This chapter will discuss one example of how AI techniques are being integrated with, and extending, existing molecular biology sequence analysis methods. AI ideas of complex representations, pattern recognition, search, and machine learning have been applied to the task of inferring and recognizing structural patterns associated with molecular function. We wish to construct such patterns, and to recognize them in unknown molecules, based on information inferred solely from protein primary (amino acid) sequences.

Our understanding of any process can be measured by the extent to which a simulation we create mimics the real behavior of that process. Deviations of a simulation indicate either limitations or errors in our knowledge. In addition, these observed differences often suggest verifiable experimental hypotheses to extend our knowledge. The biochemical approach to understanding biological processes is essentially one of simulation. A biochemist typically prepares a cell-free extract that can mediate a well-described physiological process. The extract is then fractionated to purify the components that catalyze individual reactions.

Human scientists actively seek out information that bears on questions they have decided to pursue. They design experiments, explore the implications of the knowledge they have, refine their questions and test alternative ideas. Although many discoveries are the result of unexpected observations, these surprises take place in the context of an explicit pursuit of knowledge. Viewing scientific discovery as a kind of motivated action raises some basic issues common to goal-directed behavior generally: Where do desires (to know) come from? What are the actions that can be taken (to discover)? What are the resources those actions consume, and how are they allocated? How are decisions about selecting and combining actions made?

Cells function as organized chemical engines carrying out a large number of transformations, called bioreactions or biochemical reactions, in a coordinated manner. These reactions are catalyzed by enzymes and exhibit great specificity and rates much higher than the rates of non-enzymatic reactions. Enzymes are neither transformed nor consumed, but that facilitate the underlying reactions by their presence. The coordination of the extensive network of biochemical reactions is achieved through regulation of the concentrations and the specific activities of enzymes. Single enzymecatalyzed steps in succession form long chains, called biochemical pathways, achieving the overall transformation of substrates to far removed products.

Cohen, Design Considerations for a Program to Provide Consultations in Clinical Therapeutics, in Proceedings of San Diego Biomedical Symposium, February 1974.

Cohen, Design Considerations for a Program to Provide Consultations in Clinical Therapeutics, in Proceedings of San Diego Biomedical Symposium, February 1974.

The RNA secondary structure prediction problem (2 RNA) is a critical one in molecular biology. Secondary structure can be determined directly by x-ray diffraction, but this is difficult, slow, and expensive. Moreover, it is currently impossible to crystallize most RNAs. Mathematical models for prediction have therefore been developed and these have led to serial (and some parallel) computer algorithms, but these too are expensive in terms of computation time. The general solution has asymptotic running time exponential in N (i.e., proportional to 2 N), where N is the length of the RNA sequence. Serial approximation algorithms which employ heuristics and make strong assumptions are significantly faster, on the order of N 3 or N 4, but their predictive success rates are low -- often less than 40 percent -- and even these algorithms can run for days when processing very long (thousands of bases) RNA sequences. Neural network algorithms that perform a multiple constraint satisfaction search using a massively parallel network of simple processors may provide accurate and very fast solutions.

Shortly after Watson and Crick's discovery of the structure of DNA, and at about the same time that the genetic code and the essential facts of gene expression were being elucidated, the field of linguistics was being similarly revolutionized by the work of Noam Chomsky [Chomsky, 1955, 1957, 1959, 1963, 1965]. Observing that a seemingly infinite variety of language was available to individual human beings based on clearly finite resources and experience, he proposed a formal representation of the rules or syntax of language, called generative grammar, that could provide finite--indeed, concise--characterizations of such infinite languages. Just as the breakthroughs in molecular biology in that era served to anchor genetic concepts in physical structures and opened up entirely novel experimental paradigms, so did Chomsky's insight serve to energize the field of linguistics, with putative correlates of cognitive processes that could for the first time be reasoned about 48 A

The prediction of protein structure from amino acid sequence has become the Holy Grail of computational molecular biology. Since Anfinsen [1973] first noted that the information necessary for protein folding resides completely within the primary structure, molecular biologists have been fascinated with the possibility of obtaining a complete three-dimensional picture of a protein by simply applying the proper algorithm to a known amino acid sequence. The development of rapid methods of DNA sequencing coupled with the straightforward translation of the genetic code into protein sequences has amplified the urgent need for automated methods of interpreting these one-dimensional, linear sequences in terms of three-dimensional structure and function. Although improvements in computational capabilities, the development of area detectors, and the widespread use of synchrotron radiation have reduced the amount of time necessary to determine a protein structure by X-ray crystallography, a crystal structure determination may still require one or more man-years.

He also taught the biochemistry course that I finally took, two years after finishing my Ph.D. David J. States deserves much of the credit as well. In the three years we have been working together, he greatly extended my understanding of not only what biologists know, but how they think. He has read several drafts of this chapter and made helpful suggestions. David Landsman, Mark Boguski, Kalí Tal and Jill Shirmer have also read the chapter and made suggestions. Angel Lee graciously supplied the gel used in Figure 4. Of course, all remaining mistakes are my responsibility.