Accurate Prediction of Drug-Potent Proteins in Critical Diseases
Research Competitiveness Subprogram grant, Louisiana Board of Regents
In order to design an effective drug to combat critical diseases, one needs to first identify the 3D fold (or, native state) of a protein, and then screen these folds against millions of potential ligands in databases to find some with appropriate bindings. This is somewhat true; however, it missed the fact that disorder protein could exist, which does not always have 3D fold, instead disorder protein can have flexible structures, partially or fully, and can change states during binding. Such disorder proteins are found abundant in nature. Thus, we will develop an effective disorder-predictor at first using novel machine learning approaches. Disorder predictor itself alone will be very important for identifying cancer and cardio-vascular disease causing proteins.
We are combining statistical energy function with evolution-derived properties such as predicted torsion angles and solvent accessibility surface area. The resulting energy function significantly enhances its ability for structure discrimination of near-native structures from decoys – however the incorporation of the predicted disorder within its statistics would make the energy function more realistic. With this discriminating energy function, one will also need a very effective sampling algorithm to sample the astronomical search space of protein structure. We already have developed effective sampling algorithms and will improve further.
These components altogether would essentially form effective ab initio protein structure predictor (aiPSP). aiPSP would be an essential tool to identify drug-potent proteins accurately, especially for critical disease (e.g. malaria) where homologous proteins are rare making an homologous predictor unreliable.
Hoque, Tamjidul, "Accurate Prediction of Drug-Potent Proteins in Critical Diseases" (2013). Computer Science - Grants and Contracts. Paper 19.