Optimizing machine-learning models for mutagenicity prediction through better feature selection.

Assessing a compound's mutagenicity using machine learning is an important activity in the drug discovery and development process. Traditional methods of mutagenicity detection, such as Ames test, are expensive and time and labor intensive. In this context, in silico methods that predict a compound mutagenicity with high accuracy are important. Recently, machine-learning (ML) models are increasingly being proposed to improve the accuracy of mutagenicity prediction. While these models are used in practice, there is further scope to improve the accuracy of these models. We hypothesize that choosing the right features to train the model can further lead to better accuracy. We systematically consider and evaluate a combination of novel structural and molecular features which have the maximal impact on the accuracy of models. We rigorously evaluate these features against multiple classification models (from classical ML models to deep neural network models). The performance of the models was assessed using 5- and 10-fold cross-validation and we show that our approach using the molecule structure, molecular properties, and structural alerts as feature sets successfully outperform the state-of-the-art methods for mutagenicity prediction for the Hansen et al. benchmark dataset with an area under the receiver operating characteristic curve of 0.93. More importantly, our framework shows how combining features could benefit model accuracy improvements.

Phytochemical screening, phenolic and flavonoid contents, antioxidant and cytogenotoxicity activities of Combretum leprosum Mart. (Combretaceae).

Combretum leprosum Mart. (Combretaceae), a shrub popularly known as mofumbo, is used in folk medicine for treatment of uterine bleeding, pertussis, gastric pain, and as a sedative. The aim of this study was to (1) determine the phytochemical profile,(2) identify chemical constituents and (3) examine antioxidant and cytogenotoxic activity of ethanolic extracts and fractions of stem bark and leaves. The plant material (leaf and stem bark) was submitted to extraction with ethanol, followed by partition using hexane, chloroform, and ethyl acetate. It was possible to identify and quantify the epicatechin in the ethanolic stem bark extract (0.065 mg/g extract) and rutin in the leaf extract (3.33 mg/g extract). Based upon in vitro tests a significant relationship was noted between findings from antioxidant tests and levels of total phenolic and flavonoid. Comparing all samples (extracts and fractions), the ethyl acetate fractions of stem bark (411.40 ± 15.38 GAE/g) and leaves (225.49 ± 9.47 GAE/g) exhibited higher phenolic content, whereas hexanic fraction of stem bark (124.28 ± 56 mg/g sample) and ethyl acetate fraction of leaves (238.91 ± 1.73 mg/g sample) demonstrated a higher content of flavonoids. Among the antioxidant tests, the intermediate fraction of stem bark (28.5 ± 0.60 µg/ml) and ethyl acetate fraction of leaves (40 ± 0.56 µg/ml) displayed a higher % inhibition of free radical DPPH activity, whereas intermediate fraction of stem bark (27.5 ± 0.9 µg/ml) and hydromethanol fraction of leaves (81 ± 1.4 µg/ml) demonstrated inhibition of the free radical ABTS. In biological tests (Allium cepa and micronucleus in peripheral blood), data showed that none of the tested concentrations of ethanolic extracts of leaves and stem bark produced significant cytotoxicity, genotoxicity, and mutagenic activity.Abbreviations AA%: percentage of antioxidant activity; ABTS: 2,2'-azinobis-(3-ethylbenzothiazoline-6-sulfonic acid); CEUA: Ethics Committee in the Use of Animals; TLC: Thin Layer Chromatography; DNA: deoxyribonucleic acid; DPPH: 1,1-diphenyl-2-picrylhydrazyl; ROS: Reactive oxygen species; EEB: ethanol extract of the stem bark; HFB: Hexanic fraction of stem bark; IFB: Intermediate fraction of stem bark; CFB: Chloroform fraction of stem bark; EAFB: Ethyl acetate fraction of stem bark; HMFB: Hydromethanol fraction of the stem bark; EEL: Ethanol extract from leaves; HFL: Hexane fraction of leaves; CFL: Chloroform fraction of leaves; EAFL: Ethyl acetate fraction of leaves; HMFL: Hydromethanol fraction of leaves; GAE: Gallic Acid Equivalent; IC50: 50% inhibition concentration; HCOOH: Formic acid; HCl: hydrochloric acid; HPLC: High-performance liquid chromatography; MN: micronucleus; WHO: World Health Organization; UFLC: Ultra-Fast Liquid Chromatography; UESPI: State University of Piauí.

Heteroexpression of Mycobacterium leprae hypothetical protein ML0190 provides protection against DNA-alkylating agent methyl methanesulfonate.

Repair of DNA alkylation damage is essential for maintaining genome integrity and Fe(II)/2-oxoglutarate(2OG)-dependent dioxygenase family of enzymes play crucial role in repairing some of the alkylation damages. Alkylation repair protein-B (AlkB) of Escherichia coli belongs to Fe(II)/2OG-dependent dioxygenase family and carries out DNA dealkylation repair. We report here identification of a hypothetical Mycobacterium leprae protein (accession no. ML0190) from the genomic database and show that this 615-bp open reading frame encodes a protein with sequence and structural similarity to Fe(II)/2OG-dependent dioxygenase AlkB. We identified mRNA transcript of this gene in the M. leprae infected clinical skin biopsy samples isolated from the leprosy patients. Heterologous expression of ML0190 in methyl methane sulfonate (MMS) sensitive and DNA repair deficient strain of Saccharomyces cerevisiae and Escherichia coli resulted in resistance to alkylating agent MM. The results of the present study imply that Mycobacterium leprae ML0190 is involved in protecting the bacterial genome from DNA alkylation damage.

Identification of Nontoxic Substructures: A New Strategy to Avoid Potential Toxicity Risk.

Avoidance of structural alerts (SAs) might reduce the risk of failure in drug discovery. However, there are still some marketed drugs containing SA, which indicates that SA should be analyzed carefully to avoid their excessive uses. Several detection systems, including automatic mining methods and expert systems, have been developed to identify SA. These methods only focus on toxic compounds that support the SA without consideration of nontoxic ones. Here, we proposed a frequency-based substructure detection protocol that learns from the nontoxic compounds containing SA to get nontoxic substructures (NTSs), whose appearance will reduce the probability of a compound becoming toxic. Kazius and Hansen's Ames mutagenicity dataset was used as an example to demonstrate the protocol. SARpy and ToxAlerts were first employed to obtain the potential SA. Then 2 kinds of NTS were exploited: reverse effect substructures (RESs) and conjugate effect substructures. Contribution and prediction performance of the substructures were evaluated via neural network and rule-based methods. We also compared substructure-based methods with the conventional machine learning-based methods. The results demonstrated that most substructures contributed as supposed and substructure-based methods performed better in the resistance of overfitting. This work indicated that the protocol could effectively reduce the false positive rate in prediction of chemical mutagenicity, and possibly extend to other endpoints.

Validation of Toxtree and SciQSAR in silico predictive software using a publicly available benchmark mutagenicity database and their applicability for the qualification of impurities in pharmaceuticals.

The draft ICH M7 guidance (US FDA, 2013) recommends that the computational assessment of bacterial mutagenicity for the qualification of impurities in pharmaceuticals be performed using an expert rule-based method and a second statistically-based (Q)SAR method. The public nonproprietary 6489 compound Hansen benchmark mutagenicity data set was used as an external validation data set for Toxtree, a free expert rule-based SAR software. This is the largest known external validation of Toxtree. The Toxtree external validation specificity, sensitivity, concordance and false negative rate for this mutagenicity data set was 66%, 80%, 74% and 20%, respectively. This mutagenicity data set was also used to create a statistically-based SciQSAR-Hansen mutagenicity model. In a 10% leave-group-out internal cross validation study the specificity, sensitivity, concordance and false negative rate for the SciQSAR mutagenicity model was 71%, 83%, 77% and 17%, respectively. Combining Toxtree and SciQSAR predictions and scoring a positive finding in either software as a positive mutagenicity finding reduced the false negative rate to 7% and increased sensitivity to 93% at the expense of specificity which decreased to 53%. The results of this study support the applicability of Toxtree, and the SciQSAR-Hansen mutagenicity model for the qualification of impurities in pharmaceuticals.

Food borne yeasts as DNA-bioprotective agents against model genotoxins.

Yeasts isolated from Italian beverages and foods (wine and cheeses) were identified as Saccharomyces cerevisiae and Debaryomyces hansenii by sequencing the D1/D2 domain of the 26S rRNA gene and differentiated, at strain level, by microsatellite PCR fingerprinting and RAPD-PCR. All the strains showed antioxidant activity, as demonstrated by their ability to scavenge the free radical diphenyl-1-picrylhydrazyl (DPPH). Furthermore, tested strains revealed high in vitro inhibitory activity against two model genotoxins, 4-nitroquinoline-1-oxide (4-NQO) and N-methyl-N'-nitro-N-nitrosoguanidine (MNNG), as showed by short-term methods with different target cells: SOS-Chromotest with Escherichia coli PQ37 and Comet assay with HT-29 enterocytes. High inhibitory activity towards 4-NQO was associated with cell viability, while heat-inactivated cells showed a reduced antigenotoxic capability. Surprisingly, high inhibition of MNNG genotoxicity was observed even with heat-treated cells. Moreover, the strains able to inhibit the genotoxins induced some changes in the spectroscopic properties of the original compound. This result perfectly agrees with the information obtained by the two bioassays. Interestingly, strains characterized for antioxidant and antigenotoxic properties, also presented acid-bile tolerance, indicating that food autochthonous yeasts could be expected to reach gut in viable form and thus prevent genotoxin DNA damage in situ.

Assessment of the mutagenic and antimutagenic activity of Synadenium umbellatum Pax latex by micronucleus test in mice.

Synadenium umbellatum Pax, popularly known as "cola-nota", is a medicinal plant that grows in tropical regions. The latex of this plant is used against various diseases, such as diabetes mellitus, leprosy, tripanosomiasis, leukemia, and several malignant tumors. The mutagenic, antimutagenic, and cytotoxic effects of the latex of this plant were investigated by measuring the frequency of micronuclei in mice bone marrow cells. To evaluate mutagenicity, the animals were treated with four doses of latex (10, 30, 50, and 100 mg/kg body weight). To study the antimutagenic activity, the animals were simultaneously treated with latex and mitomycin C (4 mg/kg). The cytotoxicity was evaluated by polychromatic and normochromatic erythrocytes ratio. Our results showed a significant increase of frequency of micronucleated polychromatic erythrocytes (MNPCE) compared to the negative control group (p < 0.05). Concerning antimutagenicity, the doses of 10 and 30 mg/kg co-administered with mitomycin C showed significant decrease in MNPCE frequency compared to the positive control group (p < 0.05). However, no significant reduction in MNPCE frequency (p > 0.05) was detected at the doses of 50 and 100 mg/kg. Under our experimental conditions, the results obtained indicate strong mutagenic and cytotoxic activity of S. umbellatum latex except the dose of 10 mg/kg and moderate antimutagenic effect at lower doses.

Assessment of the in vitro and in vivo genotoxicity of Thalomid (thalidomide).

Thalomid is the FDA-approved commercial formulation of thalidomide currently used in the US to treat erythema nodosum leprosum, a complication of leprosy. The genotoxicity of Thalomid thalidomide was assessed in the Ames reverse mutation, AS52/XPRT mammalian cell forward gene mutation, and mouse bone marrow micronucleus assays. The Ames and AS52 assays were performed with and without S9. In the Ames, Salmonella typhimurium strains TA1535, 1537, 98, 100, and 102 and Escherichia coli strain WP2 uvrA were used. Assays were performed by using plate incorporation and liquid pre-incubation systems at thalidomide doses of 50-10,000 microg/plate. In the AS52 assay, Chinese hamster ovary cells were plated with fortified Ham's F12 medium and incubated overnight. The medium was then incubated with 1-1000 microg/ml thalidomide. After a series of aspirations, washings, reconstitutions, and incubations, mutant AS52 cells were fixed and stained. Colonies were then counted and the relative survival frequencies compared to negative controls. In the mouse micronucleus assay, Crl:CD-1 albino mice were dosed with 500, 2,500, and 5,000 mg/kg thalidomide and sacrificed over 72 h. Femurs were flushed with fetal bovine serum and the suspensions centrifuged. The supernatant was aspirated and the cell pellet resuspended and stained. Polychromatic erythrocytes were scored for micronucleated polychromatic and normochromatic erythrocytes. Thalidomide did not increase revertant frequencies in all bacterial strains. It also did not produce any significant increase in the average mutant frequencies of AS52 cells and mouse micronucleated polychromatic erythrocytes. We conclude that Celgene's Thalomid thalidomide is non-genotoxic.

Reduction of clastogenic effect of clofazimine, an antileprosy drug, by vitamin A and vitamin C in bone marrow cells of mice.

Clofazimine (CLF), an antileprosy drug, has earlier been proved to be clastogenic in mice in vivo. It is an important constituent of the triple-drug regimen recommended by WHO for the treatment of leprosy. In this study the protective role of vitamins A and C (vit A and vit C) against the clastogenic effect of CLF in mouse bone marrow cells has been evaluated. Two doses (20 and 40 mg/kg) of vit C and two doses (2500 and 5000 IU/kg) of vit A were tested against a dose of 40 mg CLF/kg. The drug alone induced chromosomal aberrations of about 8 times the control value. Neither of the doses of vit C exhibited any clastogenic effect and, when administered simultaneously with CLF, both reduced the effect of CLF very significantly, the higher dose reducing chromosomal aberrations almost to the control value. Conversely, both doses of vit A, when administered alone, brought about significant increases in chromosome aberrations over the control value; the higher, but not the lower dose, given simultaneously with CLF, minimized the effect of CLF significantly but not as greatly as vit C. A scavenging effect of the vitamins, removing free radicals produced by CLF, is assumed to be responsible for modulation of the clastogenic effect of CLF.

Differential sensitivity of peripheral blood lymphocytes of untreated leprosy patients to mitomycin C.

The effects of a bifunctional alkylating agent mitomycin C (MMC), an effective inducer of chromosome aberrations and sister-chromatid exchanges (SCEs), have been studied in untreated leprosy patients. This was done to study the mutagen sensitivity of the leprosy patients. The frequency of chromosomal aberrations induced by MMC (conc. 0.01 microgram/ml) was 2.5% in controls, 3.6% in paucibacillary (PB), and 6.8% in multibacillary (MB) patients. The difference in the frequency of MMC-induced chromosome aberrations between the 3 groups studied was highly significant (p less than 0.01). Cultures grown with MMC showed the frequency of SCEs/cell to be 12.70 +/- 1.19 in controls, 19.97 +/- 3.51 in PB, and 29.66 +/- 5.92 in MB patients. The differences in the frequency of MMC-induced SCEs between the 3 groups were found to be highly significant (p less than 0.01). The enhanced frequencies of spontaneous and MMC-induced chromosome aberrations and SCEs observed in PB and MB patients indicate a clear differential mutagen sensitivity between PB and MB patients who are known to have different immunological status and thereby differ in the severity of the disease.

Clofazimine binding studies with deoxyribonucleic acid.

1. The antileprosy drug, clofazimine, formed stable complexes with DNA and transfer RNA. A quantitative study was made of the spectral red shifts that occurred when clofazimine interacted with DNA. The red shift appeared specific for clofazimine binding to nucleic acid polymers. 2. The degree of clofazimine interaction with DNA was related to the G+C content of the DNA strand. As compared to the human strand, clofazimine interacted with the mycobacterial strand to give a larger red shift which was consistent with the increased G+C content of mycobacterial DNA. 3. It was found that clofazimine interacted with the synthetic single-stranded polynucleotide, poly G, whereas little interaction occurred withpoly A, poly C, or poly U. It was concluded that the guanine base region was a predominant site of clofazimine binding to DNA. 4. No evidence was found to indicate that clofazimine underwent intercalative binding between the base pairs of DNA. 5. It was proposed that clofazimine underwent binding along the minor groove region of DNA at appropriate base sequences which contain guanine. The resultant effect would inhibit template function of the DNA strand.