Pharmacometabolomics in Endogenous Drugs: A New Approach for Predicting the Individualized Pharmacokinetics of Cholic Acid.

The evaluation of individual variability in endogenous drugs' metabolism and disposition is a very challenging task. We developed and validated a metabotype to pharmacokinetics (PK) matching approach by taking cholic acid as an example to predict the individualized PK of endogenous drugs. The stable isotope-labeled cholic acid was selected as the substitute analyte of cholic acid to ensure the accurate measurement of blood concentration. First, large-scale metabolite profiling studies were performed on the predose urine samples of 28 rats. Then, to examine the individualized PK of deuterium 4-cholic acid (d4-cholic acid) in these rats, we determined its plasma concentrations and calculated the differential AUC values. Subsequently, we conducted a two-stage partial least-squares analysis in which 31 baseline metabolites were screened initially for predicting the individualized AUC values of d4-cholic acid using the data of predose urine metabolites. Finally, network biology analysis was applied to give the biological interpretation of the individual variances in cholic acid metabolism and disposition, and the result further narrowed the selection of baseline metabolites from 31 to 2 (sarcosine and S-adenosyl-l-homocysteine) for such prediction. Collectively, this pharmacometabolomics research provided a new strategy for predicting individualized PK of endogenous drugs.

New IDH1 mutant inhibitors for treatment of acute myeloid leukemia.

Neomorphic mutations in isocitrate dehydrogenase 1 (IDH1) are driver mutations in acute myeloid leukemia (AML) and other cancers. We report the development of new allosteric inhibitors of mutant IDH1. Crystallographic and biochemical results demonstrated that compounds of this chemical series bind to an allosteric site and lock the enzyme in a catalytically inactive conformation, thereby enabling inhibition of different clinically relevant IDH1 mutants. Treatment of IDH1 mutant primary AML cells uniformly led to a decrease in intracellular 2-HG, abrogation of the myeloid differentiation block and induction of granulocytic differentiation at the level of leukemic blasts and more immature stem-like cells, in vitro and in vivo. Molecularly, treatment with the inhibitors led to a reversal of the DNA cytosine hypermethylation patterns caused by mutant IDH1 in the cells of individuals with AML. Our study provides proof of concept for the molecular and biological activity of novel allosteric inhibitors for targeting different mutant forms of IDH1 in leukemia.

Characterization of the potential new phthalides in Ligusticum chuanxiong Hort. using ultra-performance liquid chromatography coupled with quadrupole time of flight tandem mass spectrometry.

The characterization of unknown compounds is still a great challenge currently. A strategy for deduction of potential new phthalides through the characterization of isomers based on ultra-performance liquid chromatography coupled with quadrupole time of flight tandem mass spectrometry was proposed here to characterize the unknown compounds of Ligusticum chuanxiong Hort. (Chuanxiong). This proposed strategy consisted of four steps: (1) the high resolution MS data was collected, and the peaks were screened preliminarily by UNIFITM platform based on the in-house database; (2) the fragmentation patterns and the characteristic fragments were summarized based on the representative standards; (3) the target compounds were identified based on the fragmentation rules, standards comparison and false positive exclusion; (4) the unknown components were structurally characterized according to the accurate mass and fragmentation patterns analysis. This strategy was successfully applied to the identification and deduction of phthalides in Chuanxiong. A total of 81 phthalides were detected. Fifty-five known phthalides were identified, and 26 potential new phthalides were characterized. This research enriched the material basis of Chuanxiong, and provided a liquid chromatography tandem mass spectrometry-oriented method for the discovery of the potential new compounds.

Applying characteristic fragment filtering for rapid detection and identification of ingredients in rhubarb by HPLC coupled with linear ion trap-Orbitrap mass spectrometry.

Chemical characteristic fragment filtering in MSn chromatograms was proposed to detect and identify the components in rhubarb rapidly using high-performance liquid chromatography coupled with linear ion trap-Orbitrap mass spectrometry. Characteristic fragments consist of diagnostic ions and neutral loss fragments. Characteristic fragment filtering is a postacquisition data mining method for the targeted screening of groups with specific structures, including three steps: first, in order to comprehensively summarize characteristic fragments for global identification of the ingredients in rhubarb, representative authentic standards of dominant chemical categories contained in rhubarb were chosen, from which fragmentation rules and a characteristic fragments schedule were proposed; second, characteristic fragment filtering was used to rapidly recognize analogous skeletons; finally, combined with retention time, accurate mass, characteristic fragments, and previous literature, the structures of the filtered compounds were identified or tentatively characterized. As a result, a total of 271 compounds were detected and identified in rhubarb, including 34 anthraquinones, 83 anthrones, 46 tannins, 17 stilbenes, 24 phenylbutanones, 26 acylglucosides, 26 chromones, and 15 other compounds, 69 of which are potentially new compounds. The proposed characteristic fragment filtering strategy would be a reference for the large-scale detection and identification of the ingredients of herbal medicines.

[Rapid analysis on phenolic compounds in Rheum palmatum based on UPLC-Q-TOF/MSE combined with diagnostic ions filter].

Diagnostic ions filter method was used to rapidly detect and identify the phenolic compounds in Rheum palmatum based on ultra performance liquid chromatography coupled with quadrupole time-of-flight mass spectrometry (UPLC-Q-TOF/MSE). The representative authentic standards of phenolic compounds, including gallic acid, (+)-catechin, (－)-epicatechin, (－)-epicatechin-3-O-gallate and procyanidin B2, were subjected to analysis by UPLC-Q-TOF/MSE system with negative ion mode. Fragmentation patterns of each standard were summarized based on assigned fragment ions. The prominent product ions were selected as diagnostic ions. Subsequently, diagnostic ions filter was employed to rapidly recognize analogous skeletons. Combined with retention time, accurate mass, characteristic fragments and previous literature data, the structures of the filtered compounds were identified or tentatively characterized. A total 63 phenolic compounds (36 phenolic acid derivatives, 8 flavonoid derivatives and 19 tennis derivatives) in R. palmatum were identified, including 6 potential new compounds. The method of diagnostic ions filter could rapidly detect and identify phenolic compounds in R. palmatum This study provides a method for rapid detection of phenolic compounds in R. palmatum and is expected to complete the material basis of rhubarb.

Delipidation-based solid-phase extraction pretreatment technique for plasma broad-coverage metabolomic profiling to reveal the potential pathogenesis of yeast-induced fever in rats.

During the process of metabolomics profiling by using ultra high performance liquid chromatography coupled with time-of flight mass spectrometry, blood sample pretreatment is a crucial step for biomarker discovery. Herein, in order to prevent the potential loss of metabolites and ion suppression phenomena caused by the proteins and phospholipids contained in blood fluids, a delipidation-based solid-phase extraction pretreatment technique for plasma broad-coverage metabolomic profiling was performed. This technique can be summarized as a single extraction, a single elution of solid-phase extraction plate, followed by four times measuring with electrospray ionization in positive and negative ion mode, respectively. This approach significantly increased the number of features detected in plasma, and 1572 features in positive mode and 1352 features in negative mode were detected, respectively. Besides, the stability and repeatability of the approach were greatly improved. For these advantages, the approach was employed to elucidate the potential pathogenesis of yeast-induced fever in rats. The biomarkers associated with the pathogenesis of fever were shown to be related to amino acids metabolism and lipid metabolism. The delipidation-based solid-phase extraction pretreatment approach can provide a useful tool to reveal the pathological mechanisms of such systemic pathological process.

Mutant IDH1 enhances the production of 2-hydroxyglutarate due to its kinetic mechanism.

The human, cytosolic enzyme isocitrate dehydrogenase 1 (IDH1) reversibly converts isocitrate to α-ketoglutarate (αKG). Cancer-associated somatic mutations in IDH1 result in a loss of this normal function but a gain in a new or neomorphic ability to convert αKG to the oncometabolite 2-hydroxyglutarate (2HG). To improve our understanding of the basis for this phenomenon, we have conducted a detailed kinetic study of wild-type IDH1 as well as the known 2HG-producing clinical R132H and G97D mutants and mechanistic Y139D and (newly described) G97N mutants. In the reductive direction of the normal reaction (αKG to isocitrate), dead-end inhibition studies suggest that wild-type IDH1 goes through a random sequential mechanism, similar to previous reports on related mammalian IDH enzymes. However, analogous experiments studying the reductive neomorphic reaction (αKG to 2HG) with the mutant forms of IDH1 are more consistent with an ordered sequential mechanism, with NADPH binding before αKG. This result was further confirmed by primary kinetic isotope effects for which saturating with αKG greatly reduced the observed isotope effect on (D)(V/K)NADPH. For the mutant IDH1 enzyme, the change in mechanism was consistently associated with reduced efficiencies in the use of αKG as a substrate and enhanced efficiencies using NADPH as a substrate. We propose that the sum of these kinetic changes allows the mutant IDH1 enzymes to reductively trap αKG directly into 2HG, rather than allowing it to react with carbon dioxide and form isocitrate, as occurs in the wild-type enzyme.

[Retracted] MicroRNA­599 targets high­mobility group AT­hook 2 to inhibit cell proliferation and invasion in clear cell renal carcinoma.

Following the publication of the above paper, it was drawn to the Editors' attention by a concerned reader that cell migration and invasion assay data shown in Fig. 5C were strikingly similar to data appearing in different form in other articles by different authors, which have been retracted. Owing to the fact that the contentious data in the above article were already under consideration for publication, or had already been published, elsewhere when it was submitted to Molecular Medicine Reports, the Editor has decided that this paper should be retracted from the Journal. The authors were asked for an explanation to account for these concerns, but the Editorial Office did not receive any reply. The Editor apologizes to the readership for any inconvenience caused. [Molecular Medicine Reports 17: 7451­7459, 2018; DOI: 10.3892/mmr.2018.8755].

On the catalytic mechanism of human ATP citrate lyase.

ATP citrate lyase (ACL) catalyzes an ATP-dependent biosynthetic reaction which produces acetyl-coenzyme A and oxaloacetate from citrate and coenzyme A (CoA). Studies were performed with recombinant human ACL to ascertain the nature of the catalytic phosphorylation that initiates the ACL reaction and the identity of the active site residues involved. Inactivation of ACL by treatment with diethylpyrocarbonate suggested the catalytic role of an active site histidine (i.e., His760), which was proposed to form a phosphohistidine species during catalysis. The pH-dependence of the pre-steady-state phosphorylation of ACL with [γ-(33)P]-ATP revealed an ionizable group with a pK(a) value of ~7.5, which must be unprotonated for the catalytic phosphorylation of ACL to occur. Mutagenesis of His760 to an alanine results in inactivation of the biosynthetic reaction of ACL, in good agreement with the involvement of a catalytic histidine. The nature of the formation of the phospho-ACL was further investigated by positional isotope exchange using [γ-(18)O(4)]-ATP. The ß,γ-bridge to nonbridge positional isotope exchange rate of [γ-(18)O(4)]-ATP achieved its maximal rate of 14 s(-1) in the absence of citrate and CoA. This rate decreased to 5 s(-1) when citrate was added, and was found to be 10 s(-1) when both citrate and CoA were present. The rapid positional isotope exchange rates indicated the presence of one or more catalytically relevant, highly reversible phosphorylated intermediates. Steady-state measurements in the absence of citrate and CoA showed that MgADP was produced by both wild type and H760A forms of ACL, with rates at three magnitudes lower than that of k(cat) for the full biosynthetic reaction. The ATPase activity of ACL, along with the small yet significant positional isotope exchange rate observed in H760A mutant ACL (~150 fold less than wild type), collectively suggested the presence of a second, albeit unproductive, phosphoryl transfer in ACL. Mathematical analysis and computational simulation suggested that the desorption of MgADP at a rate of ~7 s(-1) was the rate-limiting step in the biosynthesis of AcCoA and oxaloacetate.

Biochemical characterization of human HIF hydroxylases using HIF protein substrates that contain all three hydroxylation sites.

The HIF (hypoxia-inducible factor) plays a central regulatory role in oxygen homoeostasis. HIF proteins are regulated by three Fe(II)- and α-KG (α-ketoglutarate)-dependent prolyl hydroxylase enzymes [PHD (prolyl hydroxylase domain) isoenzymes 1-3 or PHD1, PHD2 and PHD3] and one asparaginyl hydroxylase [FIH (factor inhibiting HIF)]. The prolyl hydroxylases control the abundance of HIF through oxygen-dependent hydroxylation of specific proline residues in HIF proteins, triggering subsequent ubiquitination and proteasomal degradation. FIH inhibits the HIF transcription activation through asparagine hydroxylation. Understanding the precise roles and regulation of these four Fe(II)- and α-KG-dependent hydroxylases is of great importance. In the present paper, we report the biochemical characterization of the first HIF protein substrates that contain the CODDD (C-terminal oxygen-dependent degradation domain), the NODDD (N-terminal oxygen-dependent degradation domain) and the CAD (C-terminal transactivation domain). Using LC-MS/MS (liquid chromatography-tandem MS) detection, we show that all three PHD isoenzymes have a strong preference for hydroxylation of the CODDD proline residue over the NODDD proline residue and the preference is observed for both HIF1α and HIF2α protein substrates. In addition, steady-state kinetic analyses show differential substrate selectivity for HIF and α-KG in reference to the three PHD isoforms and FIH.

A tale of two subunits: how the neomorphic R132H IDH1 mutation enhances production of αHG.

Heterozygously expressed single-point mutations in isocitrate dehydrogenase 1 and 2 (IDH1 and IDH2, respectively) render these dimeric enzymes capable of producing the novel metabolite α-hydroxyglutarate (αHG). Accumulation of αHG is used as a biomarker for a number of cancer types, helping to identify tumors with similar IDH mutations. With IDH1, it has been shown that one role of the mutation is to increase the rate of conversion from αKG to αHG. To improve our understanding of the function of this mutation, we have detailed the kinetics of the normal (isocitrate to αKG) and neomorphic (αKG to αHG) reactions, as well as the coupled conversion of isocitrate to αHG. We find that the mutant IDH1 is very efficient in this coupled reaction, with the ability to form αHG from isocitrate and NADP(+). The wild type/wild type IDH1 is also able to catalyze this conversion, though it is much more sensitive to concentrations of isocitrate. This difference in behavior can be attributed to the competitive binding between isocitrate and αKG, which is made more favorable for αKG by the neomorphic mutation at arginine 132. Thus, each partial reaction in the heterodimer is functionally isolated from the other. To test whether there is a cooperative effect resulting from the two subunits being in a dimer, we selectively inactivated each subunit with a secondary mutation in the NADP/H binding site. We observed that the remaining, active subunit was unaffected in its associated activity, reinforcing the notion of each subunit being functionally independent. This was further demonstrated using a monomeric form of IDH from Azotobacter vinelandii, which can be shown to gain the same neomorphic reaction when a homologous mutation is introduced into that protein.

Rheum tanguticum Alleviates Cognitive Impairment in APP/PS1 Mice by Regulating Drug-Responsive Bacteria and Their Corresponding Microbial Metabolites.

Drugs targeting intestinal bacteria have shown great efficacy for alleviating symptoms of Alzheimer's disease (AD), and microbial metabolites are important messengers. Our previous work indicated that Rheum tanguticum effectively improved cognitive function and reshaped the gut microbial homeostasis in AD rats. However, its therapeutic mechanisms remain unclear. Herein, this study aimed to elaborate the mechanisms of rhubarb for the treatment of AD by identifying effective metabolites associated with rhubarb-responsive bacteria. The results found that rhubarb reduced hippocampal inflammation and neuronal damage in APP/PS1 transgenic (Tg) mice. 16S rRNA sequencing and metabolomic analysis revealed that gut microbiota and their metabolism in Tg mice were disturbed in an age-dependent manner. Rhubarb-responsive bacteria were further identified by real-time polymerase chain reaction (RT-PCR) sequencing. Four different metabolites reversed by rhubarb were found in the position of the important nodes on rhubarb-responsive bacteria and their corresponding metabolites combined with pathological indicators co-network. Furthermore, in vitro experiments demonstrated o-tyrosine not only inhibited the viabilities of primary neurons as well as BV-2 cells, but also increased the levels of intracellular reactive oxygen species and nitric oxide. In the end, the results suggest that rhubarb ameliorates cognitive impairment in Tg mice through decreasing the abundance of o-tyrosine in the gut owing to the regulation of rhubarb-responsive bacteria. Our study provides a promising strategy for elaborating therapeutic mechanisms of bacteria-targeted drugs for AD.

Three epitope-distinct human antibodies from RenMab mice neutralize SARS-CoV-2 and cooperatively minimize the escape of mutants.

Coronavirus disease 2019 (COVID-19), a pandemic disease caused by the newly emerging severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), has caused more than 3.8 million deaths to date. Neutralizing antibodies are effective therapeutic measures. However, many naturally occurring mutations at the receptor-binding domain (RBD) have emerged, and some of them can evade existing neutralizing antibodies. Here, we utilized RenMab, a novel mouse carrying the entire human antibody variable region, for neutralizing antibody discovery. We obtained several potent RBD-blocking antibodies and categorized them into four distinct groups by epitope mapping. We determined the involved residues of the epitope of three representative antibodies by cryo-electron microscopy (Cryo-EM) studies. Moreover, we performed neutralizing experiments with 50 variant strains with single or combined mutations and found that the mixing of three epitope-distinct antibodies almost eliminated the mutant escape. Our study provides a sound basis for the rational design of fully human antibody cocktails against SARS-CoV-2 and pre-emergent coronaviral threats.

ATP-competitive inhibitors of the mitotic kinesin KSP that function via an allosteric mechanism.

The mitotic kinesin KSP (kinesin spindle protein, or Eg5) has an essential role in centrosome separation and formation of the bipolar mitotic spindle. Its exclusive involvement in the mitotic spindle of proliferating cells presents an opportunity for developing new anticancer agents with reduced side effects relative to antimitotics that target tubulin. Ispinesib is an allosteric small-molecule KSP inhibitor in phase 2 clinical trials. Mutations that attenuate ispinesib binding to KSP have been identified, which highlights the need for inhibitors that target different binding sites. We describe a new class of selective KSP inhibitors that are active against ispinesib-resistant forms of KSP. These ATP-competitive KSP inhibitors do not bind in the nucleotide binding pocket. Cumulative data from generation of resistant cells, site-directed mutagenesis and photo-affinity labeling suggest that they compete with ATP binding via a novel allosteric mechanism.

Research on a learning rate with energy index in deep learning.

The stochastic gradient descent algorithm (SGD) is the main optimization solution in deep learning. The performance of SGD depends critically on how learning rates are tuned over time. In this paper, we propose a novel energy index based optimization method (EIOM) to automatically adjust the learning rate in the backpropagation. Since a frequently occurring feature is more important than a rarely occurring feature, we update the features to different extents according to their frequencies. We first define an energy neuron model and then design an energy index to describe the frequency of a feature. The learning rate is taken as a hyperparameter function according to the energy index. To empirically evaluate the EIOM, we investigate different optimizers with three popular machine learning models: logistic regression, multilayer perceptron, and convolutional neural network. The experiments demonstrate the promising performance of the proposed EIOM compared with that of other optimization algorithms.

Biochemical characterization of human prolyl hydroxylase domain protein 2 variants associated with erythrocytosis.

Prolyl hydroxylase domain proteins (PHD isozymes 1-3) regulate levels of the alpha-subunit of the hypoxia inducible factor (HIF) through proline hydroxylation, earmarking HIFalpha for proteosome-mediated degradation. Under hypoxic conditions, HIF stabilization leads to enhanced transcription and regulation of a multitude of processes, including erythropoiesis. Herein, we examine the biochemical characterization of PHD2 variants, Arg371His and Pro317Arg, identified from patients with familial erythrocytosis. The variants display differential effects on catalytic rate and substrate binding, implying that partial inhibition or selective inhibition with regard to HIFalpha isoforms of PHD2 could result in the phenotype displayed by patients with familial erythrocytosis.

MicroRNA-599 targets high-mobility group AT-hook 2 to inhibit cell proliferation and invasion in clear cell renal carcinoma.

Dysregulation of microRNAs (miRNAs) is associated with the occurrence and development of clear cell renal cell carcinoma (ccRCC) through their participation in a number of critical biological processes. Therefore, an in­depth investigation into miRNAs and their biological roles within ccRCC may provide useful insights and lead to the identification of novel therapeutic methods for patients with ccRCC. miRNA­599 (miR­599) serves critical roles in different types of human cancer. However, the expression pattern, biological function and molecular mechanism of miR­599 in ccRCC remain unknown. The present study aimed to detect the expression level of miR­599 in ccRCC, examine its effect on ccRCC progression and further explore the possible underlying mechanisms. It was observed that miR­599 was significantly underexpressed in ccRCC tissues and cell lines compared with the control. Functional assays revealed that restored expression of miR­599 restricted the proliferation and invasion of ccRCC cells. Bioinformatics analysis, luciferase reporter assay, reverse transcription­quantitative polymerase chain reaction and western blot analysis demonstrated that high­mobility group AT­hook 2 (HMGA2) was a direct target of miR­599 in ccRCC. HMGA2 knockdown simulated the suppressive effects caused by miR­599 overexpression in ccRCC. Recovered HMGA2 expression partially rescued the miR­599­mediated inhibition of ccRCC proliferation and invasion. These results suggest that miR­599 may serve tumour suppressive roles in ccRCC by directly targeting HMGA2, indicating that miR­599 may have potential as a treatment for patients with ccRCC.

MicroRNA­199b promotes cell proliferation and invasion in Wilms' tumour by directly targeting Runt­related transcription factor 3.

Emerging evidence has demonstrated that the deregulation of microRNAs (miRNAs) contributes to Wilms' tumour (WT) malignant progression. Therefore, identifying the essential miRNAs for WT onset and progression may be a promising therapeutic method for patients with this disease. Dysregulation of miRNA­199b (miR­199b) serves significant roles in various types of human cancer. However, its expression patterns, possible functions and associated mechanisms in WT are largely unknown. In the present study, the expression of miR­199b in WT tissues was detected by reverse transcription­quantitative polymerase chain reaction (RT­qPCR) analysis. The biological functions of miR­199b overexpression in WT cells were determined using Cell counting kit­8 and Transwell invasion assays. The mechanisms underlying the action of miR­199b in WT cells were also investigated using bioinformatics analysis, a luciferase reporter assay, RT­qPCR and western blot analysis. It was revealed that miR­199b expression was upregulated in WT tissues. In addition, the downregulation of miR­199b attenuated the proliferation and invasion of WT cells. Runt­related transcription factor 3 (RUNX3) was mechanistically predicted as a potential target of miR­199b. Subsequent experiments demonstrated that RUNX3 was a direct target gene of miR­199b in WT. In addition, the downregulation of RUNX3 in the WT tissues was inversely correlated with the miR­199b expression level. The recovered RUNX3 expression counteracted the oncogenic roles of miR­199b in WT cells. Therefore miR­199b may serve as an oncogene in WT progression by directly targeting RUNX3, thereby suggesting that the miR­199b/RUNX3 axis may be a promising therapeutic target for patients with WT.

Deciphering biochemical basis of Qingkailing injection-induced anaphylaxis in a rat model by time-dependent metabolomic profiling based on metabolite polarity-oriented analysis.

ETHNOPHARMACOLOGICAL RELEVANCE: Qingkailing injection (QKLI) is prepared from eight traditional Chinese medicinal materials or their extracts, which is widely used in clinical practice to treat the upper respiratory inflammation, pneumonia, high fever and viral encephalitis, nonetheless, suffering from serious anaphylaxis. AIM OF STUDY: This study aims to develop an integrative metabolomics approach for deciphering the biochemical basis of QKLI induced anaphylaxis (QKLI-IA). MATERIALS AND METHODS: The accuracy of animal modeling, the coverage of detected metabolites and the timeliness of pathological reaction are three key factors for revealing the biochemical basis of disease with untargeted metabolomics. In this study, firstly, the allergic rats (responders) were first screened by passive cutaneous anaphylaxis experiment and then were utilized for modeling. To cover a wider range of metabolites, a large-scale untargeted metabolomics based on metabolites polarity-oriented analysis was performed using ultra performance liquid chromatography coupled with quadrupole time-of-flight mass spectrometry. Then, to evaluate the timeliness of QKLI-IA, a time-dependent metabolomic profiling including the early, mid and late anaphylaxis stages of QKLI-IA, was performed. RESULTS: Corresponding to early, mid and late anaphylaxis stages of QKLI-IA, 14, 9 and 4 potential biomarkers were identified, respectively. Metabolism pathway analysis revealed that QKLI-IA resulted in dynamic changes in serum amino acid, fatty acid, glycerolipid, and phospholipid metabolisms. Twenty-four metabolites were found with identical fluctuating trends across the three stages of QKLI-IA. The results indicate that the pathogenesis of QKLI-IA is closely related to arachidonic acid metabolism. CONCLUSION: This research provides a methodology reference for revealing the biochemical basis of disease using metabolomic profiling and offers a new insight to understand the pathogenesis of QKLI-IA.

Improving deep convolutional neural networks with mixed maxout units.

Motivated by insights from the maxout-units-based deep Convolutional Neural Network (CNN) that "non-maximal features are unable to deliver" and "feature mapping subspace pooling is insufficient," we present a novel mixed variant of the recently introduced maxout unit called a mixout unit. Specifically, we do so by calculating the exponential probabilities of feature mappings gained by applying different convolutional transformations over the same input and then calculating the expected values according to their exponential probabilities. Moreover, we introduce the Bernoulli distribution to balance the maximum values with the expected values of the feature mappings subspace. Finally, we design a simple model to verify the pooling ability of mixout units and a Mixout-units-based Network-in-Network (NiN) model to analyze the feature learning ability of the mixout models. We argue that our proposed units improve the pooling ability and that mixout models can achieve better feature learning and classification performance.