Title: Digest the data: towards reliable biology-specific protein-protein Interaction  

Speaker: Prof. Bin XUE, Department of Cell Biology, Microbiology and Molecular Biology, University of South Florida

Time: June 6th, Thursday |4:00- 5:30pm

Venue: Building 24#-C406

Host: Prof. Peter R. Taylor

Abstract: 

Protein-protein interaction (PPI) is a dominant factor of various biological processes.

However, determining PPIs that are specific to biological conditions is still challenging.PPI data obtained from wet-lab experiments could be specific to the tissue, cell, and/or conditions used in experiments. Further, many experimental techniques are designed for detecting stable and strong interactions, and therefore may not be fully effective when being used for intrinsically disordered proteins (IDPs) that do not have stable 3D structures and frequently interact with partners through transient and weak interactions. In this project, it was demonstrated that experimental PPI data deposited into databases are largely not consistent, and about 15% of the IDs used in several major PPI DBs are mislabeled. Therefore, a new computational pipeline was developed to integrate existing PPI databases and to remove the mislabled IDs. To improve the detection of PPI involving IDPs, a new meta-strategy was designed to predict sequence motifs of IDPs, which interact with other protein partners. The proto-type new predictor improved the prediction performance remarkably comparing to other motif predictors. In addition, a genotypic distance method based on gene expression levels was designed to identify genotypically-and-phenotypically-significant differentially-expressed-genes (gpsDEGs). The new method effectively improved the correctness of DEG identification. The gpsDEGs can be identified using microarray, RNAseq, or mass spectrometry data as input. Further combination of  PPI and gpsDEGs leads to PPIs that are specific to biological conditions. Since the gpsDEGs are mostly identified at RNA levels, it is necessary to examine gene expression at the protein level. For this purpose, a novel machine-learning tool combining meta-strategy, sample space reorganization, two-step significance-voting was developed to predict miRNA: mRNA interactions. The new method improved the accuracy of miRNA: mRNA interaction significantly. The predicted miRNA: mRNA interactions were integrated into the afore-mentioned PPI to show the complex regulation between miRNA and mRNA, as well as the impact on PPIs at the protein level.