Mitochondrial protein cleavage site predictor

Summary form only given. We have constructed a new tool to predict the cleavage sites of mitochondrial proteins. Trained on newer data, in preliminary results our tool compares favorably with the industry standard tool TargetP. In most eukaryotic cells, over 1000 nuclear-encoded proteins are translocated into the mitochondria to perform their function. Roughly half of these have an N-terminal sorting signal known as the MTS (Matrix Targeting Signal) and most of those are cleaved after mitochondrial import by MPP (Mitochondrial Protein Peptidase) and other peptidases. The position of these cleavage sites is necessary to know the amino acid sequence of the mature protein. TargetP is a classic tool for predicting MPP cleavage sites. However, recently large-scale experiments in yeast gave us an opportunity to create a new classifier based on much more data. Ultimately we hope to predict all of the cleavage events experienced by proteins as they mature in the mitochondria, but in our poster we will describe the framework shown in figure 2, encompassing the peptidases MPP, Octl, and Icp55. MPP typically cleaves one residue after an arginine. Subsequently, some proteins are further cleaved by Octl (cleaving eight residues) or by Icp55 (cleaving one residue). Figure 1 shows the sequence logos for the inferred cleavage sites of each peptidase and an extra group of currently unexplained data. As in Vogtle et al, we have inferred some intermediate sites and the identity of the peptidases based on the position of arginines. Currently we have had the most success with predicting MPP sites. Our first attempt uses a profile HMM (Hidden Markov Model) as implemented by the HMMER2 package (hmmer.janelia.org/). Preliminary cross-validation results suggest that our new predictor outperforms TargetP for this task. We are also exploring other techniques including SVM classifiers and Dynamic Bayesian networks.