Discovery of plant anti-inflammatory biomarkers by machine learning algorithms and metabolomic studies

Chagas de Paula, Daniela and Oliveria, T. B. and Zhang, Tong and Edrada-Ebel, Ruangelie and B Da Costa, Fernando (2013) Discovery of plant anti-inflammatory biomarkers by machine learning algorithms and metabolomic studies. Planta Medica, 79 (13). SL27. ISSN 0032-0943 (https://doi.org/10.1055/s-0033-1351853)

Full text not available in this repository.Request a copy

Abstract

NSAIDs are the most used anti-inflammatory (AI) drugs in the world. However, side effects still occur and some inflammatory pathologies lack efficient treatment. Cyclooxygenase (COX) and lipoxygenase (LOX) pathways are of utmost importance in inflammatory processes and therefore novel inhibitors for both of them are needed. Dual inhibitors on COX-1 and 5-LOX should be AI medicines with high efficacy and low side effects [1]. As AI activity of species from Asteraceae is well-known, we screened 55 leaf extracts (EtOH-H2O 7:3, v/v) against COX-1 and 5-LOX. Among the tested extracts, 13 of them (26.6%, IC50 range from 0.03 – 36 µg/mL) displayed the desired inhibition. Each extract was further analysed by HPLC-HRFTMS. The data of all samples were processed employing a differential expression analysis software (MZmine 2.6) coupled to the Dictionary of Natural Products for dereplication studies. The 6,052 characteristic peaks in the extracts according to their respective AI properties were selected by genetic search (Weka 3) and 1,261 of them remained. An additional selection by decision trees J48 (Weka 3) was carried out and 11 substances were determined as biomarkers for the dual inhibition. Finally, a model to predict new biologically active extracts was built by multilayer perceptron using the biomarkers data (70% of active and non-active samples comprised the training group and 30% the test group). In summary, we developed a new and robust model for prediction of the bioactivity of natural compounds, resulting in high percentage of correct predictions (90%), high precision (100%) for dual inhibition, and low error values (mean absolute error = 0.2) as also shown in the validation test. Thus, the biomarkers of the plant extracts were statistically correlated with their AI activities and therefore can be useful to predict new AI extracts as well as their AI compounds.