Prediction of Anti-inflammatory Plants and Discovery of Their Biomarkers by Machine Learning Algorithms and Metabolomic Studies
Prediction of Anti-inflammatory Plants and Discovery of Their Biomarkers by Machine Learning Algorithms and Metabolomic Studies
复制标题
DOI：
10.1055/s-0034-1396206
复制
发表时间：
2015-04-01
期刊：
影响因子：
2.7
通讯作者：
Da Costa, Fernando Batista
中科院分区：
文献类型：
--
作者：
Chagas-Paula, Daniela Aparecida;Oliveira, Tiago Branquinho;Da Costa, Fernando Batista
非甾体类抗炎药是世界上使用最多的抗炎药。然而，副作用仍然存在，一些炎症性病理缺乏有效的治疗。环加氧酶和脂加氧酶途径在炎症过程中至关重要；因此，目前需要新的抑制剂来治疗这两种疾病。环加氧酶-1和5-脂加氧酶双抑制剂是一种疗效高、副作用低的抗炎药物。本研究采用高效液相色谱-高分辨率- orbitrap -质谱法分析了57株体外双抑制环氧化酶-1和5-脂氧化酶的菊科植物叶片提取物（EtOH-H2O 7:3, v/v），并利用机器学习算法进行了计算机研究。所有样本的数据均通过与天然产物词典耦合的差异表达分析软件进行去复制研究。根据提取物的抗炎特性，采用遗传算法选择6052个色谱峰（ESI正、负模式）；经过这一过程，剩下了1241个。使用决策树分类器进行了一项研究，并确定了11种化合物因其抗炎潜力而成为生物标志物。最后，利用生物标志物数据，利用多层感知器（人工神经网络）和反向传播算法，建立了一个利用液相色谱-质谱信息预测Asteraceae物种中新的生物活性提取物的模型。结果，获得了一种新的鲁棒的预测天然化合物抗炎活性的人工神经网络模型，其预测正确率高（81%），双抑制精度高（100%），误差值低（平均绝对误差=0.3），验证试验也显示了这一点。因此，Asteraceae提取物的生物标志物与其抗炎活性具有统计学相关性，因此可以仅使用液相色谱-质谱分析数据来预测新的抗炎提取物及其抗炎化合物。
Nonsteroidal anti-inflammatory drugs are the most used anti-inflammatory medicines in the world. Side effects still occur, however, and some inflammatory pathologies lack efficient treatment. Cyclooxygenase and lipoxygenase pathways are of utmost importance in inflammatory processes; therefore, novel inhibitors are currently needed for both of them. Dual inhibitors of cyclooxygenase-1 and 5-lipoxygenase are anti-inflammatory drugs with high efficacy and low side effects. In this work, 57 leaf extracts (EtOH-H2O 7:3, v/v) from Asteraceae species with in vitro dual inhibition of cyclooxygenase-1 and 5-lipoxygenase were analyzed by high-performance liquid chromatography-high-resolution-ORBITRAP-mass spectrometry analysis and subjected to in silico studies using machine learning algorithms. The data from all samples were processed by employing differential expression analysis software coupled to the Dictionary of Natural Products for dereplication studies. The 6052 chromatographic peaks (ESI positive and negative modes) of the extracts were selected by a genetic algorithm according to their respective anti-inflammatory properties; after this procedure, 1241 of them remained. A study using a decision tree classifier was carried out, and 11 compounds were determined to be biomarkers due to their anti-inflammatory potential. Finally, a model to predict new biologically active extracts from Asteraceae species using liquid chromatography-mass spectrometry information with no prior knowledge of their biological data was built using a multilayer perceptron (artificial neural networks) with the back-propagation algorithm using the biomarker data. As a result, a new and robust artificial neural network model for predicting the anti-inflammatory activity of natural compounds was obtained, resulting in a high percentage of correct predictions (81%), high precision (100%) for dual inhibition, and low error values (mean absolute error=0.3), as also shown in the validation test. Thus, the biomarkers of the Asteraceae extracts were statistically correlated with their anti-inflammatory activities and can therefore be useful to predict new anti-inflammatory extracts and their anti-inflammatory compounds using only liquid chromatography-mass spectrometry data.