Consensus Prediction of Human Gut Microbiota-Mediated Metabolism Susceptibility for Small Molecules by Machine Learning, Structural Alerts, and Dietary Compounds-Based Average Similarity Methods.

The human gut microbiota (HGM) colonizing human gastrointestinal tract (HGT) confers a repertoire of dynamic and unique metabolic capacities that are not possessed by the host and therefore is tentatively perceived as an alternative metabolic ″organ″ besides the liver in the host. Nevertheless, the significant contribution of HGM to the overall human metabolism is often overlooked in the modern drug discovery pipeline. Hence, a systematic evaluation of HGM-mediated drug metabolism is gradually important, and its computational prediction becomes increasingly necessary. In this work, a new data set containing both the HGM-mediated metabolism susceptible (HGMMS) and insusceptible (HGMMI) compounds (329 vs 320) was manually curated. Based on this data set, the first machine learning (ML) model, a new structural alerts (SA) model, and the K-nearest neighboring dietary compounds-based average similarity (AS) model were proposed to directly predict the HGM-mediated metabolism susceptibility for small molecules, and exhibit promising performance on three independent test sets. Finally, consensus prediction (ML/SA/AS) for DrugBank molecules revealed an intriguing phenomenon that a typical Michael acceptor ″α,ß-unsaturated carbonyl group″ is a very common warhead for the design of covalent inhibitors and inclined to be metabolized by HGM in anaerobic HGT to generate the reduced metabolite without the reactive warhead, which could be a new concern to medicinal chemists. To the best of our knowledge, we gleaned the first HGMMS/HGMMI data set, developed the first HGMMS/HGMMI classification model, implemented a relatively comprehensive program based on ML/SA/AS approaches, and found a new phenomenon on the HGM-mediated deactivation of an extensively used warhead for covalent inhibitors.

Quantitative Prediction of Hemolytic Toxicity for Small Molecules and Their Potential Hemolytic Fragments by Machine Learning and Recursive Fragmentation Methods.

Hemolytic toxicity, as one of the key toxicity endpoints for small molecules, can cause lysis of the erythrocyte membrane and subsequent release of hemoglobin into blood plasma, leading to multiple acute and chronic adverse effects. Hence, it is necessary to assess the hemolytic toxicity of small molecules in an early stage of drug discovery and development process, and it is more significant to quantitatively predict the hemolytic toxicity of small molecules before costly and time-consuming experiments. Nevertheless, this endpoint has never been quantitatively predicted due to the lack of an appropriate dataset. In this work, we manually collected a quantitative hemolytic toxicity dataset containing 805 small molecules with experimental values of HD50 (50% hemolytic dose) from a variety of literature, built the first machine learning-based regression model to quantitatively predict the hemolytic toxicity of small molecules, and developed a pragmatic software for automatic prediction. Based on this model, we further implemented an automatic recursive fragmentation module to predict the hemolytic fragments with high fragment efficiency for the given compound(s), which may be of particular interest to experimental medicinal chemists. Therefore, we anticipate that this quantitative model may help medicinal chemists boost the development of promising lead compounds with low hemolytic toxicity or fuel the discovery of highly hemolytic chemical probes to delve into the in-depth mechanism of the hemolytic process.

Prediction of Hemolytic Toxicity for Saponins by Machine-Learning Methods.

Saponins are a type of compounds bearing a hydrophobic steroid/triterpenoid moiety and hydrophilic carbohydrate branches. The majority of the saponins demonstrate a broad range of prominent pharmacological activities. Nevertheless, many saponins also possess harmful hemolytic toxicity, which can cause the lysis of erythrocytes and thereby hamper their applications in medicine. As such, the organic synthesis of diverse saponins with versatile therapeutic effects and without hemolytic toxicity has gained considerable interests among medicinal/organic chemists. To date, the non-hemolytic saponins of interests have usually been designed by the traditional trial-and-error method or discovered by serendipity. It would be more efficient to develop an in silico method to rationally design promising saponins without hemolytic toxicity prior to the laborious organic synthesis, despite the fact that there is, so far, no computational model to predict the hemolytic toxicity of saponins. To this end, we manually curate 331 hemolytic and 121 non-hemolytic saponins from the literature for the first time and build the first machine-learning-based hemolytic toxicity classification model for the saponins, which provides encouraging performance with 95% confidence intervals for accuracy (0.906 ± 0.009), precision (0.904 ± 0.012), specificity (0.711 ± 0.039), sensitivity (0.978 ± 0.010), F1-score (0.939 ± 0.006), and Matthews correlation coefficient (0.756 ± 0.025) on the test set by averaging over 19 different random data-partitioning schemes. Moreover, we have developed a free program called "e-Hemolytic-Saponin" for the automatic prediction and design of hemolytic/non-hemolytic saponins. To the best of our knowledge, we herein compile the first comprehensive saponin dataset focused on hemolytic toxicity, build the first informative model of hemolytic toxicity for the saponins, and implement the first convenient software that will enable organic/medicinal chemists to automatically predict and design the saponins of interests.

Computational Prediction of a New ADMET Endpoint for Small Molecules: Anticommensal Effect on Human Gut Microbiota.

The human gut microbiota (HGM), which are evolutionarily commensal in the human gastrointestinal system, are crucial to our health. However, HGM can be broadly shaped by multifaceted factors such as intake of drugs. About one-quarter of the existing drugs for humans, which are designed to target human cells rather than HGM, can notably alter the composition of HGM. Therefore, the anticommensal effect of human drugs should be avoided to the maximum extent possible in the drug discovery and development process. Nevertheless, the anticommensal effect of small molecules is a new ADMET (absorption, distribution, metabolism, excretion, and toxicity) end point, which was never predicted with the computational method before. In this work, we present the first machine-learning based consensus classification model with the accuracy (0.811 ± 0.012), precision (0.759 ± 0.032), specificity (0.901 ± 0.019), sensitivity (0.628 ± 0.036), F1-score (0.687 ± 0.023), and AUC (0.814 ± 0.030) respectively on the test set. Furthermore, we develop an easy-to-use "e-Commensal" program for the automatic prediction. Based on this program, virtual-screening of the food-constituent database (FooDB) indicates that 5888 of 23 202 food-relevant compounds are forecasted to possess an anticommensal effect on HGM. Several top-ranked anticommensal compounds in our prediction are further scrutinized and confirmed by experiments in the existing literature. To the best of our knowledge, this is the first classification model and stand-alone software for the prediction of commensal or anticommensal compounds impacting HGM.

Synthesis, biological evaluation, and molecular docking study of novel allyl-retrochalcones as a new class of protein tyrosine phosphatase 1B inhibitors.

We describe herein the design, synthesis, and biological evaluation of a series of novel protein tyrosine phosphatase 1B (PTP1B) inhibitor retrochalcones having an allyl chain at the C-5 position of their B ring. Biological screening results showed that the majority of these compounds exhibited an inhibitory activity against PTP1B. Thus, preliminary structure-activity relationship (SAR) and quantitative SAR analyses were conducted. Among the compounds, 23 was the most potent inhibitor, exhibiting the highest in vitro inhibitory activity against PTP1B with an IC50 of 0.57 µM. Moreover, it displayed a significant hepatoprotective property via activation of the IR pathway in type 2 diabetic db/db mice. In addition, the results of our docking study showed that 23, as a specific inhibitor of PTP1B, effectively transformed the WPD loop from "close" to "open" in the active site. These results may reveal suitable compounds for the development of PTP1B inhibitors.

Osthole Protects against Acute Lung Injury by Suppressing NF-κB-Dependent Inflammation.

Inflammation is a key factor in the pathogenesis of ALI. Therefore, suppression of inflammatory response could be a potential strategy to treat LPS-induced lung injury. Osthole, a natural coumarin extract, has been reported to protect against acute kidney injury through an anti-inflammatory mechanism, but its effect on ALI is poorly understood. In this study, we investigated whether osthole ameliorates inflammatory sepsis-related ALI. Results from in vitro studies indicated that osthole treatment inhibited the LPS-induced inflammatory response in mouse peritoneal macrophages through blocking the nuclear translocation of NF-κB. Consistently, the in vivo studies indicated that osthole significantly prolonged the survival of septic mice which was accompanied by inflammation suppression. In the ALI mouse model, osthole effectively inhibited the development of lung tissue injury, leukocytic recruitment, and cytokine productions, which was associated with inhibition of NF-κB nuclear translocation. These findings provide evidence that osthole was a potent inhibitor of NF-κB and inflammatory injury and suggest that it could be a promising anti-inflammatory agent for therapy of septic shock and acute lung injury.

Proposed Hydrogen-Bonding Index of Donor or Acceptor Reflecting Its Intrinsic Contribution to Hydrogen-Bonding Strength.

In this work, we tentatively propose that the hydrogen-bonding strength EHB (referring to the minimal hydrogen-bonding energy) and its corresponding hydrogen-bond (HB) distance (referring to the optimal HB distance dHB) for simple mono-HB systems have an exponential relationship on the basis of MP2 and DFT computational results. We take a step further and propose that the hydrogen-bonding indices of the donor (Idonor) and acceptor (Iacceptor), reflecting their intrinsic contributions to hydrogen-bonding strength, also have an exponential relation with the hypothetical effective hydrogen-bond radii of the donor (rdonor) and acceptor (racceptor), respectively. On the basis of extensive quantum-mechanical calculations, relevant assumptions about the hydrogen-bonding index are rationalized. Moreover, the hydrogen-bonding index is also suggested as an additional prefiltering criterion for virtual screening besides the widely accepted Lipinski's rule of five. Finally, a "Hydrogen-Bond Index Estimator (HBIE)" module has been implemented in our Visual Force Field Derivation Toolkit (VFFDT) program to approximately and rapidly estimate the hydrogen-bonding indices of any small molecules in batch and screen possible stronger donors or acceptors from the small-molecule database. To the best of our knowledge, the concept of the hydrogen-bonding index and its potential application are proposed here for the first time.

Electron affinity of tricyclic, bicyclic, and monocyclic compounds containing cyanoenones correlates with their potency as inducers of a cytoprotective enzyme.

UNLABELLED: Tricyclic, bicyclic, and monocyclic compounds containing cyanoenones induce various anti-inflammatory and cytoprotective enzymes through activation of the Keap1/Nrf2/ARE (antioxidant response element) pathway. The potency of these compounds as Nrf2 activators was determined using a prototypic cytoprotective enzyme NAD(P)H: quinone oxidoreductase 1 (NQO1) in Hepa1c1c7 murine hepatoma cells. The electron affinity (EA) of the compounds, expressed as the energy of their lowest unoccupied molecular orbital [E (LUMO)], was evaluated using two types of quantum mechanical calculations: the semiempirical (AM1) and the density functional theory (DFT) methods. We observed striking linear correlations [r=0.897 (AM1) and 0.936 (DFT)] between NQO1 inducer potency of these compounds and their E (LUMO) regardless of the molecule size. Importantly and interestingly, this finding demonstrates that the EA is the essentially important factor that determines the reactivity of the cyanoenones with Keap1.

VFFDT: A New Software for Preparing AMBER Force Field Parameters for Metal-Containing Molecular Systems.

Force fields are fundamental to molecular dynamics simulations. However, the incompleteness of force field parameters has been a long-standing problem, especially for metal-related systems. In our previous work, we adopted the Seminario method based on the Hessian matrix to systematically derive the zinc-related force field parameters for AMBER. In this work, in order to further simplify the whole protocol, we have implemented a user-friendly Visual Force Field Derivation Toolkit (VFFDT) to derive the force field parameters via simply clicking on the bond or angle in the 3D viewer, and we have further extended our previous program to support the Hessian matrix output from a variety of quantum mechanics (QM) packages, including Gaussian 03/09, ORCA 3.0, QChem, GAMESS-US, and MOPAC 2009/2012. In this toolkit, a universal VFFDT XYZ file format containing the raw Hessian matrix is available for all of the QM packages, and an instant force field parametrization protocol based on a semiempirical quantum mechanics (SQM) method is introduced. The new function that can automatically obtain the relevant parameters for zinc, copper, iron, etc., which can be exported in AMBER Frcmod format, has been added. Furthermore, our VFFDT program can read and write files in AMBER Prepc, AMBER Frcmod, and AMBER Mol2 format and can also be used to customize, view, copy, and paste the force field parameters in the context of the 3D viewer, which provides utilities complementary to ANTECHAMBER, MCPB, and MCPB.py in the AmberTools.

Pharmacokinetics and pharmacodynamics of orally administered acetylenic tricyclic bis(cyanoenone), a highly potent Nrf2 activator with a reversible covalent mode of action.

The acetylenic tricyclic bis(cyanoenone) TBE-31 is a highly potent cysteine targeting compound with a reversible covalent mode of action; its best-characterized target being Kelch-like ECH-associated protein-1 (Keap1), the cellular sensor for oxidants and electrophiles. TBE-31 reacts with cysteines of Keap1, impairing its ability to target nuclear factor-erythroid 2 p45-related factor 2 (Nrf2) for degradation. Consequently, Nrf2 accumulates and orchestrates cytoprotective gene expression. In this study we investigated the pharmacokinetic and pharmacodynamic properties of TBE-31 in C57BL/6 mice. After a single oral dose of 10 µmol/kg (∼200 nmol/animal), the concentration of TBE-31 in blood exhibited two peaks, at 22.3 nM and at 15.5 nM, 40 min and 4 h after dosing, respectively, as determined by a quantitative stable isotope dilution LC-MS/MS method. The AUC0-24h was 195.5 h/nmol/l, the terminal elimination half-life was 10.2 h, and the kel was 0.068 h(-1). To assess the pharmacodynamics of Nrf2 activation by TBE-31, we determined the enzyme activity of its prototypic target, NAD(P)H: quinone oxidoreductase 1 (NQO1) and found it elevated by 2.4- and 1.5-fold in liver and heart, respectively. Continuous feeding for 18 days with diet delivering the same daily doses of TBE-31 under conditions of concurrent treatment with the immunosuppressive agent azathioprine had a similar effect on Nrf2 activation without any indications of toxicity. Together with previous reports showing the cytoprotective effects of TBE-31 in animal models of carcinogenesis, our results demonstrate the high potency, efficacy and suitability for chronic administration of cysteine targeting reversible covalent drugs.

Synthesis of (13) C2 (15) N2 -labeled anti-inflammatory and cytoprotective tricyclic bis(cyanoenone) ([(13) C2 (15) N2 ]-TBE-31) as an internal standard for quantification by stable isotope dilution LC-MS method.

Tricyclic bis(cyanoenone), TBE-31, one of the most potent activators of the Keap1/Nrf2/antioxidant response element pathway, has been developed as a new anti-inflammatory and cytoprotective agent. (13) C2 (15) N2 -labeled TBE-31 ([(13) C2 (15) N2 ]-TBE-31), which has two (13) C and two (15) N atoms in two cyano groups, was designed to develop a method for quantification of cell, tissue, and plasma levels of TBE-31 that involves chromatography/mass spectrometry coupled with the use of a stable isotope-labeled internal standard. [(13) C2 (15) N2 ]-TBE-31 was successfully synthesized in four steps from a previously reported intermediate, which is prepared in 11 steps from cyclohexanone, by introduction of two (13) C atoms with ethyl [(13) C]formate and two (15) N atoms with hydroxyl[(15) N]amine. The stable isotope dilution liquid chromatography-mass spectrometry method for quantification of TBE-31 was successfully developed using [(13) C2 (15) N2 ]-TBE-31 to compensate for any variables encountered during sample processing and analysis.

Demethylzeylasteral exerts potent efficacy against non-small-cell lung cancer via the P53 signaling pathway.

Lung cancer has one of the highest mortality rates worldwide, with non-small-cell lung cancer (NSCLC) constituting approximately 85% of all cases. Demethylzeylasteral (DEM), extracted from Tripterygium wilfordii Hook F, exhibits notable anti-tumor properties. In this study, we revealed that DEM could effectively induce NSCLC cell apoptosis. Specifically, DEM can dose-dependently suppress the viability and migration of human NSCLC cells. RNA-seq analysis revealed that DEM regulates the P53-signaling pathway, which was further validated by assessing crucial proteins involved in this pathway. Biacore analysis indicated that DEM has high affinity with the P53 protein. The CDX model demonstrated DEM's anti-tumor actions. This work provided evidence that DEM-P53 interaction stabilizes P53 protein and triggers downstream anti-tumor activities. These findings indicate that DEM treatment holds promise as a potential therapeutic approach for NSCLC, which warrants further clinical assessment in patients with NSCLC.

Tethered spectroscopic probes estimate dynamic distances with subnanometer resolution in voltage-dependent potassium channels.

Measurements of inter- and intramolecular distances are important for monitoring structural changes and understanding protein interaction networks. Fluorescence resonance energy transfer and functionalized chemical spacers are the two predominantly used strategies to map short-range distances in living cells. Here, we describe the development of a hybrid approach that combines the key advantages of spectroscopic and chemical methods to estimate dynamic distance information from labeled proteins. Bifunctional spectroscopic probes were designed to make use of adaptable-anchor and length-varied spacers to estimate molecular distances by exploiting short-range collisional electron transfer. The spacers were calibrated using labeled polyproline peptides of defined lengths and validated by molecular simulations. This approach was extended to estimate distance restraints that enable us to evaluate the resting-state model of the Shaker potassium channel.

Synthesis and biological evaluation of biotin conjugates of (±)-(4bS,8aR,10aS)-10a-ethynyl-4b,8,8-trimethyl-3,7-dioxo-3,4b,7,8,8a,9,10,10a-octahydro-phenanthrene-2,6-dicarbonitrile, an activator of the Keap1/Nrf2/ARE pathway, for the isolation of its protein targets.

The tricycle 1 ((±)-(4bS,8aR,10aS))-10a-ethynyl-4b,8,8-trimethyl-3,7-dioxo-3,4b,7,8,8a,9,10,10a-octahydrophenanthrene-2,6-dicarbonitrile), a potent activator of the Keap1/Nrf2/ARE pathway, has the potential to be a first in class drug for the treatment of diabetic nephropathy. To identify the protein targets for the development of 1, the (1:1)-diasteromeric mixture of biotinylated tricycles 3a and 3b were designed and synthesized. For the synthesis of 3a and 3b, a new important precursor, hydroxylated tricycle (±)-16 was synthesized from 4 by a C1 α-methyl group oxidation protocol, which involves cyclopalladation of the C1 α-methyl group from a C2-oxime. For the induction of the phase 2 cytoprotective enzyme NQO1 in Hepa1c1c7 murine hepatoma cells, the diasteromeric mixture 3a and 3b shows high potency (CD, 75 nM) although this potency is lower than that of 1 and 16. Thus, biotinylated tricycles 3a and 3b may be promising tools for the isolation of the protein targets of 1.

Catalytic enantioselective halolactonization of enynes and alkenes.

New organocatalysts have been developed for the enantioselective halolactonization of (Z)-1,3-enynes and 1,1-disubstituted alkenes. In the case of 1,3-enynes, the carboxylate nucleophile and halogen electrophile were added to the conjugated π-system from the same face. Up to 99% ee was achieved for the 1,4-syn-bromolactonization of conjugated (Z)-1,3-enynes. Based on the results from the enyne halolactonization, a second generation of catalysts was designed for simple olefins. Up to 91% ee was observed for chlorolactonization of 1,1-disubstituted alkenes. The catalysts developed for the enantioselective halolactonization of both enynes and alkenes are composed of a cinchona alkaloid skeleton tethered to a urea group.

(±)-(4bS,8aR,10aS)-10a-Ethynyl-4b,8,8-trimethyl-3,7-dioxo-3,4b,7,8,8a,9,10,10a-octa-hydro-phenanthrene-2,6-dicarbonitrile.

The anti-inflammatory and cytoprotective tricyclic title compound, C(21)H(18)N(2)O(2), also known as TBE-31, crystallizes with two nearly superimposable mol-ecules in the asymmetric unit. In both mol-ecules, the three ring systems conform to an envelope-chair-planar arrangement. The central ring, in a cyclohexane chair conformation, contains an axial ethynyl group that bends slightly off from a nearby axial methyl group because of the 1,3-diaxial repulsion between the two groups. In the crystal, weak C-H⋯N and C-H⋯O inter-actions form chains along [001].

e-Graphene: A Computational Platform for the Prediction of Graphene-Based Drug Delivery System by Quantum Genetic Algorithm and Cascade Protocol.

Graphene, as a novel category of carbon nanomaterials, has attracted a great attention in the field of drug delivery. Due to its large dual surface area, graphene can efficiently load drug molecules with high capacity via non-covalent interaction without chemical modification of the drugs. Hence, it ignites prevalent interests in developing a new graphene/graphene oxide (GO)-based drug delivery system (GDDS). However, current design of GDDS primarily depends on the prior experimental experience with the trial-and-error method. Thus, it is more appealing to theoretically predict possible GDDS candidates before experiments. Toward this end, we propose to fuse quantum genetic algorithm (QGA) and quantum mechanics (QM)/semi-empirical quantum mechanics (SQM)/force field (FF) to globally search the optimal binding interaction between the graphene/GO and drug in a given GDDS and develop a free computational platform "e-Graphene" to automatically predict/screen potential GDDS candidates. To make this platform more pragmatic for the rapid yet relatively accurate prediction, we further propose a cascade protocol via firstly conducting a fast QGA/FF calculation with fine QGA parameters and automatically passing the best chromosomes from QGA/FF to initialize a higher level QGA/SQM or QGA/QM calculation with coarse QGA parameters (e.g., small populations and short evolution generations). By harnessing this platform and protocol, systematic tests on a typical GDDS containing an anticancer drug SN38 illustrate that high fabrication rates of hydroxyl, epoxy, and carboxyl groups on a pristine graphene model will compromise the stability of GDDS, implying that an appropriate functionalization rate is crucial for the delicate balance between the stability and solubility/biocompatibility of GDDS. Moreover, automatic GDDS screen in the DrugBank database is performed and elicits four potential GDDS candidates with enhanced stability than the commonly tested GDDS containing SN38 from the computational point of view. We hope that this work can provide a useful program and protocol for experimental scientists to rationally design/screen promising GDDS candidates prior to experimental tests.

Enantioselective bromolactonization of conjugated (Z)-enynes.

A catalytic enantioselective syn-1,4-bromolactonization of conjugated (Z)-enynes was reported. Diastereomeric ratios >20:1 and up to 99% enantiomeric excesses were observed. Di-, tri-, and tetra-substituted bromoallenes were prepared together with lactone heterocycles efficiently and stereoselectively. Preliminary investigations suggest that the chiral catalyst may serve as a bifunctional reagent by interacting with both a carboxylic acid nucleophile and NBS electrophile.

In Silico Prediction of Hemolytic Toxicity on the Human Erythrocytes for Small Molecules by Machine-Learning and Genetic Algorithm.

Hemolytic toxicity of small molecules, as one of the important ADMET end points, can cause the lysis of erythrocytes membrane and leaking of hemoglobin into the blood plasma, which leads to various side effects. Thus, it is very crucial to assess the hemolytic potential of small molecules during the early stage of drug development process. However, so far there is no computational model to predict the human hemolytic toxicity of small molecules. To this end, we manually curate the hemolytic toxicity data set for the small molecules experimentally evaluated on the human erythrocytes, develop the first machine-learning (ML) based models to predict the human hemolytic toxicity of small molecules, harness the genetic algorithm (GA) and ML based model to optimize human hemolytic toxicity based on the molecular fingerprint to derive "optimal virtual fingerprints (OVFs)" with the desired hemolytic/nonhemolytic property, and finally implement a free software for the users to predict/optimize the human hemolytic toxicity with ML and GA in the automatic manner.

e-Sweet: A Machine-Learning Based Platform for the Prediction of Sweetener and Its Relative Sweetness.

Artificial sweeteners (AS) can elicit the strong sweet sensation with the low or zero calorie, and are widely used to replace the nutritive sugar in the food and beverage industry. However, the safety issue of current AS is still controversial. Thus, it is imperative to develop more safe and potent AS. Due to the costly and laborious experimental-screening of AS, in-silico sweetener/sweetness prediction could provide a good avenue to identify the potential sweetener candidates before experiment. In this work, we curate the largest dataset of 530 sweeteners and 850 non-sweeteners, and collect the second largest dataset of 352 sweeteners with the relative sweetness (RS) from the literature. In light of these experimental datasets, we adopt five machine-learning methods and conformational-independent molecular fingerprints to derive the classification and regression models for the prediction of sweetener and its RS, respectively via the consensus strategy. Our best classification model achieves the 95% confidence intervals for the accuracy (0.91 ± 0.01), precision (0.90 ± 0.01), specificity (0.94 ± 0.01), sensitivity (0.86 ± 0.01), F1-score (0.88 ± 0.01), and NER (Non-error Rate: 0.90 ± 0.01) on the test set, which outperforms the model (NER = 0.85) of Rojas et al. in terms of NER, and our best regression model gives the 95% confidence intervals for the R2(test set) and ΔR2 [referring to |R2(test set)- R2(cross-validation)|] of 0.77 ± 0.01 and 0.03 ± 0.01, respectively, which is also better than the other works based on the conformation-independent 2D descriptors (e.g., 2D Dragon) according to R2(test set) and ΔR2. Our models are obtained by averaging over nineteen data-splitting schemes, and fully comply with the guidelines of Organization for Economic Cooperation and Development (OECD), which are not completely followed by the previous relevant works that are all on the basis of only one random data-splitting scheme for the cross-validation set and test set. Finally, we develop a user-friendly platform "e-Sweet" for the automatic prediction of sweetener and its corresponding RS. To our best knowledge, it is a first and free platform that can enable the experimental food scientists to exploit the current machine-learning methods to boost the discovery of more AS with the low or zero calorie content.