A Fuzzy System Classification Approach for QSAR Modeling of α- Amylase and α-Glucosidase Inhibitors.

INTRODUCTION: This report proposes the application of a new Machine Learning algorithm called Fuzzy Unordered Rules Induction Algorithm (FURIA)-C in the classification of druglike compounds with antidiabetic inhibitory ability toward the main two pharmacological targets: α-amylase and α-glucosidase. METHODS: The two obtained QSAR models were tested for classification capability, achieving satisfactory accuracy scores of 94.5% and 96.5%, respectively. Another important outcome was to achieve various α-amylase and α-glucosidase fuzzy rules with high Certainty Factor values. Fuzzyrules derived from the training series and active classification rules were interpreted. An important external validation step, comparing our method with those previously reported, was also included. RESULTS: The Holm's test comparison showed significant differences (p-value<0.05) between FURIA-C, Linear Discriminating Analysis (LDA), and Bayesian Networks, the former beating the two latter according to the relative ranking score of the Holm's test. CONCLUSION: From these results, the FURIA-C algorithm could be used as a cutting-edge technique to predict (classify or screen) the α-amylase and α-glucosidase inhibitory activity of new compounds and hence speed up the discovery of new potent multi-target antidiabetic agents.

Beyond model interpretability using LDA and decision trees for α-amylase and α-glucosidase inhibitor classification studies.

In this report are used two data sets involving the main antidiabetic enzyme targets α-amylase and α-glucosidase. The prediction of α-amylase and α-glucosidase inhibitory activity as antidiabetic is carried out using LDA and classification trees (CT). A large data set of 640 compounds for α-amylase and 1546 compounds in the case of α-glucosidase are selected to develop the tree model. In the case of CT-J48 have the better classification model performances for both targets with values above 80%-90% for the training and prediction sets, correspondingly. The best model shows an accuracy higher than 95% for training set; the model was also validated using 10-fold cross-validation procedure and through a test set achieving accuracy values of 85.32% and 86.80%, correspondingly. Additionally, the obtained model is compared with other approaches previously published in the international literature showing better results. Finally, we can say that the present results provided a double-target approach for increasing the estimation of antidiabetic chemicals identification aimed by double-way workflow in virtual screening pipelines.

N-linear algebraic maps for chemical structure codification: a suitable generalization for atom-pair approaches?

The present manuscript introduces, for the first time, a novel 3D-QSAR alignment free method (QuBiLS-MIDAS) based on tensor concepts through the use of the three-linear and four-linear algebraic forms as specific cases of n-linear maps. To this end, the k(th) three-tuple and four-tuple spatial-(dis)similarity matrices are defined, as tensors of order 3 and 4, respectively, to represent 3Dinformation among "three and four" atoms of the molecular structures. Several measures (multi-metrics) to establish (dis)-similarity relations among "three and four" atoms are discussed, as well as, normalization schemes proposed for the n-tuple spatial-(dis)similarity matrices based on the simple-stochastic and mutual probability algebraic transformations. To consider specific interactions among atoms, both for the global and local indices, n-tuple path and length cut-off constraints are introduced. This algebraic scaffold can also be seen as a generalization of the vector-matrix-vector multiplication procedure (which is a matrix representation of the traditional linear, quadratic and bilinear forms) for the calculation of molecular descriptors and is thus a new theoretical approach with a methodological contribution. A variability analysis based on Shannon's entropy reveals that the best distributions are achieved with the ternary and quaternary measures corresponding to the bond and dihedral angles. In addition, the proposed indices have superior entropy behavior than the descriptors calculated by other programs used in chemo-informatics studies, such as, DRAGON, PADEL, Mold2, and so on. A principal component analysis shows that the novel 3D n-tuple indices codify the same information captured by the DRAGON 3D-indices, as well as, information not codified by the latter. A QSAR study to obtain deeper criteria on the contribution of the novel molecular parameters was performed for the binding affinity to the corticosteroid-binding globulin, using Cramer's steroid database. The achieved results reveal superior statistical parameters for the Bond Angle and Dihedral Angle approaches, consistent with the results obtained in variability analysis. Finally, the obtained QuBiLS-MIDAS models yield superior performances than all 3D-QSAR methods reported in the literature using the 31 steroids as training set, and for the popular division of Cramer's database in training (1-21) and test (22-31) sets, comparable to superior results in the prediction of the activity of the steroids are obtained. From the results achieved, it can be suggested that the proposed QuBiLS-MIDAS N-tuples indices are a useful tool to be considered in chemo-informatics studies.

New antitrichomonal drug-like chemicals selected by bond (edge)-based TOMOCOMD-CARDD descriptors.

Bond-based quadratic indices, new TOMOCOMD-CARDD molecular descriptors, and linear discriminant analysis (LDA) were used to discover novel lead trichomonacidals. The obtained LDA-based quantitative structure-activity relationships (QSAR) models, using nonstochastic and stochastic indices, were able to classify correctly 87.91% (87.50%) and 89.01% (84.38%) of the chemicals in training (test) sets, respectively. They showed large Matthews correlation coefficients of 0.75 (0.71) and 0.78 (0.65) for the training (test) sets, correspondingly. Later, both models were applied to the virtual screening of 21 chemicals to find new lead antitrichomonal agents. Predictions agreed with experimental results to a great extent because a correct classification for both models of 95.24% (20 of 21) of the chemicals was obtained. Of the 21 compounds that were screened and synthesized, 2 molecules (chemicals G-1, UC-245) showed high to moderate cytocidal activity at the concentration of 10 microg/ml, another 2 compounds (G-0 and CRIS-148) showed high cytocidal activity only at the concentration of 100 microg/ml, and the remaining chemicals (from CRIS-105 to CRIS-153, except CRIS-148) were inactive at these assayed concentrations. Finally, the best candidate, G-1 (cytocidal activity of 100% at 10 microg/ml) was in vivo assayed in ovariectomized Wistar rats achieving promising results as a trichomonacidal drug-like compound.

Bond-based linear indices in QSAR: computational discovery of novel anti-trichomonal compounds.

Trichomonas vaginalis (Tv) is the causative agent of the most common, non-viral, sexually transmitted disease in women and men worldwide. Since 1959, metronidazole (MTZ) has been the drug of choice in the systemic treatment of trichomoniasis. However, resistance to MTZ in some patients and the great cost associated with the development of new trichomonacidals make necessary the development of computational methods that shorten the drug discovery pipeline. Toward this end, bond-based linear indices, new TOMOCOMD-CARDD molecular descriptors, and linear discriminant analysis were used to discover novel trichomonacidal chemicals. The obtained models, using non-stochastic and stochastic indices, are able to classify correctly 89.01% (87.50%) and 82.42% (84.38%) of the chemicals in the training (test) sets, respectively. These results validate the models for their use in the ligand-based virtual screening. In addition, they show large Matthews' correlation coefficients (C) of 0.78 (0.71) and 0.65 (0.65) for the training (test) sets, correspondingly. The result of predictions on the 10% full-out cross-validation test also evidences the robustness of the obtained models. Later, both models are applied to the virtual screening of 12 compounds already proved against Tv. As a result, they correctly classify 10 out of 12 (83.33%) and 9 out of 12 (75.00%) of the chemicals, respectively; which is the most important criterion for validating the models. Besides, these classification functions are applied to a library of seven chemicals in order to find novel antitrichomonal agents. These compounds are synthesized and tested for in vitro activity against Tv. As a result, experimental observations approached to theoretical predictions, since it was obtained a correct classification of 85.71% (6 out of 7) of the chemicals. Moreover, out of the seven compounds that are screened, synthesized and biologically assayed, six compounds (VA7-34, VA7-35, VA7-37, VA7-38, VA7-68, VA7-70) show pronounced cytocidal activity at the concentration of 100 mug/ml at 24 h (48 h) within the range of 98.66%-100% (99.40%-100%), while only two molecules (chemicals VA7-37 and VA7-38) show high cytocidal activity at the concentration of 10 mug/ml at 24 h (48 h): 98.38% (94.23%) and 97.59% (98.10%), correspondingly. The LDA-assisted QSAR models presented here could significantly reduce the number of synthesized and tested compounds and could increase the chance of finding new chemical entities with anti-trichomonal activity.