KARP@SUMEX-AIM.STANFORD.EDU (Peter Karp) (10/27/88)
The following summarizes my current research in Artificial Intelligence in the Stanford Computer Science Department. I am a PhD student working in the areas of Machine Learning and Qualitative Reasoning. My thesis work involves building a computer system to reproduce some of the reasoning used by Dr. Charles Yanofsky and other workers in discovering the bacterial gene regulation mechanism called attenuation. The work has three major components. First, I performed a historical study of Yanofsky's research to understand what reasoning processes biologists used to make these discoveries. This study includes a comprehensive survey of the literature and many interviews with Yanofsky, his co-workers, and researchers in other laboratories. Second, I constructed a knowledge-based simulation system which models the E. coli tryptophan operon (the system in which attenuation was studied). This system embodies a theory of the trp operon which can predict the results of experiments on the trp operon. This knowledge base is currently rather small in size (roughly 200 objects), but constructing it provided important lessons in the representation of biological knowledge, and raised a number of issues which must be addressed in the construction of large knowledge bases. The system includes models of entities such as the trp operon, and the processes of of transcription and translation, and the biosynthesis of tryptophan. These models are fairly abstract, e.g., they contain no DNA or protein sequence information but represent DNA functional units such as operators, promoters and ribosome binding sites. The system was constructed on a Xerox Lisp machine running Interlisp using Intellicorp's KEE frame (objected oriented) knowledge representation tool. Third, I am building a hypothesis generation system which takes as input (a) a description of an experiment whose outcome is predicted incorrectly by the simulation system above, and (b) the model of the trp operon used by the simulation system. Its output is a set of hypotheses which alter both the theory and the initial conditions of the experiment (e.g., by postulating the existence of mutations) to produce a correct prediction. These hypotheses are synthesized by a program which views hypothesis generation as a design problem, and employs Artificial Intelligence techniques from design and planning to generate hypotheses. Further information can be obtained from the references below. My dissertation will be a comprehensive description of this research, and is in preparation. P. Karp, "A Process-Oriented Model of Bacterial Gene Regulation", 1988 Stanford Knowledge Systems Laboratory Technical Report KSL-88-18, 14 pages. Friedland, P., and Kedes, L. "Discovering the secrets of DNA", Communications of the ACM, 28(11):1164-1186, November, 1985. -------