Pines' demon observed as a 3D acoustic plasmon in Sr2RuO4.

The characteristic excitation of a metal is its plasmon, which is a quantized collective oscillation of its electron density. In 1956, David Pines predicted that a distinct type of plasmon, dubbed a 'demon', could exist in three-dimensional (3D) metals containing more than one species of charge carrier1. Consisting of out-of-phase movement of electrons in different bands, demons are acoustic, electrically neutral and do not couple to light, so have never been detected in an equilibrium, 3D metal. Nevertheless, demons are believed to be critical for diverse phenomena including phase transitions in mixed-valence semimetals2, optical properties of metal nanoparticles3, soundarons in Weyl semimetals4 and high-temperature superconductivity in, for example, metal hydrides3,5-7. Here, we present evidence for a demon in Sr2RuO4 from momentum-resolved electron energy-loss spectroscopy. Formed of electrons in the ß and γ bands, the demon is gapless with critical momentum qc = 0.08 reciprocal lattice units and room-temperature velocity v = (1.065 ± 0.12) × 105 m s-1 that undergoes a 31% renormalization upon cooling to 30 K because of coupling to the particle-hole continuum. The momentum dependence of the intensity of the demon confirms its neutral character. Our study confirms a 67-year old prediction and indicates that demons may be a pervasive feature of multiband metals.

Neofunctionalization of an OMT cluster dominates polymethoxyflavone biosynthesis associated with the domestication of citrus.

Polymethoxyflavones (PMFs) are a class of abundant specialized metabolites with remarkable anticancer properties in citrus. Multiple methoxy groups in PMFs are derived from methylation modification catalyzed by a series of hydroxylases and O-methyltransferases (OMTs). However, the specific OMTs that catalyze the systematic O-methylation of hydroxyflavones remain largely unknown. Here, we report that PMFs are highly accumulated in wild mandarins and mandarin-derived accessions, while undetectable in early-diverging citrus species and related species. Our results demonstrated that three homologous genes, CreOMT3, CreOMT4, and CreOMT5, are crucial for PMF biosynthesis in citrus, and their encoded methyltransferases exhibit multisite O-methylation activities for hydroxyflavones, producing seven PMFs in vitro and in vivo. Comparative genomic and syntenic analyses indicated that the tandem CreOMT3, CreOMT4, and CreOMT5 may be duplicated from CreOMT6 and contributes to the genetic basis of PMF biosynthesis in the mandarin group through neofunctionalization. We also demonstrated that N17 in CreOMT4 is an essential amino acid residue for C3-, C5-, C6-, and C3'-O-methylation activity and provided a rationale for the functional deficiency of OMT6 to produce PMFs in early-diverging citrus and some domesticated citrus species. A 1,041-bp deletion in the CreOMT4 promoter, which is found in most modern cultivated mandarins, has reduced the PMF content relative to that in wild and early-admixture mandarins. This study provides a framework for reconstructing PMF biosynthetic pathways, which may facilitate the breeding of citrus fruits with enhanced health benefits.

Computational approaches identify a transcriptomic fingerprint of drug-induced structural cardiotoxicity.

Structural cardiotoxicity (SCT) presents a high-impact risk that is poorly tolerated in drug discovery unless significant benefit is anticipated. Therefore, we aimed to improve the mechanistic understanding of SCT. First, we combined machine learning methods with a modified calcium transient assay in human-induced pluripotent stem cell-derived cardiomyocytes to identify nine parameters that could predict SCT. Next, we applied transcriptomic profiling to human cardiac microtissues exposed to structural and non-structural cardiotoxins. Fifty-two genes expressed across the three main cell types in the heart (cardiomyocytes, endothelial cells, and fibroblasts) were prioritised in differential expression and network clustering analyses and could be linked to known mechanisms of SCT. This transcriptomic fingerprint may prove useful for generating strategies to mitigate SCT risk in early drug discovery.

Tandem Proton Transfer in Carboxylated Supramolecular Polymer for Highly Efficient Overall Photosynthesis of Hydrogen Peroxide.

Proton supply is as critical as O2 activation for artificial photosynthesis of H2O2 via two-electron oxygen reduction reaction (2e- ORR). However, proton release via water dissociation is frequently hindered because of the sluggish water oxidation reaction (WOR), extremely limiting the efficiency of photocatalytic H2O2 production. To tackle this challenge, carboxyl-enriched supramolecular polymer (perylene tetracarboxylic acid - PTCA) is elaborately prepared by molecular self-assembly for overall photosynthesis of H2O2. Interestingly, the interconversion between carboxyl as Brønsted acid and its conjugated base realizes rapid proton circulation. Through this efficient tandem proton transfer process, the spatial effect of photocatalytic reduction and oxidation reaction is greatly enhanced with reduced reaction barrier. This significantly facilitates 2e- photocatalytic ORR to synthesize H2O2 and in the meanwhile promotes 4e- photocatalytic WOR to evolve O2. Consequently, the as-developed PTCA exhibits a remarkable H2O2 yield of 185.6 µM h-1 in pure water and air atmosphere under visible light illumination. More impressively, an appreciable H2O2 yield of 78.6 µM h-1 can be well maintained in an anaerobic system owing to in-situ O2 generation by 4e- photocatalytic WOR. Our study presents a novel concept for artificial photosynthesis of H2O2 via constructing efficient proton transfer pathway to enable rapid proton circulation.

E2F1 proteolysis via SCF-cyclin F underlies synthetic lethality between cyclin F loss and Chk1 inhibition.

Cyclins are central engines of cell cycle progression in conjunction with cyclin-dependent kinases (CDKs). Among the different cyclins controlling cell cycle progression, cyclin F does not partner with a CDK, but instead forms via its F-box domain an SCF (Skp1-Cul1-F-box)-type E3 ubiquitin ligase module. Although various substrates of cyclin F have been identified, the vulnerabilities of cells lacking cyclin F are not known. Thus, we assessed viability of cells lacking cyclin F upon challenging them with more than 180 different kinase inhibitors. The screen revealed a striking synthetic lethality between Chk1 inhibition and cyclin F loss. Chk1 inhibition in cells lacking cyclin F leads to DNA replication catastrophe. Replication catastrophe depends on accumulation of the transcription factor E2F1 in cyclin F-depleted cells. We find that SCF-cyclin F controls E2F1 ubiquitylation and degradation during the G2/M phase of the cell cycle and upon challenging cells with Chk1 inhibitors. Thus, Cyclin F restricts E2F1 activity during the cell cycle and upon checkpoint inhibition to prevent DNA replication stress. Our findings pave the way for patient selection in the clinical use of checkpoint inhibitors.

Gastric Peroral Endoscopic Pyloromyotomy for Infants With Congenital Hypertrophic Pyloric Stenosis.

INTRODUCTION: Congenital hypertrophic pyloric stenosis (CHPS), the most common infantile disease requiring surgical intervention, is routinely treated with open or laparoscopic pyloromyotomy. Recently, gastric peroral endoscopic pyloromyotomy (G-POEM) has been used for adult gastroparesis. We aimed to evaluate the efficacy and safety of G-POEM in treating infantile CHPS. METHODS: We reviewed data from 21 G-POEM-treated patients at 3 tertiary children's endoscopic centers in China between January 2019 and December 2020. Clinical characteristics, procedure-related parameters, perioperative management, and follow-up outcomes were summarized. RESULTS: G-POEM was performed successfully in all patients. The median operative duration was 49 (14-150) minutes. The submucosal tunnels were successfully established along the greater curvature of the stomach in 19 cases, and 2 cases were switched to the lesser curvature because of difficulty. No perioperative major adverse events occurred. Minor adverse events included inconsequential mucosal injury in 5 cases and unsatisfactory closure of the mucosal incision in 1 case. Upper gastrointestinal contrast radiography in all patients showed smooth passage of the contrast agent through the pylorus on postoperative day 3. The growth curves of the patients reached normal levels 3 months after the procedure. No recurrent clinical symptoms occurred in any patient during the median follow-up period of 25.5 (14-36) months. DISCUSSION: G-POEM is feasible, safe, and effective for infants with CHPS, with satisfactory clinical responses over a short-term follow-up. Further multicenter studies should be performed to compare the long-term outcomes of this minimally invasive technique with open or laparoscopic pyloromyotomy.

Boosting Photo-Electro-Fenton Process Via Atomically Dispersed Iron Sites on Graphdiyne for InVitro Hydrogen Peroxide Detection.

Hydrogen peroxide (H2 O2 ) is essential in oxidative stress and signal regulation of organs of animal body. Realizing in vitro quantification of H2 O2 released from organs is significant, but faces challenges due to short lifetime of H2 O2 and complex bio-environment. Herein, rationally designed and constructed a photoelectrochemical (PEC) sensor for in vitro sensing of H2 O2 , in which atomically dispersed iron active sites (Hemin) modified graphdiyne (Fe-GDY) serves as photoelectrode and catalyzes photo-electro-Fenton process. Sensitivity of Fe-GDY electrode is enhanced 8 times under illumination compared with dark condition. The PEC H2 O2 sensor under illumination delivers a wide linear range from 0.1 to 48 160 µm and a low detection limit of 33 nm, while demonstrating excellent selectivity and stability. The high performance of Fe-GDY is attributed to, first, energy levels matching of GDY and Hemin that effectively promotes the injection of photo-generated electrons from GDY to Fe3+ for reduced Fe2+ , which facilitates the Fe3+ /Fe2+ cycle. Second, the Fe2+ actively catalyzes H2 O2 to OH- through the Fenton process, thereby improving the sensitivity. The PEC sensor demonstrates in vitro quantification of H2 O2 released from different organs, providing a promising approach for molecular sensing and disease diagnosis in organ levels.

Network-based modeling of herb combinations in traditional Chinese medicine.

Traditional Chinese medicine (TCM) has been practiced for thousands of years for treating human diseases. In comparison to modern medicine, one of the advantages of TCM is the principle of herb compatibility, known as TCM formulae. A TCM formula usually consists of multiple herbs to achieve the maximum treatment effects, where their interactions are believed to elicit the therapeutic effects. Despite being a fundamental component of TCM, the rationale of combining specific herb combinations remains unclear. In this study, we proposed a network-based method to quantify the interactions in herb pairs. We constructed a protein-protein interaction network for a given herb pair by retrieving the associated ingredients and protein targets, and determined multiple network-based distances including the closest, shortest, center, kernel, and separation, both at the ingredient and at the target levels. We found that the frequently used herb pairs tend to have shorter distances compared to random herb pairs, suggesting that a therapeutic herb pair is more likely to affect neighboring proteins in the human interactome. Furthermore, we found that the center distance determined at the ingredient level improves the discrimination of top-frequent herb pairs from random herb pairs, suggesting the rationale of considering the topologically important ingredients for inferring the mechanisms of action of TCM. Taken together, we have provided a network pharmacology framework to quantify the degree of herb interactions, which shall help explore the space of herb combinations more effectively to identify the synergistic compound interactions based on network topology.

Prediction of inotropic effect based on calcium transients in human iPSC-derived cardiomyocytes and machine learning.

Functional changes to cardiomyocytes are undesirable during drug discovery and identifying the inotropic effects of compounds is hence necessary to decrease the risk of cardiovascular adverse effects in the clinic. Recently, approaches leveraging calcium transients in human induced pluripotent stem cell-derived cardiomyocytes (hiPSC-CMs) have been developed to detect contractility changes, induced by a variety of mechanisms early during drug discovery projects. Although these approaches have been able to provide some predictive ability, we hypothesised that using additional waveform parameters could offer improved insights, as well as predictivity. In this study, we derived 25 parameters from each calcium transient waveform and developed a modified Random Forest method to predict the inotropic effects of the compounds. In total annotated data for 48 compounds were available for modelling, out of which 31 were inotropes. The results show that the Random Forest model with a modified purity criterion performed slightly better than an unmodified algorithm in terms of the Area Under the Curve, giving values of 0.84 vs 0.81 in a cross-validation, and outperformed the ToxCast Pipeline model, for which the highest value was 0.76 when using the best-performing parameter, PW10. Our study hence demonstrates that more advanced parameters derived from waveforms, in combination with additional machine learning methods, provide improved predictivity of cardiovascular risk associated with inotropic effects.

A short period of early life oxytocin treatment rescues social behavior dysfunction via suppression of hippocampal hyperactivity in male mice.

Early sensory experiences interact with genes to shape precise neural circuits during development. This process is vital for proper brain function in adulthood. Neurological dysfunctions caused by environmental alterations and/or genetic mutation may share the same molecular or cellular mechanisms. Here, we show that early life bilateral whisker trimming (BWT) subsequently affects social discrimination in adult male mice. Enhanced activation of the hippocampal dorsal CA3 (dCA3) in BWT mice was observed during social preference tests. Optogenetic activation of dCA3 in naive mice impaired social discrimination, whereas chemogenetic silencing of dCA3 rescued social discrimination deficit in BWT mice. Hippocampal oxytocin (OXT) is reduced after whisker trimming. Neonatal intraventricular compensation of OXT relieved dCA3 over-activation and prevented social dysfunction. Neonatal knockdown of OXT receptor in dCA3 mimics the effects of BWT, and cannot be rescued by OXT treatment. Social behavior deficits in a fragile X syndrome mouse model (Fmr1 KO mice) could also be recovered by early life OXT treatment, through negating dCA3 over-activation. Here, a possible avenue to prevent social dysfunction is uncovered.

A head-to-head comparison of the measurement properties of EQ-5D-3L and EQ-5D-5L in Chinese family caregivers of cancer patients.

BACKGROUND: Although both EQ-5D-3L(3L) and EQ-5D-5L(5L) have demonstrated good measurement properties in several patient populations, there is currently limited evidence comparing the measurement properties of 3L and 5L in family caregivers (FCs) of cancer patients. PURPOSE: This study aimed to compare the measurement properties of 3L and 5L in a sample of family caregivers of cancer patients. METHODS: A consecutive sample of FCs of cancer patients recruited from three tertiary hospitals were invited to complete the two versions of the EQ-5D in two rounds of interviews. We compared i) the ceiling effect using the McNemar's test, ii) test-retest reliability using intraclass correlation coefficient (ICC) and Cohen's Kappa, iii) convergent validity using Spearman's rank correlation coefficient, iv) known-group validity using F-statistic, v) and discriminant capacity using ordinal logistic regression. RESULTS: A total of 416 FCs completed the baseline questionnaire and 120 caregivers completed the follow-up questionnaire. Ceiling effects were smaller in 5L (12.5%) than in 3L (20.7%). The convergent validity (r = 0.344-0.771), known-groups validity (Fratio5L/3L = 2.06-4.09), discriminant capacity (ES = 0.341-0.396), and test-retest reliability (ICC = 0.725) of the 5L were slightly better than those of the 3L in China. CONCLUSION: The current study found both 3L and 5L to be suitable for use by FCs of cancer patients. However, 5L showed superior measurement properties compared to 3L and therefore could be the preferred instrument when EQ-5D data of cancer patients FCs is required.

Halogen-Incorporated Sn Catalysts for Selective Electrochemical CO2 Reduction to Formate.

Electrochemically reducing CO2 to valuable fuels or feedstocks is recognized as a promising strategy to simultaneously tackle the crises of fossil fuel shortage and carbon emission. Sn-based catalysts have been widely studied for electrochemical CO2 reduction reaction (CO2 RR) to make formic acid/formate, which unfortunately still suffer from low activity, selectivity and stability. In this work, halogen (F, Cl, Br or I) was introduced into the Sn catalyst by a facile hydrolysis method. The presence of halogen was confirmed by a collection of ex situ and in situ characterizations, which rendered a more positive valence state of Sn in halogen-incorporated Sn catalyst as compared to unmodified Sn under cathodic potentials in CO2 RR and therefore tuned the adsorption strength of the key intermediate (*OCHO) toward formate formation. As a result, the halogen-incorporated Sn catalyst exhibited greatly enhanced catalytic performance in electrochemical CO2 RR to produce formate.

Single-Atom Iron Anchored on 2-D Graphene Carbon to Realize Bridge-Adsorption of O-O as Biomimetic Enzyme for Remarkably Sensitive Electrochemical Detection of H2O2.

Single-atom catalysis is mainly focused on its dispersed high-density catalytic sites, but delicate designs to realize a unique catalysis mechanism in terms of target reactions have been much less investigated. Herein an iron single atomic site catalyst anchored on 2-D N-doping graphene (Fe-SASC/G) was synthesized and further employed as a biomimetic sensor to electrochemically detect hydrogen peroxide, showing an extremely high sensitivity of 3214.28 µA mM-1 cm-2, which is much higher than that (6.5 µA mM-1 cm-2) of its dispersed on 1-D carbon nanowires (Fe-SASC/NW), ranking the best sensitivity among all reported Fe based catalyst at present. The sensor was also used to successfully in situ monitor H2O2 released from A549 living cells. The mechanism was further systematically investigated. Results interestingly indicate that the distance between adjacent single Fe atomic catalytic sites on 2-D graphene of Fe-SASC/G matches statistically well with the outer length of bioxygen of H2O2 to promote a bridge adsorption of -O-O- for simultaneous 2-electron transfer, while the single Fe atoms anchored on distant 1-D nanowires in Fe-SASC/NW only allow an end-adsorption of oxygen atoms for 1-electron transfer. These results demonstrate that Fe-SASC/G holds great promise as an advanced electrode material in selective and sensitive biomimetic sensor and other electrocatalytic applications, while offering scientific insights in deeper single atomic catalysis mechanisms, especially the effects of substrate dimensions on the mechanism.

Low-Loss Tunable Infrared Plasmons in the High-Mobility Perovskite (Ba,La)SnO3.

BaSnO3 exhibits the highest carrier mobility among perovskite oxides, making it ideal for oxide electronics. Collective charge carrier oscillations known as plasmons are expected to arise in this material, thus providing a tool to control the nanoscale optical field for optoelectronics applications. Here, the existence of relatively long-lived plasmons supported by high-mobility charge carriers in La-doped BaSnO3 (BLSO) is demonstrated. By exploiting the high spatial and energy resolution of electron energy-loss spectroscopy with a focused beam in a scanning transmission electron microscope, the dispersion, confinement ratio, and damping of infrared localized surface plasmons (LSPs) in BLSO nanoparticles are systematically investigated. It is found that LSPs in BLSO exhibit a high degree of spatial confinement compared to those sustained by noble metals and have relatively low losses and high quality factors with respect to other doped oxides. Further analysis clarifies the relation between plasmon damping and carrier mobility in BLSO. The results support the use of nanostructured degenerate semiconductors for plasmonic applications in the infrared region and establish a solid alternative to more traditional plasmonic materials.

Bit-error rate and average capacity of an absorbent and turbulent underwater wireless communication link with perfect Laguerre-Gauss beam.

The proper selection of signal source structure and parameters is one of the effective methods to suppress the random disturbance of underwater channel and enhance the performance of link communication. In this paper, the bit error probability and average capacity of a weakly turbulent absorbing seawater link with a perfect Laguerre-Gaussian beam are studied. The bit error rate of orbital angular momentum (OAM) channel under on-off key modulation is derived. The average capacity model of the optical wireless communication link is established on the basis that the OAM channel formed by vortex carrier of topological charge m is symmetric channel. The relationship between OAM channel capacity, carrier wavelength and seawater absorption is numerically analyzed by seawater spectral absorption coefficient, and it is concluded that in the range of "window transmission wavelength" and long channel, the conclusion of the longer signal source wavelength is beneficial to the performance of non-absorptive turbulent channel is no longer valid. Some other numerical results are worked out to show that the underwater communication link with perfect Laguerre-Gaussian beam can obtain high channel capacity by adopting low OAM topological charge, smaller aperture of transmitter as well as receiver.

CsMYB96 confers resistance to water loss in citrus fruit by simultaneous regulation of water transport and wax biosynthesis.

A Citrus sinensis R2R3 MYB transcription factor (CsMYB96) has previously been shown to be strongly associated with the expression of many genes related to wax biosynthesis in the fruit. In this study, CsMYB96 was found to alleviate water loss by simultaneously regulating the expression of genes encoding plasma membrane intrinsic proteins (CsPIPs) and wax-related genes. Expression profiling indicated that CsPIP1;1 and CsPIP2;4 had high expression that was representative of other aquaporins, and they were down-regulated in the peel of post-harvest citrus fruit. CsPIP2;4 was further characterized as the predominant CsPIP, with high expression and high-water channel activity. Transient overexpression of CsPIP2;4 accelerated water loss in citrus fruit. In silico analysis further indicated that the expression of CsMYB96 had a significant negative correlation with that of CsPIPs. In vivo and in vitro experiments confirmed that CsMYB96 was able to directly repress the expression of CsPIPs. In addition, CsMYB96 was able to activate wax-related genes and promote wax biosynthesis for defense against water loss. Transient and stable overexpression of CsMYB96 reduced water loss from both citrus fruit and Arabidopsis.

IDL-PPBopt: A Strategy for Prediction and Optimization of Human Plasma Protein Binding of Compounds via an Interpretable Deep Learning Method.

The prediction and optimization of pharmacokinetic properties are essential in lead optimization. Traditional strategies mainly depend on the empirical chemical rules from medicinal chemists. However, with the rising amount of data, it is getting more difficult to manually extract useful medicinal chemistry knowledge. To this end, we introduced IDL-PPBopt, a computational strategy for predicting and optimizing the plasma protein binding (PPB) property based on an interpretable deep learning method. At first, a curated PPB data set was used to construct an interpretable deep learning model, which showed excellent predictive performance with a root mean squared error of 0.112 for the entire test set. Then, we designed a detection protocol based on the model and Wilcoxon test to identify the PPB-related substructures (named privileged substructures, PSubs) for each molecule. In total, 22 general privileged substructures (GPSubs) were identified, which shared some common features such as nitrogen-containing groups, diamines with two carbon units, and azetidine. Furthermore, a series of second-level chemical rules for each GPSub were derived through a statistical test and then summarized into substructure pairs. We demonstrated that these substructure pairs were equally applicable outside the training set and accordingly customized the structural modification schemes for each GPSub, which provided alternatives for the optimization of the PPB property. Therefore, IDL-PPBopt provides a promising scheme for the prediction and optimization of the PPB property and would be helpful for lead optimization of other pharmacokinetic properties.

CitWRKY28 and CitNAC029 promote the synthesis of cuticular wax by activating CitKCS gene expression in citrus fruit.

KEY MESSAGE: CitWRKY28 and CitNAC029 are involved in cuticular wax synthesis as indicated by the comparative analysis of fruit aliphatic wax content between Citrus reticulata and Citrus trifoliata and gene co-expression analysis. Cuticular wax covers the fruit surface, playing important roles in reduction of fruit water loss and resistance to pathogen invasion. However, there is limited research on the synthesis and transcriptional regulation of cuticular wax in citrus fruit. In this study, we characterized the variations of aliphatic wax in HJ (Citrus reticulata) and ZK (Citrus trifoliata) from young fruit to mature fruit, as well as performed transcriptome sequencing on 27 samples at different fruit developmental stages. The results revealed that the ZK fruit always had a higher aliphatic wax content than the HJ fruit during development. qRT-PCR analysis demonstrated that two KCS genes, CitKCS1 and CitKCS12, had the most significant difference in expression between HJ and ZK. Furthermore, a heterologous expression assay in Arabidopsis indicated that CitKCS1 and CitKCS12 are involved in cuticular wax synthesis. Subsequently, gene co-expression network analysis screened CitWRKY28 and CitNAC029. Dual luciferase and EMSA assays indicated that CitWRKY28 might bind to the promoter of CitKCS1 and CitKCS12 and CitNAC029 might bind to that of CitKCS1 to activate their expression. Moreover, CitWRKY28 and CitNAC029 could promote the accumulation of cuticular wax in Arabidopsis leaves. Our findings provide new insights into the synthesis and regulation of cuticular wax and valuable information for further mining of wax-related genes in citrus fruit.

Wander and spread of a perfect Laguerre-Gauss beam under turbulent absorbent seawater.

The wander and long-term spread of a beam caused by turbulence are two important factors affecting channel targeting and information receiving in optics communication systems. In this paper, the wander and long-term spread of a perfect Laguerre-Gauss (PLG)/circular perfect Laguerre-Gauss (CPLG) beam in turbulent absorbing seawater are studied. The analytical expression of the wander for a CPLG beam in the weak turbulent fluctuation region and the analytical expression of the long-term spread for a CPLG beam in a weak to strong turbulent fluctuation region are derived by using the Rytov approximation and the generalized Huygens-Fresnel integral, respectively. Through numerical analysis, we find that the optimal beam diameter and self-convergence effect of the PLG beam exist under given communication link conditions, the long-term spread of PLG beam is smaller than that of the LG beam, but the wander evolution trend of the PLG beam with increasing propagation distance is opposite to that of the LG beam. PLG and CPLG beams have stronger resistance disturbance of turbulence than that of Laguerre-Gauss and circular Laguerre-Gauss beams, respectively.

In silico prediction of chemical aquatic toxicity by multiple machine learning and deep learning approaches.

Fish is one of the model animals used to evaluate the adverse effects of a chemical exposed to the ecosystem. However, its low throughput and relevantly high expense make it impossible to test all new chemicals in manufacture. Hence, using in silico models to prioritize compounds to be tested has been widely applied in environmental risk assessment and drug discovery. In this study, we constructed the local predictive models for four fish species, including bluegill sunfish, rainbow trout, fathead minnow, and sheepshead minnow, and the global models with all four fish data. A total of 1874 unique compounds with their labels, that is, toxic (LC50 < 10 ppm) or nontoxic, were collected from ECOTOX and literature. Both conventional machine learning methods and the deep learning architecture, graph convolutional network (GCN), were used to build predictive models. The classification accuracy of the best local model for each fish species was higher than 0.83. For the global models, two strategies including consistency prediction and probability threshold were adopted to improve the predictive capability at the cost of limiting applicability domain. For 63% of compounds in domain, the accuracy was around 0.97. By comparison of the deep learning and machine learning methods, we found that the single-task GCN showed specific advantages in performance, and multitask GCN showed no advantages over the conventional machine learning methods. The data and models are available on GitHub (https://github.com/ChemPredict/ChemicalAquaticToxicity).