Artificial Intelligence and Molecular Biology (American Association for Artificial Intelligence)
Format: PDF / Kindle (mobi) / ePub
The enormous amount of data generated by the Human Genome Project and other large-scale biological research has created a rich and challenging domain for research in artificial intelligence. These original contributions provide a current sampling of AI approaches to problems of biological significance; they are the first to treat the computational needs of the biology community hand-in-hand with appropriate advances in artificial intelligence. Focusing on novel technologies and approaches, rather than on proven applications, they cover genetic sequence analysis, protein structure representation and prediction, automated data analysis aids, and simulation of biological systems. A brief introductory primer on molecular biology and Al gives computer scientists sufficient background to understand much of the biology discussed in the book.Lawrence Hunter is Director of the Machine Learning Project at the National Library of Medicine, National Institutes of Health.
purpose of this section is to describe some of the basic experimental methods of molecular biology. These methods are important not only in understanding the source of possible errors in the data, but also because computational methods for managing laboratory activities and analyzing raw data are another area where AI can play a role (see the chapters by Edwards, et al and Glasgow, et al, in this volume). I will also describe some of the many online information resources relevant to computational
the automaton has to guess whether it has encountered the center of the palindrome at any point, and can begin popping the stack. Any nondeterministic FSA may be converted to a deterministic FSA, though obviously the same cannot be said of PDAs. Thus, the deterministic subset of CFLs properly contains the RLs. 2.2.3 Ambiguity. Another useful distinction within the CFLs concerns the notion of ambiguity. Formally, we say that a grammar is ambiguous if there is some string for which more than one
cylinder at the bottom), β-strands (the arrows in the center), and β-turns (the “kink” iconified at the upper right). Pattern-directed inference systems like Ariadne [Lathrop, Webster and Smith, 1987] have been used to detect amino acid sequences that are likely to produce such structures, combining statistical evidence for the features themselves with a hierarchical model of their higher-level ordering, captured in what amounts to a regular expression. Such an approach must necessarily deal with
architecture may be defined by specifying the following (list taken from [McClelland, Rumelhart and the PDP research group, 1986]): 138 ARTIFICIAL INTELLIGENCE & MOLECULAR BIOLOGY • A set of processing units • A set of activation states for the units • An output function for each unit • A pattern of connectivity among units • A propagation rule for propagating patterns of activity through the network • An activation rule for combining a unit’s inputs with its activation level to produce a new
dismutase. Furthermore, to base positive disulfide bond predictions on high cysteine content and even parity result in failures for ferredoxins, metallothioneins, and some cytochromes. Clearly, predictions based on these simple rules fail to capture the unique micro-environments a protein structure imposes on its cysteines to define their disulfide bonding states. Recently, Muskal et al.  used a network of the architecture seen in Figure 9 to predict a cysteine’s disulfide bonding state,