Bayesian Inference on Hidden Knowledge in High-Throughput Molecular Biology Data

Along with the information overload brought about by the Internet in communication, economics, and sociology, high-throughput biology techniques produce vast amount of data, which are usually represented in form of matrices and considered as knowledge networks. A spectral based approach has been proved useful in extracting hidden information within such networks to estimate missing data. In this paper, we propose the use of a simple nonparametric Bayesian model to fully automate this approach and better utilize the available data at each stage of the learning process. Although the algorithm is developed with a general purpose in mind, within the scope of this paper, we evaluate its performance by applying on three different examples from the field of proteomics and genetic networks. The comparison with other general or data-specific methods has shown favor to ours. Systematic tests on synthetic data are also performed, showing the approach’s robustness in handling large percentage of missing data both in term of prediction accuracy and convergence rate. Finally, we describe a procedure to explore the nature of different types of noise containing within investigated systems.