Biological prediction by integrated probabilistic models

Harvard Medical School, Boston MA

Investigators

Linked publications & trials

Abstract

[unreadable] DESCRIPTION (provided by applicant): The use of computational techniques for making biological predictions of gene attributes or interactions serves an important role in the scientific community. Due to the rapid growth of available data, it is not feasible to perform every desired experiment using wet-lab techniques. Computational predictions can thus be used to prioritize hypotheses, permitting more efficient use of experimental resources. Existing prediction techniques often make use of similar information to the variable being predicted (e.g. when predicting a gene attribute, use other known attributes of that gene; or more commonly the guilt-by-association approach with other genes, such as sequence homology). A larger probabilistic model that makes use of more information is to be extended during this research, with the goal of utilizing as much available information as possible to heighten the accuracy of the predictive methods. In particular, the incorporation of multi-organism data, the analysis of "missing data" (i.e. unperformed experiments), and alternative modeling approaches will be studied. Furthermore, predictions that are interesting and novel will be pursued using traditional wet-lab experiments (either in-house or through collaborators) in an attempt to make discoveries of practical value to biologists (following the precedent set by the Roth Lab, e.g. the rRNA processing collaboration with Tim Hughes). [unreadable] [unreadable] [unreadable]