The microarray explorer tool for data mining of cDNA microarrays: application for the mammary gland.

The Microarray Explorer (MAExplorer) is a versatile Java-based data mining bioinformatic tool for analyzing quantitative cDNA expression profiles across multiple microarray platforms and DNA labeling systems. It may be run as either a stand-alone application or as a Web browser applet over the Internet. With this program it is possible to (i) analyze the expression of individual genes, (ii) analyze the expression of gene families and clusters, (iii) compare expression patterns and (iv) directly access other genomic databases for clones of interest. Data may be downloaded as required from a Web server or in the case of the stand-alone version, reside on the user's computer. Analyses are performed in real-time and may be viewed and directly manipulated in images, reports, scatter plots, histograms, expression profile plots and cluster analyses plots. A key feature is the clone data filter for constraining a working set of clones to those passing a variety of user-specified logical and statistical tests. Reports may be generated with hypertext Web access to UniGene, GenBank and other Internet databases for sets of clones found to be of interest. Users may save their explorations on the Web server or local computer and later recall or share them with other scientists in this groupware Web environment. The emphasis on direct manipulation of clones and sets of clones in graphics and tables provides a high level of interaction with the data, making it easier for investigators to test ideas when looking for patterns. We have used the MAExplorer to profile gene expression patterns of 1500 duplicated genes isolated from mouse mammary tissue. We have identified genes that are preferentially expressed during pregnancy and during lactation. One gene we identified, carbonic anhydrase III, is highly expressed in mammary tissue from virgin and pregnant mice and in gene knock-out mice with underdeveloped mammary epithelium. Other genes, which include those encoding milk proteins, are preferentially expressed during lactation.

Flicker image comparison of 2-D gel images for putative protein identification using the 2DWG meta-database.

With the availability of two-dimensional (2-D) gel electrophoresis databases that have many characterized proteins, it may be possible to compare a researcher's gel images with those in relevant databases. This may lead to the putative identification of unknown protein spots in a researcher's gel with those characterized in a given database, saving the researcher time and money by suggesting monoclonal antibodies to try in confirming these identifications. We have developed two tools to help with this comparison: (1) Flicker, http:/(/)www.lecb.ncifcrf.gov/flicker/, a Java applet program running in the researcher's Web browser, to visually compare their gels against gels on the Internet; and (2) the 2DWG meta-database, http:/(/)www.lecb.ncifcrf.gov/2dwgDB /, a searchable database of locations of 2-D electrophoretic gel images found on the Internet. Recent additions to Flicker allow users to click on a protein spot in a gel that is linked to a federated 2D gel database, such as SWISS-2DPAGE, and have it retrieve a report from that Web database for that protein.

Xconf: a network-based image conferencing system.

People often need to get together to share and discuss small amounts of image and textual data, but this is difficult when they are not located in the same place. One solution to this problem is Xconf, a multimedia computer conferencing groupware tool using existing national and international networks (the Internet). Simultaneous conferencing supports real-time interaction between multiple remote computer displays. Xconf multimedia may include conversational text, images, pointers to objects in images, and group execution of programs. Conferencing may take place with or without images. Interaction is tightly coupled with all users aware of global changes to the shared session and alternatively, individuals may monitor a specific subgroup of other users to concentrate on what they are discussing. Collaborative groups who have access to both computer networks and networked based X-Window System graphics displays can participate in a conference. Xconf is an X-Window "client" program which provides multimedia conferencing support for a group of X-Window displays. Because it is centralized, no software other than the standard X-Window System is required on any of the participants display systems. Key data structures and algorithms for image conferencing are present.

Comparing two-dimensional electrophoretic gel images across the Internet.

Scientists around the world often work on similar data so the need to share results and compare data arises periodically. We describe a method of comparing two two-dimensional (2-D) protein gels of similar samples created in different laboratories to help identify or suggest protein spot identification. Now that 2-D gels and associated databases frequently appear on the Internet, this opens up the possibility of visually comparing one's own experimental 2-D gel image data with data from another gel in a remote Internet database. In general, there are a few ways to compare images: (i) slide one gel (autoradiograph or stained gel) over the other while back-illuminated, or (ii) build a 2-D gel computer database from both gels after scanning and analyzing these gels. These are impractical since in the first case the gel from the Internet database is not locally available. In the second, the costs of building a multi-gel database solely to answer the question of whether a spot is the same spot may be excessive if only a single visual comparison is needed. We describe a distributed gel comparison program (URL: http://www-lmmb.ncifcrf.gov/flicker) which runs on any World Wide Web (WWW) connected computer and is invoked from a Java-capable web browser. One gel image is read from any Internet 2-D gel database (e.g. SWISS-2DPAGE) and the other may reside on the investigator's computer. Images may be more easily compared by first applying spatial warping or other transforms interactively on the user's computer. First, regions of interest are "landmarked" with several corresponding points in each gel image, then one gel image is warped to the geometry of the other. As the two gels are rapidly alternated, or flickered, in the same window, the user can slide one gel past the other to visually align corresponding spots by matching local morphology. This flicker-comparison technique may be applied to analyzing other types of one-dimensional and 2-D biomedical images.

The 2DWG meta-database of two-dimensional electrophoretic gel images on the Internet.

The 2DWG meta-database is a searchable database of two-dimensional (2-D) electrophoretic gel images found on the Internet. A meta-database contains information about locating data in other databases - but not that data itself. This database was constructed because of a need for an enriched set of World Wide Web (WWW) locations (URLs) of 2-D gel images on the Internet. These gel images are used in conjunction with the National Cancer Institute (NCI) Flicker Server to manipulate and visually compare 2-D gel images across the Internet. User's gels may also be compared with those in the database. The 2DWG is organized as a spreadsheet table with each gel image being represented by a row sorted by tissue type. Data for each gel includes tissue type, species, cell-line, image URL, database URL, gel protocol, organization URL, image properties, map URL if it exists, etc. The 2DWG may be searched to find relevant subsets of gels. Searching is done using the dbEngine - a WWW database search engine which accesses selected rows of gels from the full 2DWG table. The 2DWG meta-database is accessible on the WWW at http://www-lecb.ncifcrf.gov/2dwgDB/ and the NCI Flicker server at http://www-lecb.ncifcrf.gov/flicker/

Automatic detection of noisy spots in two-dimensional Southern blots. International Electrophoresis Society Meeting, Washington DC, March 16-19, 1991.

One of the problems of automatically quantitating 2D DNA gels, is clearly detecting spots visualized by Southern hybridization blots using DNA probes (Rogan et al. in this conference). Spots appear noisy due to multiple transfer steps. If one applies standard 2D PAGE protein gel spot segmentation methods, spots fragment due to highly textured image noise and a weak radioautograph signal and thus are poorly detected. We have observed that these spots are all of a minimum size. Therefore an image processing filter which both takes minimum spot size into account and has immunity to image texture-type noise should be able to reliably detect this class of spots. The 'Busse' Laplacian filter used in the GELLAB-II system, is a modification of the standard (1-21) digital approximation of the Laplacian. In the Busse Laplacian, the sampling interval is n pixels (n greater than 1) instead of 1. In addition, 3 x 3 averaged 'super pixel' values are used instead of single pixels for each element of the Laplacian convolution. This gives the needed noise immunity by filtering out the high spatial frequency image noise while preserving the low spatial frequency character of the spots. We have used this filter successfully on 2D DNA Southern blot image data.

Data-base techniques for multiple two-dimensional polyacrylamide gel electrophoresis analyses.

Two-dimensional protein electrophoresis can benefit from a powerful set of computer-supported image processing and data structure/management procedures. Detection of quantitative differences is complicated by local inhomogeneities in the polyacrylamide base; biochemical changes and variations in temperature and preparative technique also make the between-gel density and x-y coordinate correspondences quite imprecise. The program presented here provides local alignment and computer-controlled variable "flicker" rates for multiple gels, with use of an interactive display system. Manual spot densitometry, referred to a National Bureau of Standards density wedge, can be complemented by a set of automatic densitometry routines for previously established lists of spots. The ability to establish a set of local landmarks, either by included standards or user identification, provides a basis for automatic n-way gel comparison for subsets or for the entire set of spots. Automatic segmentatin algorithms allow isolatin of spots and separation of touching and partially overlapping regions. Various analytical and statistical facilities are part of the user's access to the interactively developed data base. The data-structure and image-manipulation techniques developed here allow for user-directed and heuristic comparisons with online presentation of intermediate and final results.

Database and search techniques for two-dimensional gel protein data: a comparison of paradigms for exploratory data analysis and prospects for biological modeling.

Two-dimensional (2-D) polyacrylamide gel electrophoresis can detect thousands of polypeptides, separating them by apparent molecular weight (Mr) and isoelectric point (pI). Thus it provides a more realistic and global view of cellular genetic expression than any other technique. This technique has been useful for finding sets of key proteins of biological significance. However, a typical experiment with more than a few gels often results in an unwiedly data management problem. In this paper, the GELLAB-II system is discussed with respect to how data reduction and exploratory data analysis can be aided by computer data management and statistical search techniques. By encoding the gel patterns in a "three-dimensional" (3-D) database, an exploratory data analysis can be carried out in an environment that might be called a "spread sheet for 2-D gel protein data". From such databases, complex parametric network models of protein expression during events such as differentiation might be constructed. For this, 2-D gel databases must be able to include data from other domains external to the gel itself. Because of the increasing complexity of such databases, new tools are required to help manage this complexity. Two such tools, object-oriented databases and expert-system rule-based analysis, are discussed in this context. Comparisons are made between GELLAB and other 2-D gel database analysis systems to illustrate some of the analysis paradigms common to these systems and where this technology may be heading.

An efficient disk based data structure for rapid searching of quantitative two-dimensional gel databases.

Fast access of two-dimensional (2-D) gel quantitative databases is important for rapid searching for protein differences between sets of 2-D gels from an experiment. The GELLAB-II system organizes corresponding spots from the gels in the database into reference or "Rspot" sets. These Rspot numeric names index fixed regions in the paged composite gel database file. This is adequate for an existing database, but has several problems. (i) Building the initial database requires guessing how much disk space to pre-allocate for each corresponding spot (i.e. spots from different gels). If it ever runs out of pre-allocated space during this process, it must expand the size of each corresponding set of spots copying the old database data into the new in-place on the disk. (ii) When adding new gels or editing the database, if a new spot is created, the system may also go into this expansion mode. The time spent and wasted disk space can be appreciable--depending on the size of the database (order of 100 gel database). (iii) Because each set of corresponding spots is the same size, we waste space in most spot sets since they do not require the additional space a few spot sets require which contain additional fragmented spots. We present a new low-level disk object-based structure and algorithm, paged indexed buckets (PIB), which optimizes disk space usage while having similar retrieval speed to the original method.

A fast spot segmentation algorithm for two-dimensional gel electrophoresis analysis.

An important issue in the automation of two-dimensional gel electrophoresis image analysis is the detection and quantification of protein spots. A spot segmentation algorithm must detect, define the extent of, and measure the integrated density of spots under a wide variety of actual gel image conditions. Besides these functions, the algorithm must be memory efficient to be able to process very large gel images and do this in a reasonable amount of computation time on low-cost computers, such as workstations and personal computers. We have developed a fast spot segmentation algorithm, extending the GELLAB-II segmenter, which extracts spots in a single raster scanning pass through the gel image. The performance analysis of the algorithm will be given in the paper as well as a discussion of the algorithm.

Some extensions to the GELLAB two-dimensional electrophoretic gel analysis system.

Several important extensions to the GELLAB 2D electrophoresis gel analysis system are discussed, including concepts that lead to the generation of consistent gel maps for a particular cell line, as illustrated by some recent results for the P388D1 cell system, GELLAB is an interactive graphics computer system that facilitates (a) image accessioning, (b) spot-data extraction from the image with position and density quantitation, (c) pairing of corresponding spots between gels, and (d) spot analysis for a large number of corresponding spots (over 3000 per gel) for up to 128 gels. Multiple gels are accessioned into the system used to create a composite gel data base. These then may be partitioned into multiple classes of gels and searched for statistically significant interclass spot differences, which can be visualized in reference map images of selected spot sets as well as mosaic images of particular spots in all gels. In addition to the visual feedback that is available during the gel analysis, various statistical, numerical, and other displays may be generated in the CGELP interactive program of the GELLAB system.

Splitting merged spots in two-dimensional polyacrylamide gel electrophoresis gel images.

We describe a heuristic computer algorithm using boundary analysis for improving spot finding and spot quantitation of large saturated or near-saturated spots in two-dimensional polyacrylamide electrophoresis gels. This spot quantitation is done using spot segmentation, which consists of spot finding and subsequent quantification steps. Occasionally, clusters of large saturated spots may become merged during spot finding. To correct this, the merged spots must be cut apart before quantitation. It is generally obvious from viewing the merged spot's border where they should be cut--at opposing saddlepoints (concavities in the boundary). The algorithm uses an analysis of the missegmented spot's boundary when a saturated spot is detected. If a near-saturated spot is larger than a given size, the spot segmenter program attempts to merge saturated fragments. When merging occurs, the segmenter program analyses the boundary to see if the spot should be split. The new algorithm first finds all robust concavities and then tries to match complementary ones. These paired concavities are then used to guide cutting of the missegmented spot into two or more separate spot regions. Finally, control is returned to the segmenter program to reprocess the data as a set of smaller separated spots.

A World Wide Web (WWW) server database engine for an organelle database, MitoDat.

We describe a simple database search engine "dbEngine" which may be used to quickly create a searchable database on a World Wide Web (WWW) server. Data may be prepared from spreadsheet programs (such as Excel, etc.) or from tables exported from relationship database systems. This Common Gateway Interface (CGI-BIN) program is used with a WWW server such as available commercially, or from National Center for Supercomputer Algorithms (NCSA) or CERN. Its capabilities include: (i) searching records by combinations of terms connected with ANDs or ORs; (ii) returning search results as hypertext links to other WWW database servers; (iii) mapping lists of literature reference identifiers to the full references; (iv) creating bidirectional hypertext links between pictures and the database. DbEngine has been used to support the MitoDat database (Mendelian and non-Mendelian inheritance associated with the Mitochondrion) on the WWW.

Identification of coordinate pairs of polypeptides: a technique for screening of putative precursor product pairs in 2D gels.

Two-dimensional electrophoresis, combined with computerized quantification of individual proteins, provides a sensitive and accurate approach to detection of small differences in the protein pattern. As an extension of "GELLAB," a data base system for the analysis of two-dimensional electrophoretic gels, we report here on a screening algorithm for detection of certain post-translational modification events. It is based on the assumption that, if there is a post-translational modification of a protein in the transition between two functional states of a biological system, in many cases the sum of the protein concentration of a precursor-product pair in one is equivalent to that in the other. In addition to the identification of candidate pairs with a precursor-product relationship, cues to the nature of the post-translational modification that might relate them might also be derived from comparisons of the isoelectric point and apparent molecular mass. This subset of the GELLAB system may be thought of as a screening tool in the search for and investigation of post-translational protein-processing events so that the more time-consuming structural investigations to follow can be performed on a smaller set of proteins.

ECGF and heparin determine differentiation of cloned cerebral endothelial cells in vitro.

Protein expression patterns of morphologically different cloned capillary endothelial cells from porcine and murine brain cortices were examined. Type I cells, grown in medium containing heparin and endothelial cell growth factor (ECGF), exhibited a polygonal, cobblestone appearance and appeared to replicate the cells of the blood-brain barrier endothelium. Type II cells, grown in medium without heparin and ECGF, were elongated and appeared to replicate capillaries in central nervous system tissue. Cells of both phenotypes stained positive by the specific endothelial cell marker Bandeiraea simplicifolia lectin. The expression of alpha smooth-muscle actin (mRNA and protein) was taken as a marker for type II cells. By use of 2-D gel images and the GELLAB II system, a data base was created revealing that two proteins (90 kDa, pI 5.1, and 35 kDa, pI 5.7) were exclusively expressed in type I cells. Furthermore, the synergistic action of ECGF and heparin in respect to the phenotypic determination of cerebral endothelial cells was demonstrated.

Coordinate regulation of the expression of axonal proteins by the axonal microenvironment.

The axonal functions that act in the formation of the neuronal network have been shown to occur in close interdependence with the tissue that surrounds the growing axons. However, little is known about the molecular building blocks underlying axonal functions, although more than 400 axonal proteins have been identified. In view of the existence of such a large number of axonal proteins, we have initiated a project to determine the molecules involved in the implementation of particular axonal functions by a selective approach. On the assumption that plasticity in the expression of axonal functions in response to specific features of the local axonal environment may be based on changes in the expression of particular axonal proteins, the axonal proteins of dorsal root ganglion (DRG) neurons were screened for those whose expression responds to environmental influences. DRG neurons were grown in a compartmental cell system that offers separate access to neuronal somas and to their axons and the axons were locally exposed to different populations of cells from the peripheral or central nervous system. The axonal proteins were metabolically labeled and subjected to two-dimensional gel electrophoresis. Computerized quantitation of the individual axonal proteins revealed that the cocultured cells modulate the synthesis of a few axonal proteins of DRG neurons differentially. The data on the abundance of the newly expressed proteins under varying local environmental conditions were condensed as expression profiles. Comparison of expression profiles and cluster analysis of quantitative gel analysis data revealed that the environmentally modulated proteins subdivide into clusters with common distinct expression profiles under the influence of nonneuronal cells from the peripheral nervous system, nonneuronal cells of the central nervous system, and spinal cord cells, which are composed of neurons and nonneuronal cells. By means of this new, characteristic attribute assigned to environmentally modulated axonal proteins, working hypotheses were made as to their functional role.

Differential modulation of the expression of axonal proteins by non-neuronal cells of the peripheral and central nervous system.

Axonal behavior during the formation of the neuronal network of the nervous system has been shown to be under environmental control. Hence, as a first step in a project aiming to elucidate the molecular basis of axonal functions, we have identified axonal proteins whose synthesis is subject to environmentally induced changes. Neurons from chicken embryonic dorsal root ganglia (DRG) were grown in a compartmental cell culture system that allows selective examination of axonal proteins. Non-neuronal cells of the peripheral or central nervous system were co-cultured with the DRG axons. The axonal proteins expressed under these different environmental conditions were examined by metabolic labeling and two-dimensional SDS-polyacrylamide gel electrophoresis. Computerized quantification revealed that 12 out of 400 axonal proteins responded to changes in the local axonal environment by a change in their relative abundance. Some proteins changed in response to both types of co-cultures whereas some changed specifically under the influence of either peripheral or central non-neuronal cells.