In the past, nearest neighbor algorithms for learning from examples have worked best in domains in which all features had numeric values. In such domains, the examples can be treated as points and distance metrics can use standard definitions. In symbolic domains, a more sophisticated treatment of the feature space is required. We introduce a nearest neighbor algorithm for learning in domains with symbolic features. Our algorithm calculates distance tables that allow it to produce real-valued distances between instances, and attaches weights to the instances to further modify the structure of feature space. We show that this technique produces excellent classification accuracy on three problems that have been studied by machine learning researchers: predicting protein secondary structure, identifying DNA promoter sequences, and pronouncing English text. Direct experimental comparisons with the other learning algorithms show that our nearest neighbor algorithm is comparable or superior ...

ABSTRACT We have developed an auto-matic algorithm STRIDE for protein secondary structure assignment from atomic coordinates based on the combined use of hydrogen bond energy and statistically derived backbone tor-sional angle information. Parameters of the pattern recognition procedure were optimized using designations provided by the crystallog-raphers as a standard-of-truth. Comparison to the currently most widely used technique DSSP by Kabsch and Sander (Biopolymers 222577-2637, 1983) shows that STRIDE and DSSP as-sign secondary structural states in 58 and 31% of 226 protein chains in our data sample, re-spectively, in greater agreement with the spe-cific residue-by-residue definitions provided by the discoverers of the structures while in 11 % of the chains, the assignments are the same. STRIDE delineates every 11 th helix and every 32nd strand more in accord with published assignments. Q 1995 Wiley-Liss, Inc. Key words: protein structure analysis, hydro-gen bond, torsional angle, a-helix, p-sheet

*Corresponding authors

The prediction of protein secondary structure by use of carefully structured neural networks and multiple sequence alignments has been investigated. Separate networks are used for predicting the three secondary structures ff-helix, fi-strand and coil. The networks are designed using a priori knowledge of amino acid properties with respect to the secondary structure and of the characteristic periodicity in ff-helices. Since these single-structure networks all have less than 600 adjustable weights over-fitting is avoided. To obtain a three-state prediction of ff-helix, fi-strand or coil, ensembles of single-structure networks are combined with another neural network. This method gives an overall prediction accuracy of 66.3% when using seven-fold cross-validation on a database of 126 non-homologous globular proteins. Applying the method to multiple sequence alignments of homologous proteins increases the prediction accuracy significantly to 71.3% with corresponding Matthews' correlation c...

We describe a method for using machine learning to refine algorithms represented as generalized finite-state automata. The knowledge in an automaton is translated into an artificial neural network, and then refined with backpropagation on a set of examples. Our technique for translating an automaton into a network extends kbann, a system that translates a set of propositional rules into a corresponding neural network. The extended system, FSkbann, allows one to refine the large class of algorithms that can be represented as state-based processes. As a test, we use FSkbann to refine the Chou-Fasman algorithm, a method for predicting how globular proteins fold. Empirical evidence shows the refined algorithm FSkbann produces is statistically significantly more accurate than both the original Chou-Fasman algorithm and a neural network trained using the standard approach. Introduction As machine learning has been increasingly applied to complex real-world problems, many researchers have...

Abstract. Wild-type and mutant chicken integrin/31 subunit (/31o) cDNAs were expressed in NIH 3T3 cells and assayed for localization in focal adhesions of cells plated on fibronectin substrates. Focal adhesion localization in stable transfected cells was assayed by indirect immunofluorescent staining with chicken-specific anti-/3L: antibodies. Mutant/31c integrins containing internal deletions of 13 amino acids adjacent to the membrane, A759-771, and 20 centrally located amino acids, A771-790, localized in focal adhesions demonstrating that sequences required for direction to focal adhesion structures were not limited to one region of the cytoplasmic domain. Point mutations revealed three clusters of amino acids which contribute to localization in focal adhesions. These three clusters or signals

Abstract. Protein localization to the TGN was investigated by examining the subcellular distribution of chimeric proteins in which the cytoplasmic and/or transmembrane domains of the TGN protein, TGN38, were substituted for the analogous domains of the plasma membrane protein, Tac. Using immunofluorescence and immunoelectron microscopy, the COOHterminal cytoplasmic domain of TGN38 was found to be sufficient for localization of the chimeric proteins to the TGN. Deletion analysis identified an l 1-amino acid segment containing the critical sequence, YQRL, as being sufficient for TGN localization. TGN localization was abrogated by mutation of the tyrosine or leu-cine residues in this sequence to alanine, or of the arginine residue to aspartate. In addition to specifying TGN localization, the ll-amino acid segment was active as an internalization signal, although the property of internalization alone was insufficient to confer TGN localization. Overexpression of chimeric proteins containing TGN localization determinants resulted in their detection at the plasma membrane and in intracellular vesicles, and abolished detection of endogenous TGN38. These results suggest that discrete cytoplasmic determinants can mediate protein localization to the TGN, and reveal a novel role for tyrosine-based motifs in this process. T HE Golgi complex plays a central role in the intracellular transport, processing and sorting of proteins in the secretory pathway. Morphological and functional studies have established that the Golgi complex of higher eukaryotes is organized into at least three contiguous but distinct regions: the cis-Golgi network (CGN) t, the Golgi stack and the TGN (reviewed by Mellman and Simons, 1992; Rothman and Orci, 1992). The CGN is a collection of tubules connected to the cis-Golgi cisternae which serves as the entry site for proteins transiting from the ER into the Golgi complex. Itinerant proteins entering the CGN are either recycled back to the ER or routed forward into the Golgi

Abstract. Chicken integrin B1 eDNA and its sitedirected mutants were cloned into a mammalian expression vector and introduced into mouse NIH 3T3 cells. Stable transfectants expressing the chicken/31 subunit or its site-directed mutants were identified by immunostaining with antibodies specific for the chicken integrin/31 subunit. The chicken/~1 proteins were expressed predominately in the endoplasmic reticulum of transfectants and to a lesser degree in the plasma membrane. Immunoblots and immunoprecipitations, using anti-chicken integrin antibodies, revealed three different sizes of the chicken subunit (90, 95, and 120 kD) and a mouse 140-kD cz subunit. Immunoprecipitations of the cell surface receptors showed only two peptides, an 120-kD/31 and an 140kD

Abstract. Thrombospondin is a 420,000-D glycoprotein that has recently been shown to have several properties in common with the members of a class of adhesive proteins. To characterize further the adhesive properties of thrombospondin, we have studied its ability to support cell attachment. Thrombospondin adsorbed to plastic dishes supports the attachment of human endothelial and smooth muscle cells and the monocyte-like cell line (U937) as well as normal rat kidney cells. The majority of attached cells do not spread on the solid-phase thrombospondin. The attachment of all four cell types to thrombospondin is abolished if the assay is performed in the presence of EGTA, although the cells still attach to fibronectin. If thrombospondin is adsorbed to the dishes in the presence of EGTA and then washed with buffer containing calcium before addition of the cells, attachment is still markedly inhibited, indicating that calcium affects the conformation and function of thrombospondin. Attach-ment of all four cell types is also markedly inhibited by the synthetic peptides gly-arg-gly-asp-ser-pro (GRG-DSP) and gly-arg-gly-asp-ala-cys (GRGDAC) but not by the control peptide gly-arg-gly-glu-ser-pro (GRG-ESP). Affinity chromatography of n-octylglucoside extracts of surface-labeled endothelial cells or smooth muscle cells on thrombospondin-Sepharose and GRG-DSP-Affigel columns was used to identify an integrin complex related to glycoprotein IIb-IIIa as an RGDdependent receptor for thrombospondin. In addition, a monoclonal antibody (LM609) that blocks attachment of endothelial cells to vitronectin, fibrinogen, and von Willebrand factor also inhibits attachment of endothelial cells to thrombospondin. These data indicate that the attachment of cells to thrombospondin is mediated by RGD and calcium-dependent mechanisms and is conistent with the hypothesis that the GRGDAC sequence in thrombospondin is a site for interaction with an integrin receptor of the 133 subclass.

Abstract. The complete primary structure of an integral membrane glycoprotein of the nuclear pore was deduced from the eDNA sequence. The cDNA encodes a polypeptide of 204,205 D containing a 25residue-long signal sequence, two hydrophobic segments that could function as transmembrane segments, and 13 potential N-linked oligosaccharide addition sites. Endoglycosidase H reduces the molecular mass by '~9 kD suggesting that not all of these 13 sites are used. We discuss possible models for the topology of this protein in the pore membrane as well as a possible role in the formation of pores and pore complexes. T H F. outer and inner membranes of the nuclear envelope are continuous with each other at distinct circular sites