ProtAnalyzer: A Customizable Software for Prediction of Kinase Targets in the Complete Proteome
Celina Montemayor, Elizabeth Villegas, Jeffrey Rosen. University of Wisconsin Hospital and Clinics, Madison, WI; Baylor College of Medicine, Houston, TX
Background: Tools for proteome-wide identification of kinase targets are necessary for the development of new pharmacologic approaches. In silico analyses offer a quick, inexpensive screen that allows the targeted design of validation experiments. Hidden Markov Models (HMM) are robust Bayesian statistical tools widely-employed for DNA and protein pattern recognition. We have created ProtAnalyzer, a customizable HMM-based software, and demonstrate its use by predicting novel Polo kinase 2 (Plk2) phosphorylation targets. Plk2 is a candidate tumor suppressor in triple-negative breast cancer, a disease with poor prognosis for which no targeted therapy exists at present. The discovery of Plk2 substrates is a crucial step to identify potential pharmacologic targets to treat this disease.
Design: ProtAnalyzer runs in any desktop computer. It employs the Viterbi algorithm to search for matches to a user-defined HMM in any number of protein sequences, at a specified stringency level (j). Thus, the universe of emission states encompasses all 20 aminoacids, and the number of transitions is unlimited. To model Plk2 substrates, we constructed an HMM based on the known Plk2 mechanism of action. Targets of this kinase must dock at Plk2's Polo-binding domains (PBDs), and are subsequently phosphorylated in a Ser/Thr residue within a specific aminoacid context. Published consensus sequences for both PBD-binding and Plk2-phosphorylation domains were employed to model an HMM that allows for detection of both motifs at flexible locations and distances within the protein. The j value for optimal sensitivity and specificity was determined by running iterations using the published Plk2 substrates as positive controls, as well as with experimentally-validated negative controls and with equivalent but random-generated sequences.
Results: The human proteome was analyzed at a j value of 0.001, yielding a list of candidate Plk2 targets, along with the exact location of their phosphorylation and PBD-binding sites. Experiments to validate these targets in vitro by bimolecular fluorescence complementation and in vivo using Plk2-/- mouse tissue are underway.
Conclusions: ProtAnalyzer is based on HMM principles and can be easily customized to scan entire proteomes for any kind of flexible aminoacid pattern. The list of Plk2 targets obtained from this demonstration represent possible new pharmacologic targets to treat triple-negative breast cancer.
Category: Special Category - Pan-genomic/Pan-proteomic approaches to Cancer
Tuesday, March 20, 2012 9:30 AM
Poster Session III # 270, Tuesday Morning