Service List
CDEN (http://liulab.hzau.edu.cn/CDEN/)
Although intensive efforts have been devoted to investigating latent tuberculosis (LTB) and active tuberculosis (PTB) infections, the similarities and differences in the host responses to these two closely associated stages remain elusive, probably due to the difficulty in identifying informative genes related to LTB using traditional methods. Herein, we developed a framework known as the consistently differential expression network (CDEN) to identify TB-related gene pairs by combining microarray profiles and protein-protein interactions. We thus obtained 774 and 693 pairs corresponding to the PTB and LTB stages, respectively. The PTB-specific genes showed higher expression values and fold-changes than the LTB-specific genes. Furthermore, the PTB-related pairs generally had higher expression correlations and would be more activated compared to their LTB-related counterparts. The module analysis implied that the detected gene pairs tended to cluster in the topological and functional modules. Functional analysis indicated that the LTB- and PTB-specific genes were enriched in different pathways and had remarkably different locations in the NF-κB signaling pathway. Finally, we showed that the identified genes and gene pairs have the potential to distinguish TB patients in different disease stages and could be considered as drug targets for the specific treatment of patients with LTB or PTB.
MTB-human PPI (http://liulab.hzau.edu.cn/MTB/)
Tuberculosis (TB) is one of the biggest infectious disease killers caused by Mycobacterium tuberculosis (MTB). Studying the protein-protein interactions (PPIs) between MTB and human can deepen our understanding of the pathogenesis of MTB and offer new clues for the treatment against MTB infection, but the experimentally validated interactions are especially scarce in this regard. Herein we proposed an integrated framework that combined template-, domain-domain interaction-, and machine learning-based methods to predict MTB-human PPIs. As a result, we established a network composed of 13,758 PPIs including 451 MTB proteins and 3,167 human proteins. Compared to known human targets of various pathogens, our predicted human targets show a similar tendency in terms of the network topological properties and enrichment in important functional genes. Additionally, these human targets commonly have longer sequence lengths, more protein domains, more disordered residues, lower evolutionary rates, and older protein ages. Functional analysis demonstrates that these proteins show strong preferences toward the phosphorylation, kinase activity, and signaling transduction processes and are clustered in the disease and immune related pathways. Dissecting the cross-talk among top-ranked pathways suggests that the cancer pathway may serve as a bridge in MTB infection. Triplet analysis illustrates that the paired targets interacting with the same partner are adjacent to each other in the intra-species network and tend to share similar expression patterns. Finally, we found 36 potential anti-MTB human targets by integrating known drug target information and molecular properties of proteins.
PCTpred(http://liulab.hzau.edu.cn/PCTpred/)
Post-translational modification (PTM) based regulation can be mediated not only by modification of a single residue but also by cross-talk among different modifications. Accurate prediction of PTM cross-talk is a highly challenging issue and still in its infant stage. Especially, less attention has been paid to the structural preferences (except intrinsic disorder and spatial proximity) of cross-talk pairs and the characteristics of individual residues involved in cross-talk, which might restrict the improvement of prediction accuracy. Here we report a structure-based algorithm called PCTpred for improving PTM cross-talk prediction. The comprehensive residue- and residue pair-based features were designed for paired PTM sites at the sequence and structural level. Through feature selection, we reserved 23 newly introduced descriptors combined with three traditional ones to establish a sequence-based predictor PCTseq and a structure-based predictor PCTstr, both of which were integrated to build our final prediction model. According to pair- and protein-based evaluations, PCTpred obtained area under the curves of around 0.9 and 0.8, respectively. Even when removing the distance preference of samples or using the input of modelled structures, our prediction performance was maintained or moderately reduced. More importantly, PCTpred showed stable and reliable improvements over the state-of-the-art methods based on various evaluations. The source code and dataset are available at https://github.com/Liulab-HZAU/PCTpred.
RPAIAnalyst (http://liulab.hzau.edu.cn/RPAIAnalyst/)
Accurate discrimination between biological and crystal interfaces remains challenging in structural bioinformatics. Most existing algorithms focused on the features from a single side of the interface. However, very few works have exploited the properties of residue pairs across protein interfaces. Here we developed a prediction model called RPAIAnalyst to solve this problem. We first defined a novel co-evolutionary feature by combining direct coupling analysis and image processing technique. The residue pairs across biological interfaces were significantly enriched in co-evolving residues compared to those across crystal contacts, resulting in a promising classification accuracy when using this feature individually. Regarding the preference of residue pairs across interfaces, we picked out the most informative residue couplings and the optimized feature set can improve the prediction performance. Further, for the other traditional properties, we not only constructed the descriptors from one side of the interface but also generated novel descriptors that consider coupling across the interface. The complementarity between them can contribute to the superior performance. By integrating all residue pair based features, we established our final prediction model which showed stable and excellent performance on different datasets. Compared to other methods, our algorithm not only yielded comparable or better prediction results but also provided complementary signature.
CRHunter (http://liulab.hzau.edu.cn/CRHunter/)
Enzymes play indispensable roles in catalyzing biochemical reactions in living organisms. Intensive efforts have been devoted to the computational prediction of catalytic residues in enzymes individually utilizing feature- or template-based strategy, but there are no studies that systematically compare the strengths and limitations of these two strategies and further consider whether their combination can be utilized to enhance the prediction performance. Herein we established the first integrative algorithm, called CRHunter, by simultaneously utilizing the complementarity between feature- and template-based strategies and that between structural and sequence information. The Delaunay triangulation and Laplacian transformation were first used to characterize enzyme structures, resulting in several novel structural features. Combining them with traditional descriptors, we developed two support vector machine feature predictors individually based on structural and sequence information. Meanwhile, we invented two template predictors by respectively using structure and profile alignments. Evaluated on the datasets with different levels of homology, our feature predictors can achieve relatively stable performance, whereas our template predictors yield poor results as the homological constraint increases. Even so, the hybrid algorithm CRHunter consistently achieves the highest prediction accuracy among all our proposed predictors, indicating the importance of integrating different strategies. We further demonstrate that our proposed methodology can also be applicable to the simulated structures of enzymes, which is extremely useful for the query proteins having only sequence information. Compared to the state-of-the-art methods, our algorithm shows obvious advantages on various datasets, suggesting that CRHunter is an effective and efficient web tool for predicting catalytic residues.
RBRDetector (http://liulab.hzau.edu.cn/rbrdetector/)
RBRDetector is a novel structure-based algorithm to identify RNA-binding residues by combining feature- and template-based prediction strategies. Based on the well-designed evolutionary and structural features, we develop a feature-based method that is an ensemble of ten SVM-based predictors by integrating two types of residue microenvironment information with five different training sets obtained by re-sampling. Alternatively, we propose a template-based method by structurally aligning the query protein to the RNA-binding proteins with known structures. Further, considering the complementary relationship between these two methods, we construct an integrative prediction model by combining them with a piecewise function. The flowchart of our RBRDetector algorithm is given below. The details about how to implement our integrative predictor can be found in our manuscript. The extensive experiments on various structural data demonstrate that our integrative algorithm shows clearly better predictive capability compared to its component methods as well as other state-of-the-art algorithms, indicating that RBRDetector is a powerful tool for predicting RNA-binding residues.