Logical data fusion for biological hypothesis evaluation

We use techniques from finite model theory to construct a framework for hypothesis creation and ranking to aid biologists with hypothesis evaluation and experimental design. Most bioinformatics research is geared toward pattern recognition and biological database management. Our work has somewhat different aims. First, we seek to determine the structure of the space of biological hypotheses that can be composed about a given system. Second, we seek to combine a wide variety of experimental data and literature sources for use in "proofreading" such hypotheses. This data fusion problem has been a major stumbling block in modeling biological pathways. Consequently, most modeling frameworks make use of only one or two types of data, typically promoter sequences and microarray data. We present a modeling framework that is contradiction based and that performs data fusion on the logical level for an arbitrary number of sources. This greatly facilitates the incorporation of new data sources as they become available. Once a new hypothesis has been constructed, data from existing experimental databases can be fused to rank the hypothesis based on corroborating and contradictory experimental evidence. We demonstrate the logical underpinnings of this process, and show how inflationary and deflationary logical extensions alter the process.