Combining Bayesian and Meta-heuristic approaches for the Protein Inference Problem

Abstract

Protein inference refers assembling peptides identified from tandem mass spectra into a list of proteins. Due to the existence of degenerate peptides, it is very difficult to determine which proteins are present in the sample. This problem is called protein inference problem and it represents a major challenge in shotgun proteomics as well as in proteomics research. Many approaches have been introduced for solving protein inference problem. In this paper, we have combined Bayesian and Meta-heuristic approaches for solving protein inference problem. Meta-heuristic approaches provide a very fast and efficient heuristic search strategy to infer proteins with reasonable accuracy and precision. It provides the flexibility to infer proteins either parsimoniously or optimistically or somewhere between the two by taking some tuning parameters. On the other hand Bayesian model provides a probabilistic model that incorporates the predicted peptide detectabilities as the prior probabilities of peptide identification. We proposeacombinationofthesetwoapproaches. WeshoweditbycombiningMAgPI(A Memetic Algorithm Based Approach in Protein Inference Problem) as Meta-heuristics approach and Gibbs Sampler for protein inferencing as Bayesian approach. In our system,GibbsSamplerisprocessingtheinputofMAgPIandfinallyMAgPIisrefiningthe input. As, our input is going through two refining processes, the final output is better thanothersinseveralaspects. Anotherimportantfactisthatinoursystemcomputation time of MAgPI is less than before as after first stage of refining, the size of candidate solutionbecomesveryshort. WeusedSigma49datasettotestourmethodandgotgood result in several aspects.