| The complex gene interactions within cells can be formulated by the theory of network biology. Underlying efficient plant immune response is the sophisticated collaboration among myriad immune-related genes. Therefore, network biology approaches can be used to dissect the shared and distinct aspects of different types of plant immunity. Pattern-triggered immunity (PTI) and effector-triggered immunity (ETT) are two main plant immune responses to counter pathogen invasion. Genome-wide gene network organizing principles leading to quantitative differences between PTI and ETI have remained elusive. I combined an advanced machine learning method and modular network analysis to systematically characterize the organizing principles of Arabidopsis PTI and ETT at three network resolutions. At the single network node/edge level, I ranked genes and gene interactions based on their ability to distinguish immune response from normal growth and successfully identified many immune-related genes associated with PTI and ETI. Topological analysis revealed that the top ranked gene interactions tend to link network modules. At the subnetwork level, I identified a subnetwork shared by PTI and ETI encompassing1,156genes and1,289interactions. This subnetwork is enriched in interactions linking network modules and is also a hotspot of attack by pathogen effectors. The subnetwork likely represents a core component in the coordination of multiple biological processes to favor defense over development. Finally, I constructed modular network models for PTI and ETI to explain the quantitative differences in the global network architecture. The results indicate that the defense modules in ETI are organized into relatively independent structures, explaining the robustness of ETI to genetic mutations and effector attacks. Taken together, the multiscale comparisons of PTI and ETI provide a systems biology perspective on plant immunity and emphasize coordination among network modules to establish a robust immune response.In addition to plant immune system, the arms race between plants and pathogens also profoundly influences the infection strategies of pathogens. As one of the most important virulence factors in gram-negative pathogenic bacteria, type-Ⅲ effectors (T3SEs) play a crucial role in plant-pathogen interactions by suppressing plant immune response. But the fast evolutionary rate of T3SEs impedes the use of traditional bioinformatics methods such as sequence similarity searches to correctly identify them from newly sequenced pathogen genomes. In my work, by combing profile-based amino acid pair information and machine-learning algorithm I developed an accurate T3SE predictor BEAN (Bacterial Effector Analyzer). The innovation of BEAN is using a hidden Markov model-based sequence searching method (i.e., HHblits) to detect a T3SE’s homologous sequences and extract the profile-based k-spaced amino acid pair composition (HH-CKSAAP) from the N-terminal sequences. This encoding approach can amplify the signals of weakly conserved residues in the T3SE. As a result, BEAN achieves a performance with sensitivity=78%and specificity=96%, and is better than other four reported T3SE predictors in independent test. The test on a plant pathogen genome also confirmed its value for genome-wide study. To further characterize the composition of type-Ⅲ secretion signal, I performed evolutionary rate analysis and position bias analysis on the predictive amino acid pairs. The results imply weakly conserved sequence motifs may be important composition of type-Ⅲ secretion signal.In summary, this work demonstrates the power of network biology and machine learning in deepening our understanding of plant-pathogen interactions. It is hoped that these approaches can be readily applied to answer other important biological questions. |
PDF Full Text Request