Online Analysis Methods For Multi-source Plant Electrophysiological Data

Plant electrical signals are response to environment and external stimulus and have important physiological significance on plant. The increasing data are mainly applied in researcher’s personal computers and rarely open to others. Meanwhile, some basic analysis methods are still not enough for deep insight into the plant electrical signals. In fact, these factors prevent the research development of plant electrophysiology. Hence, the main content of the thesis are following.1. Plant action potential (AP) are induced by non-damaging stimulus and it is the most common signals. However, it is difficult for researchers and beginners to distinguish plant electrical signals types by eyes only. Here, for AP with refractory period, variability, discontinuity, noise and artifacts, we proposed a wave detection algorithm based on first-order derivative with threshold and a classification algorithm based on template matching. Firstly, dynamic thresholds can help to detect all similar AP waves. Then, these waves were classified by incremental template matching to get the real AP waves. Meanwhile, we extracted 19 features from AP waves in time-domain, frenquency-domain, statistical and nonlinear analysis methods. The common classifiers includes backpropagation artificial neural networks (BP-ANNs), supported vector machine (SVM) and deep learning method were also used to learn the training data and classify the AP waves. The results showed that template matching algorithm had the highest performance which accuracy is 96% among the four classifers.2. By employing tranditional Granger Causality Analysis (GCA) and transfer entropy (TE) methods, I explored the plant electrical signals causality networks inference on the optical recording data and MEA data. The analysis were executed on 3 levels. On cell level, building the causal networks using fluorescence image series of plant guard cells. On multi-cell level, discovering the causality networks in phloem using fluorescence images of Helianthus annuus. L (H. annuus) stem tissue. On leaf level, inferring the causal networks employing MEA data of H. annuus leaf. The contrast experiment was used to validate whether there was causal network for electrical activity in leaf if the leaf petiole was killed.3. The Web-based multi-source plant electrophysiological data sharing and online analysis platform was realized. Firstly, Hadoop distributed file system (HDFS) was used to store plant fluorescence image series, extracellular recording data files and MEA data. For small files storage, such as images and extracellular files, MapFile was used to merge small files. And testing the read/write performance of merging files and metadata caching. Secondly, online plant electrical signals extraction and visualization were implemented. MapReduce and Spark were used to extract the plant electrical signals from image series. The visualization includes spatio-temporal visualization and causality network visualization. Thirdly, designing Web-based metadata and data storage, online visualization and annotation. Fourthly, the adaptive derivative threshold algorithm and 12 features extraction were integrated into online analysis. Afterwards, the template matching algorithm was used to classify AP waves online.