toxicological end points, imputation extracts more valuable information from each data point compared to well-established single end point methods, such as ML-based Quantitative Structure Activity Relationship (QSAR) approaches, providing a final improvement of up to around 0.2 in the coefficient of determination. A significant aspect of this methodology is its resilience to the inclusion of extraneous chemical or experimental data. While additional data typically introduces a considerable level of noise and can hinder performance of single end point QSAR modeling, imputation models remain unaffected. This implies a reduction in the need for laborious manual preprocessing tasks such as feature selection, thereby making data preparation for ML analysis more efficient. This successful test, conducted on open-source data, validates the efficacy of imputation approaches in toxicity data analysis. This work opens the way for applying similar methods to other types of sparse toxicological data matrices, and so we discuss the development of regulatory authority guidelines to accept imputation models, a key aspect for the wider adoption of these methods.

Mixtures of persistent organic pollutants are found in vital organs of late gestation human fetuses

Persistent organic pollutants (POPs) are industrial chemicals with long half-lives. Early life exposure to POPs has been associated with adverse effects. Fetal exposure is typically estimated based on concentrations in maternal serum or placenta and little is known on the actual fetal exposure. We measured the concentrations of nine organochlorine pesticides (OCPs), ten polychlorinated biphenyl (PCB) congeners, and polybrominated diphenyl ether (PBDE) congeners by gas chromatography - tandem mass spectrometry in maternal serum, placenta, and fetal tissues (adipose tissue, liver, heart, lung and brain) in 20 pregnancies that ended in stillbirth (gestational weeks 36-41). The data were combined with our earlier data on perfluoroalkyl substances (PFASs) in the same cohort (Mamsen et al. 2019). HCB, p,p'-DDE, PCB 138 and PCB 153 were quantified in all samples of maternal serum, placenta and fetal tissues. All 22 POPs were detected in all fetal adipose tissue samples, even in cases where they could not be detected in maternal serum or placenta. Tissue:serum ratios were significantly higher in later gestations, male fetuses, and pregnancies with normal placental function. OCPs showed the highest tissue:serum ratios and PFAS the lowest. The highest chemical burden was found in adipose tissue and lowest in the brain. Overall, all studied human fetuses were intrinsically exposed to mixtures of POPs. Tissue:serum ratios were significantly modified by gestational age, fetal sex and placental function. Importantly, more chemicals were detected in fetal tissues compared to maternal serum and placenta, implying that these proxy samples may provide a misleading picture of actual fetal exposures.

LightGBM: An Effective and Scalable Algorithm for Prediction of Chemical Toxicity-Application to the Tox21 and Mutagenicity Data Sets

Machine learning algorithms have attained widespread use in assessing the potential toxicities of pharmaceuticals and industrial chemicals because of their faster speed and lower cost compared to experimental bioassays. Gradient boosting is an effective algorithm that often achieves high predictivity, but historically the relative long computational time limited its applications in predicting large compound libraries or developing in silico predictive models that require frequent retraining. LightGBM, a recent improvement of the gradient boosting algorithm, inherited its high predictivity but resolved its scalability and long computational time by adopting a leaf-wise tree growth strategy and introducing novel techniques. In this study, we compared the predictive performance and the computational time of LightGBM to deep neural networks, random forests, support vector machines, and XGBoost. All algorithms were rigorously evaluated on publicly available Tox21 and mutagenicity data sets using a Bayesian optimization integrated nested 10-fold cross-validation scheme that performs hyperparameter optimization while examining model generalizability and transferability to new data. The evaluation results demonstrated that LightGBM is an effective and highly scalable algorithm offering the best predictive performance while consuming significantly shorter computational time than the other investigated algorithms across all Tox21 and mutagenicity data sets. We recommend LightGBM for applications of in silico safety assessment and also other areas of cheminformatics to fulfill the ever-growing demand for accurate and rapid prediction of various toxicity or activity related end points of large compound libraries present in the pharmaceutical and chemical industry.

Conformal prediction of HDAC inhibitors

The growing interest in epigenetic probes and drug discovery, as revealed by several epigenetic drugs in clinical use or in the lineup of the drug development pipeline, is boosting the generation of screening data. In order to maximize the use of structure-activity relationships there is a clear need to develop robust and accurate models to understand the underlying structure-activity relationship. Similarly, accurate models should be able to guide the rational screening of compound libraries. Herein we introduce a novel approach for epigenetic quantitative structure-activity relationship (QSAR) modelling using conformal prediction. As a case study, we discuss the development of models for 11 sets of inhibitors of histone deacetylases (HDACs), which are one of the major epigenetic target families that have been screened. It was found that all derived models, for every HDAC endpoint and all three significance levels, are valid with respect to predictions for the external test sets as well as the internal validation of the corresponding training sets. Furthermore, the efficiencies for the predictions are above 80% for most data sets and above 90% for four data sets at different significant levels. The findings of this work encourage prospective applications of conformal prediction for other epigenetic target data sets.

Concentrations of perfluoroalkyl substances (PFASs) in human embryonic and fetal organs from first, second, and third trimester pregnancies

BACKGROUND

The persistent environmental contaminants perfluoroalkyl substances (PFASs) have gained attention due to their potential adverse health effects, in particular following early life exposure. Information on human fetal exposure to PFASs is currently limited to one report on first trimester samples. There is no data available on PFAS concentrations in fetal organs throughout all three trimesters of pregnancy.

METHODS

We measured the concentrations of perfluorooctanesulfonic acid (PFOS), perfluorooctanoic acid (PFOA), perfluorononanoic acid (PFNA), perfluorodecanoic acid (PFDA), perfluoroundecanoic acid (PFUnA), and perfluorohexane sulfonic acid (PFHxS) in human embryos and fetuses with corresponding placentas and maternal serum samples derived from elective pregnancy terminations and cases of intrauterine fetal death. A total of 78 embryos and fetuses aged 7-42 gestational weeks were included and a total of 225 fetal organs covering liver, lung, heart, central nervous system (CNS), and adipose tissue were analyzed, together with 71 placentas and 63 maternal serum samples. PFAS concentrations were assayed by liquid chromatography/triple quadrupole mass spectrometry.

RESULTS

All evaluated PFASs were detected and quantified in maternal sera, placentas and embryos/fetuses. In maternal serum samples, PFOS was detected in highest concentrations, followed by PFOA > PFNA > PFDA = PFUnA = PFHxS. Similarly, PFOS was detected in highest concentrations in embryo/fetal tissues, followed by PFOA > PFNA = PFDA = PFUnA. PFHxS was detected in very few fetuses. In general, PFAS concentrations in embryo/fetal tissue (ng/g) were lower than maternal serum (ng/ml) but similar to placenta concentrations. The total PFAS burden (i.e. the sum of all PFASs) was highest in lung tissue in first trimester samples and in liver in second and third trimester samples. The burden was lowest in CNS samples irrespective of fetal age. The placenta:maternal serum ratios of PFOS, PFOA and PFNA increased across gestation suggesting bioaccumulation in the placenta. Further, we observed that the ratios were higher in pregnancies with male fetuses compared to female fetuses.

CONCLUSIONS

Human fetuses were intrinsically exposed to a mixture of PFASs throughout gestation. The compounds were detected in all analyzed tissues, suggesting that PFASs reach and may affect many types of organs. Collectively, our results demonstrate that PFASs pass the placenta and deposit to embryo and fetal tissues, calling for risk assessment of gestational exposures.

Fluoxetine Affects Differentiation of Midbrain Dopaminergic Neurons In Vitro

Recent meta-analyses found an association between prenatal exposure to the antidepressant fluoxetine (FLX) and an increased risk of autism in children. This developmental disorder has been related to dysfunctions in the brains' rewards circuitry, which, in turn, has been linked to dysfunctions in dopaminergic (DA) signaling. The present study investigated if FLX affects processes involved in dopaminergic neuronal differentiation. Mouse neuronal precursors were differentiated into midbrain dopaminergic precursor cells (mDPCs) and concomitantly exposed to clinically relevant doses of FLX. Subsequently, dopaminergic precursors were evaluated for expression of differentiation and stemness markers using quantitative polymerase chain reaction. FLX treatment led to increases in early regional specification markers orthodenticle homeobox 2 (Otx2) and homeobox engrailed-1 and -2 (En1 and En2). On the other hand, two transcription factors essential for midbrain dopaminergic (mDA) neurogenesis, LIM homeobox transcription factor 1 α (Lmx1a) and paired-like homeodomain transcription factor 3 (Pitx3) were downregulated by FLX treatment. The stemness marker nestin (Nes) was increased, whereas the neuronal differentiation marker β3-tubulin (Tubb3) decreased. Additionally, we observed that FLX modulates the expression of several genes associated with autism spectrum disorder and downregulates the estrogen receptors (ERs) α and β Further investigations using ERβ knockout (BERKO) mDPCs showed that FLX had no or even opposite effects on several of the genes analyzed. These findings suggest that FLX affects differentiation of the dopaminergic system by increasing production of dopaminergic precursors, yet decreasing their maturation, partly via interference with the estrogen system.

Phthalate esters on urban airborne particles: Levels in PM10 and PM2.5 from Mexico City and theoretical assessment of lung exposure

Endocrine disrupting chemicals (EDCs) from the environment are associated with reproductive abnormalities (i.e. decreased sperm concentration; increased endometriosis) and alterations of the cardiovascular system (i.e. increased blood pressure and risk of coronary disease). Some phthalates esters have been identified as EDCs, for which inhalation is considered as one of the routes of exposure. However, only little is known regarding inhalational exposure to EDCs via urban airborne particles. In the present study, we report the monthly concentration of 8 phthalate esters measured in PM₁₀ and PM_2.5 collected and recovered during 7 months in a highly populated area of Mexico City. Using the levels of PM₁₀ and PM_2.5 reported by the automatized network of environmental monitoring of Mexico City for the sampling site, we estimated exposure levels for people of different ages and gender. Two endocrine disrupting compounds, the phthalate esters DEHP and DnBP, were found on the particles in higher concentrations during the warmer months of the year. The highest concentration was reported for DEHP (229.7μg/g of particles) in PM_2.5 collected in May 2013. After calculations of the DEHP concentration in the atmosphere, and using the respiratory flow rate, we determined males were potentially exposed to larger quantities of DEHP, reaching up to 18ng/8h in April 2013. Despite the concentrations of phthalates seem to be rather small, a comprehensive characterization of its presence is necessary in order to evaluate the overall exposure to these compounds, providing a clear view of exposure on children, adolescents and pregnant women.

Chemical space, diversity and activity landscape analysis of estrogen receptor binders

Understanding the structure–activity relationships (SAR) of endocrine-disrupting chemicals has a major importance in toxicology. Despite the fact that classifiers and predictive models have been developed for estrogens for the past 20 years, to the best of our knowledge, there are no studies of their activity landscape or the identification of activity cliffs. Herein, we report the first SAR of a public dataset of 121 chemicals with reported estrogen receptor binding affinities using activity landscape modeling. To this end, we conducted a systematic quantitative and visual analysis of the chemical space of the 121 chemicals. The global diversity of the dataset was characterized by means of Consensus Diversity Plot, a recently developed method. Adding pairwise activity difference information to the chemical space gave rise to the activity landscape of the data set uncovering a heterogeneous SAR, in particular for some structural classes. At least eight compounds were identified with high propensity to form activity cliffs. The findings of this work further expand the current knowledge of the underlying SAR of estrogenic compounds and can be the starting point to develop novel and potentially improved predictive models.

Global diversity and activity landscape analysis of endocrine-disrupting chemicals identifies activity cliffs that are rationalized at the structure level.

Half-lives of PFOS, PFHxS and PFOA after end of exposure to contaminated drinking water

Background

Municipal drinking water contaminated with perfluorinated alkyl acids had been distributed to one-third of households in Ronneby, Sweden. The source was firefighting foam used in a nearby airfield since the mid-1980s. Clean water was provided from 16 December 2013.

Objective

To determine the rates of decline in serum perfluorohexane sulfonate (PFHxS), perfluorooctane sulfonate (PFOS) and perfluorooctanoate (PFOA), and their corresponding half-lives.

Methods

Up to seven blood samples were collected between June 2014 and September 2016 from 106 participants (age 4–84 years, 53% female).

Results

Median initial serum concentrations were PFHxS, 277 ng/mL (range 12–1660); PFOS, 345 ng/mL (range 24–1500); and PFOA, 18 ng/mL (range 2.4–92). The covariate-adjusted average rates of decrease in serum were PFHxS, 13% per year (95% CI 12% to 15%); PFOS, 20% per year (95% CI 19% to 22%); and PFOA, 26% per year (95% CI 24% to 28%). The observed data are consistent with a first-order elimination model. The mean estimated half-life was 5.3 years (95% CI 4.6 to 6.0) for PFHxS, 3.4 years (95% CI 3.1 to 3.7) for PFOS and 2.7 years (95% CI 2.5 to 2.9) for PFOA. The interindividual variation of half-life was around threefold when comparing the 5th and 95th percentiles. There was a marked sex difference with more rapid elimination in women for PFHxS and PFOS, but only marginally for PFOA.

Conclusions

The estimated half-life for PFHxS was considerably longer than for PFOS and PFOA. For PFHxS and PFOS, the average half-life is shorter than the previously published estimates. For PFOA the half-life is in line with the range of published estimates.

HLA-DR7 and HLA-DQ2: Transgenic mouse strains tested as a model system for ximelagatran hepatotoxicity

The oral thrombin inhibitor ximelagatran was withdrawn in the late clinical trial phase because it adversely affected the liver. In approximately 8% of treated patients, drug-induced liver injury (DILI) was expressed as transient alanine transaminase (ALT) elevations. No evidence of DILI had been revealed in the pre-clinical in vivo studies. A whole genome scan study performed on the clinical study material identified a strong genetic association between the major histocompatibility complex alleles for human leucocyte antigens (HLA) (HLA-DR7 and HLA-DQ2) and elevated ALT levels in treated patients. An immune-mediated pathogenesis was suggested. Here, we evaluated whether HLA transgenic mice models could be used to investigate whether the expression of relevant HLA molecules was enough to reproduce the DILI effects in humans. In silico modelling performed in this study revealed association of both ximelagatran (pro-drug) and melagatran (active drug) to the antigen-presenting groove of the homology modelled HLA-DR7 molecule suggesting “altered repertoire” as a key initiating event driving development of DILI in humans. Transgenic mouse strains (tgms) expressing HLA of serotype HLA-DR7 (HLA-DRB1*0701, -DRA*0102), and HLA-DQ2 (HLA-DQB1*0202,–DQA1*0201) were created. These two lines were crossed with a human (h)CD4 transgenic line, generating the two tgms DR7xhCD4 and DQ2xhCD4. To investigate whether the DILI effects observed in humans could be reproduced in tgms, the mice were treated for 28 days with ximelagatran. Results revealed no signs of DILI when biomarkers for liver toxicity were measured and histopathology was evaluated. In the ximelagatran case, presence of relevant HLA-expression in a pre-clinical model did not fulfil the prerequisite for reproducing DILI observed in patients. Nonetheless, for the first time an HLA-transgenic mouse model has been investigated for use in HLA-associated DILI induced by a low molecular weight compound. This study shows that mimicking of genetic susceptibility, expressed as DILI-associated HLA-types in mice, is not sufficient for reproducing the complex pathogenesis leading to DILI in man.

Application of shape-based and pharmacophore-based in silico screens for identification of Type II protein kinase inhibitors

The identification of new, potent and selective inhibitors of important protein kinase targets is a major goal of drug discovery. Here we analyze the crystal structures of 55 protein kinase complexes with Type II inhibitors and find they adopt a conserved twisted V-shape, with an angle of 121 ± 8° and twist of 78 ± 8°. The tightly conserved twist appears important in ensuring ligands curve around the protein backbone and towards the deep pocket. From this, we develop predictive pharmacophore- and shape-based screens to identify Type II inhibitors from a database which also contains Type I inhibitors as decoys. Both approaches exhibit a good level of discrimination for Type II molecules. The most effective pharmacophore model requires six features and three excluded volume regions. Shape-based screening using ROCS generally performs at least as well as pharmacophore approaches. There is only a moderate dependence of shape-based or pharmacophore-based screens on the underlying conformer generator (MOE, Macromodel, Omega and SPE), as well as on ligand linkage chemistry (amide and urea). Finally, we apply our approach to retrieval of Type II inhibitors from a modified version of the DUD database, containing over 104,000 compounds. We observe good enrichment, providing further evidence that the in silico screens developed here will constitute useful guides for identification of small molecule inhibitors targetting protein kinases in their inactive conformational state.