INTERNET - WHERE IS CHEMISTRY GOING? 658
CHIMIA 52 (1998) Nr. I I (November)
Chimia 52 (1998) 658-663 scribed by name (benzonitrile, phenyl cy- © Neue Schweizerische Chemische Gesellschaft anide,cyanobenzene, benzoic-acid nitrile), ISSN 0009-4293 formula (C6HsCN, C7HsN), or any of sev- eral ID codes (100-47-0, DI2450000), as well as by the structural formula itself in The ChemFinder WebServer: any orientation and Kekule form. A search engine query for 'benzoni- Indexing Chemical Data on the trile' would locate any mention of that specific word, but would find no informa- Internet tion about 'cyanobenzene' unless both words were present on the same page. Finding information about this compound Jonathan S. Brecher* would require many searches - ideally, searches for each of the terms listed above, plus many more. Such queries depend on Abstract. The Internet offers a wealth of data on all topics, but chemical information some alphanumeric descriptor being avail- provides unique challenges for searching. General-purpose indexing techniques are able for the substance in question; if only summarized, with a discussion of why those techniques are unsuited for use with a graphical structure is available, a name chemical data. The ChemFinder WebServer, a WWW-basedchemical index, addresses must be derived before any searching can some of the weaknesses in nonchemical search engines. begin. The broad categorization that is the goal of directories makes them also un- suitable for this type of search for informa- Introduction These programs start at some page, which tion about a specific substance. they read and parse into an internal The recent growth of the Internet has database. Having parsed the starting page, transformed what was once a small net- the agent identifies all other pages refer- work devoted to scientific research into a enced in the first one, and repeats the ubiquitous part of modem society. Re- process with each of them in turn. Search cently estimated [1] as containing 320 engines excel atcreating word- and phrase- million public World-Wide Web pages, searchable indexes. As such, they are its size makes the Internet an uniquely often the best choice when looking for valuable information resource. This sheer information that can be summarized by size and decentralized nature, however, a single phrase that would not likely Fig. 1. Benz.onitrile also makes it difficult to find any particu- appear in contexts other than the one lar piece of information on demand. To- desired. Popular search engines include Methodology of the ChemFinder wards this end, public directories and AltaVista (http://altavista.digital.com), WebServer search engines have been created that al- HotBot (http://www.hotbot.com). Excite low the location of information by topics, (http://www.excite.com). and Lycos Begun in 1995, the ChemFinder Web- keywords, and related identifiers. We de- (http://www.lycos.com) . Server has been designed to address the scribe here the CS ChemFinder WebServ- Directories differ from search engines deficiencies in general-purpose informa- er (http://chemfinder.camsoft.com),an in that they do not attempt to index every tion indexes by providing a centralized Internet index specifically tailored to store word on every page, but rather try to means of searching for chemical informa- and present information about individual classify pages into broad (and usually hi- tion on the Internet. A comprehensive chemical substances. erarchical) categories. Creating directo- database has been dynamically assembled ries tends to be very labor-intensive, with from numerous public sources and is con- the final decision on where and whether to tinually being augmented. As new chem- Architecture of Search Engines and reference any given page usually given to ical information is located on the Internet, Directories trained individuals. Because of this hu- it is analyzed and merged into this master man involvement, directories are always ChemFinder WebServer database. Internet information indexes may be smaller than search engines in absolute Only core information about each sub- broadly categorized as search engines or size, but are expected to be denser in terms stance is retained by the server: the name as directories, which take different ap- of quality information content. The inher- of the substance and the address of the proaches to the same goal of making it ent categorization of directories makes page that contains information about it. easier to find information. Search engines them the indexes of choice when looking When available, select additional com- function via automated traversal of the for general information about a topic. mon data are retained, including Chemical WWWusing software agents known var- Yahoo! (http://www.yahoo.com) is the Abstracts Service Registry Numbers and iously as spiders, (ro)bots, or worms [2]. prototypical example of a directory, and is basic physical property information such the most visited site on the WWW [3]. as boiling points and melting points. Struc- Locating chemical information pro- tural information is rarely available on vides challenges not encountered with oth- other sites. Even when available, it is *Correspondence: J.S. Brecher er types of information. First among these generally provided only as a static image CambridgeSoft Corporation 100 Cambridge Park Drive is the variety of ways in which a single in .gif or .jpg format, and not in any format Cambridge, MA 02140, USA substance may be described. A given com- that preserves the chemical information in E-Mail: [email protected] pound (Fig. 1) may reasonably be de- a way that can be interpreted automatical- INTERNET - WHERE IS CHEMISTRY GOING? 659 CHI MIA 52 (1998) Nr. II (November) ly. Accordingly, structural information is the completed database is made available small number of sites [10] provides exten- not retained, but used as a reference when for searching. sive physical property data and full Mate- structures are redrawn. The ChemFinder WebServer is a de- rials Safety Data Sheets (MSDSs) for all During the indexing process, several scriptive index, not a peer-reviewed data products. More typically, chemical manu- methods are used to improve the quality of source; many errors remain even after the facturers will provide limited physical the data. Principal among these is normal- verification process. These are addressed property or usage information for some of ization of the chemical names. During on a case-by-case basis as they are found, their products [11]. Many manufacturers normalization, all spaces, parentheses, and often by users of the ChemFinder Web- provide only a textual list of their sub- other punctuation are removed, and the Server service. In some cases, the 'error' stances [12], or simply a list of classes of names are converted entirely to lower case. is encountered more frequently in com- substances [13], neither of which would Regional spelling variations (aluminum mon use than is the correct data. For exam- qualify for indexing in the ChemFinder vs. aluminium, sulfur vs . sulphur) are rec- ple, the name 'alanine' is chemically un- WebServer for reasons discussed earlier. ognized and standardized. Names written specific about the stereochemistry at C(2), Governmental and other regulatory data in CAS-style inverted form are un-invert- and could refer to the D-form, the L-form, are a third category of information that is ed (acetic acid, sodium salt vs. sodium or any mixture of the two. In practice, the well represented on the Internet. Broad acetate), and several dozen of the most term almost always refers to the natural and unrestricted distribution of regulatory common typographical errors are auto- form. Where common usage contradicts information is often mandated by the or- matically corrected. The normalized name official nomenclature rules, the error is ganization that compiles it, and the Inter- is stored along with the original form. In not corrected. The ChemFinder WebServer net provides an ideal way to distribute addition to name normalization, CAS Reg- lists 'alanine' and 'L-alanine' as synony- information to a wide audience. Regulato- istry Numbers, if present, are checked for mous, and does not list 'alanine' as a ry information typically includes little more validity using their internal checksum [4]. synonym for 'D-alanine'. than a compound name and (sometimes) The new data are compared to the data A site qualifies for inclusion only if it CAS Registry Number, but the fact that a already present in the ChemFinder Web- references individual chemical substanc- substance is regulated is often an impor- Server through comparisons of the nor- es. In its current design, the ChemFinder tant datum in its own right. Given the malized names and the CAS Registry Num- WebServer does not index information bureaucratic division between governmen- bers. Any new substances that fail to match about chemical conferences, chemical tal departments, specific pieces of regula- existing records are examined individual- education, chemical software, or any of a tory data can be difficult to locate, but are ly for typographical errors. Approximate- host of other chemistry-related topics. available from many branches of the US ly 5% of all data indexed by the Chem- Additionally, a site must provide at least federal and state governments [14], from Finder WebServer is recognized to be one piece of information about the sub- the governments of many other countries erroneous by this stage [5]. Where the stance. A simple textual list of compound [15], and from international organizations intended data are obvious (e.g., when the names - with no other information - would [16]. errors are typographical in nature), the not qualify for indexing in the ChemPind- The molecular biology community has errors are corrected; otherwise the datum er WebServer, since purely textual data is long had an active presence on the Inter- is discarded. already well suited for indexing in exist- net, with data on substances large (The During this premerge process, the data ing text-based search engines. Human Genome Project [17]), and small are also analyzed for transitive logic er- (The Pherolist: List of Sex Pheromones of rors. For example, one site might refer- Lepidoptera and Related Attractants [18]), ence 'Ritalin' and 'methylphenidate' as Chemical Information Sources on the and in between (The Protein Data Bank synonyms. Another might state that 'me- Internet [19]). Because of the large size of many of thylphenidate hydrochloride' is the gener- these substances, and because the molec- ic name for Ritalin. From this, it could be As an index, the ChemFinder Web- ular biology community has already de- concluded that 'methylphenidate' and Server would have little purpose if there veloped excellent databases such as listed 'methylphenidate hydrochloride' are in- were no data available to be indexed. above for handling these data, we have terchangeable names that refer to the same Chemical data is fairly abundant on the made a conscious decision not to index compound. This is clearly an incorrect Internet, but can be classified into one of biological macromolecules in the Chem- conclusion, and other reference sources only a small number of forms. Finder WebServer. Exceptions are made must be consulted to resolve the discrep- Health and safety data comprise one of on a case-by-case basis, and generally ancy. the largest classes of chemical data en- only for 'small' biological molecules such The remaining entries are then merged countered. Data of this type are typically as those listed in The Pherolist, and for into the main database. During the merge oriented towards consumers or health-care 'popular' biological substances such as process, any physical property informa- professionals, rather than laboratory chem- insulin and hemoglobin. tion, if present, is compared to values ists. These data are found at sites focusing In recent years, there has been an in- already in the database. Numerical differ- on the environmental impact of chemicals creasing availability of commercial sourc- ences of one unit (generally, degree Cel- [6] and sites discussing specific diseases es of chemical information on the WWW. sius) or less are assumed to be insignifi- [7]. Extensive information is available Many journals are now available in elec- cant and are discarded. Differences of about individual medications [8], particu- tronic format, and the Chemical Abstracts greater than one unit are examined on an larly those involved in treating high-pro- Service provides Internet-based access to individual basis, and other reference works file diseases [9]. their data as well. With few exceptions, are consulted to identify the correct val- Most chemical manufacturers and dis- however, this information is available only ues. Chemical structures are added, with tributors have a presence on the Internet, by subscription. We have decided to limit molecular weights and formulas calculat- usually on the WWW, but the information the ChemFinder WebServer to indexing ed directly from the structures. Finally, content of these sites is highly variable. A unrestricted data sources only. INTERNET - WHERE IS CHEMISTRY GOING? 660
CHIMIA 52 (1998) Nr. II (November)
Although most chemical information crystallographic data, the Protein Data program, the CACTVS System Chemis- on the Internet can be considered as one of Bank is a particularly valuable resource, try Spider, traversed aportion ofthe WWW the types discussed, there is still much although, as a collection ofbiological mac- looking for chemical structures stored in information that cannot be neatly catego- romolecules, it has not been indexed in the any of several chemical formats. Over rized. Often these sites will be the work of ChemFinder WebServer. 2250 such files were identified, with some one individual who has spent the time to Klotho, the Biochemical Compounds compounds represented by more than one document some substances of personal Declarative Database, was one of the first file. A search of this type is necessarily interest. University courses occasionally small-molecule data sources on theWWW, hindered by the overall scarcity of chem- provide information about molecules dis- becoming available in February, 1994 [21]. ical structures stored in chemical file for- cussed in class. Information about a chem- Klotho's collection of biological mole- mats publicly available on the Internet. ical substance might appear in a context cules demonstrated the potential of the The entries in the WWW Chemical Struc- not specifically directed to chemists. All WWW in presenting and distributing tures Database have also been indexed in of these would qualify for indexing in the chemical information, and was a source of the ChemFinder WebServer, with refer- ChemFinder WebServer. inspiration in creating the ChemFinder ences to their original source. Additionally, four sites deserve spe- WebServer. The entries in Klotho have Finally, special mention must be made cial mention for their comprehensiveness been indexed in the ChemFinder Web- of the NlST Chemistry WebBook [23], or innovation. The Protein Data Bank Server. which since August, 1996, has provided mentioned earlier was one of the first The WWW Chemical Structures Da- one of the few large (over 30000 records public sources of chemical information on tabase was the first (and, apparently, only) currently) databases of substances. It is the Internet, and has been in operation for attempt to utilize the techniques of a gen- also one of the few databases to provide over 20 years [20]. As a doubly rare source eral purpose search engine in creating a complete references for eachphysical prop- of both peer-reviewed information and chemistry-specific database [22].Its search erty value listed. The entries in the NIST Chemistry WebBook have also been in- dexed in the ChemFinder WebServer.
Use of the ChemFinder WebServer
The ChemFinder WebServer is pub- licly available via two similar form-based Welcome to the ChemFinder Web Server! WWW interfaces. Typical of other gener- al-purpose search engines, the default Etter • chm1i
Table 1. Effects of Normalization. All names d compounds who t... Oeave fields blank as necessary) shown will normalize to the same string, and will ~I return the same records when used as a query. J-chlorobcnzoic acid. _odium 'all tructure IS 3-chloro-bcnlOi a ill. ~odlUm ,odium 3-(chloro)-benzo3Ie ill-chi ro bcnll at . sodillm :alt m-chlortlenz at'. sodium
d Melting pt IS
Table 2. Disallowed Field Combinations for Searching
Molecular formula \\ith mole lIlar weight
act name \\ ith an} Olh r field
j\ • ReglW}' umber \\-ith any other Ii'ld lructurc with mol cular formulu or m Iccular:' Fig. 3. The advanced query interface of the ChemFinder WebSen>er, with diagram-based structure wdght searching INTERNET - WHERE IS CHEMISTRY GOING? 661 CHIMIA 52 (1998) Nr. II (November) executed by users of the ChemFinder Web- Server, and because a single text phrase can be identified as one of the four input C I types mentioned above with a high degree of certainty. Numerical values that con- form to the CAS Registry Number check- Fmd compo1Ulds whose- •. Oeave fields blank as necessary) sum algorithm are assumed to be CAS Registry Numbers. Other numerical val- Name lis .::.1 ues are identified as molecular weights. Alphanumeric phrases that can be parsed as molecular formulas (ignoring case) are and Formula is I treated as formulas, while all other queries are searched as names. Since a chemical and CAS RNls and :MW1S I formula could also be interpreted as a Melting Boilin and IS I and g IS chemical name, a formula search that fails pt pt to match any entries in the database is and StnJcture IS the SMILES stnng automatically re-queried as a search by chemical name. This simple interface is popular, accounting for 72% of all search- Clear the 101m I es. As with general-purpose search en- gines, substances can be searched by full Fig. 4. The historic query interface of the ChemFinder WebServer, with SMILES-based structure or partial chemical name; however, all searching name searches are normalized according to the same rules used in the database merging process (Table 1). Since the Poor Query ChemFinder WebServer has a very di- verse user base - both geographically and If you specify a Molecular Formula, it is never useful to specify a Molecular Weight. At best, in educational level - normalization is any Molecular Weight you enter will be redundant with the Molecular Formula. At worst, they very important in addressing common ty- will be contradictory, and prevent you from getting any useful hits. pographical and translation errors. This Press the button below to continue your query, searching for all fields except Molecular Weight, normalization greatly increases the per- or press the Back button on your browser to modify your query. centage of successful searches. An advanced search form (Fig. 3) of- Search for all fields except Molecular Weight fers greater control of the search process, allowing chemical names, formulas, mo- Fig. 5. An example of the message returned when entering redundant search terms lecular weights, and CAS Registry Num- bers to be searched explicitly. Queries are also supported for boiling and melting points. Using this form (or its available historic sibling, Fig. 4), chemically relat- SImpl,' Quv,.y ed data can be searched in chemically • d CS emFIn er •A~vtln«'uQv,-,'y meaningful ways. Molecular weights and Ch With Plug-In numerical physical properties can be Scorching and Information Integration Old Quc'ry Form searched by range as well as by exact values, with automatic recognition of sig- Search produced 6 results: nificant figures. Formulas can be searched by element ranges as well as by exact Acetonitrile-2-13C formula with any ordering of elements. Acetonitrile-d3 Chemical structures can be drawn directly Acetonitrile, diisobutvlamino-. in the browser window when using the CS Acetonitrile, (methyhmino)di- ChemDraw Netscape Plug-In (or entered Acetonitrile in the textual SMILES [24] format in the Acetonitrile, (phenvlsulfonyl)- older form). Searches such as these are commonplace when looking for chemical Fig. 6. Results of a search that matches more than one substance information, but rarely or never used oth- erwise, and are not supported in general- purpose search engines and directories. due to terms that were inherently contra- benzene and many other substances with a The advanced search forms allow con- dictory. A search of this type might have molecular weight of 76. The server has current searching over multiple fields. This looked for a substance whose name was since been modified to disallow certain feature was intended to permit the creation 'benzene' and whose molecular weight field combinations (Table 2). If a search is of more specific searches, but an analysis was 76. Since the molecular weight of attempted with one of these combinations, of failed searches over several months benzene truly is 78, this search would fail the query is not executed, but the user is revealed that many searches were failing even though the database contained both presented with an explanation of why the INTERNET - WHERE IS CHEMISTRY GOING? 662 CHIMIA 52 (1998) Nr. II (November)
Table 3. Top 20 Most-Common Searches at the ChemFinder WebServer (data for April, 1998). Acetonitrile-dJ [2206-26-0] Note that several of the most-common searches were for classes of compounds and notfor specif- ic compounds themselves; these searches would Synonyms: Deuterated acetonitrile not have returned any data.
(DHQD)2PH L ethanol This picbm is • liw chm1kal imtZ:< K n 4 The ChomDnw P!ultin h ameth len dianllne 1ets you somh by drlwing =, in your web browser. benz n Have you dawnloadtd ityrt? -l,5-dihydroxy-l.3-ben7cnedlsulfoOl acid phenol
fl.mm:r:l ~I[~"" Specifk Gravity 0.844 uceton' 1:~tr."r.I~ •••~ Vapor Density benzaldehyde benzoic aCId ~ • .I."]lII.+ •• r:.D.~t:l~ Water Solubility meLhan I I~r.~UW;l:m'fl'':'' EPA Code sodium h dro ide ItI·}.i'mm.I~ RTECS ulfuri acid [ ••.•li.liII~~:.lf,. LACHRYMATOR/HYGROSCOPIC. ab lute (hanoi aCid More information about this compound i" available from acetic id
ABCR GmbH&Co KG Acetonitrile-d3, 99.8% o Acetonitrile-d3, isotopic purity, 99% ester ••..•••1 . 1 ••.•••• 1 f".1 1\ •.•.n""' •..••.•• ulfamli acid mon lhydmk
Fig. 7. Detailed information about an individual substance
supplant that available at the other sites Usage of the ChemFinder WebServer, in unique sites/day, averaged listed, but is often sufficient to answer weekly basic identity questions ('What is the name of the following structure?'). 1;'3000 :!i! The ChemFinder WebServer is not a ~ 2500 single entity, but is composed of a suite of ': 2000 server-side applications. Queries are trans- :I .2" 1500 c: mitted from a WWWbrowserto the Web- ~ 1000 Site WWW server software, published by Cl l!! 500 O'Reilly and Associates. WebSite parses Ql > c( the initial query from HTTPformatinto its CD CD ,.. ,.. ,.. ,.. ,.. ,.. ,.. ,.. ,.. ,.. ,.. ,.. co co en en S; III ~ ~ ~ ~ ~ ~ ~ ~ '",.. '~ ...•..•co !a distinct fields, and transfers it via the Com- N ~ ~ i2 ,.. ~ j::: Oil ~ N ,.. ~ ~ c;;,.. ,..~ C>l,.. ~ ~ (;; ~ ill co'" (j; ~,.. ~,.. C>l,.. ';::: ~ (;; ~ mon Gateway Interface (CGI) [25] to a Week Ending Visual Basic application. The COl con- firms the validity of the query as discussed Fig. 8. ChemFinder WebServer daily usage. Several strong seasonal trends are readily apparent, above, then transmits the data to CS Chem- including a slight decrease in traffic during the (northern hemisphere) summer, followed by a sharp Finder Pro (CambridgeSoft Corporation) increase in traffic in September and October. The year-end holidays are particularly obvious, while via OLE Automation. CS ChemFinder is a the lesser decrease at the end of November corresponds to the American Thanksgiving holiday. chemical database designed specifically to search and store chemical data, and functions as the core of the ChemFinder search terms should not be used in combi- for chemical information on the Internet, WebServer. It performs the actual query, nation (Fig. 5). This has resulted in a the ChemFinder WebServer provides a and retrieves the textual, numeric, and general increase in search quality. list of links at the bottom of this page, structural data, which it returns to the CGI. Queries resulting in more than one indicating other locations that provide The CGI then formats the data and returns matching substance produce a list of names additional information about the substance it to WebSite, which finally returns it to of those substances (Fig. 6), with links to shown. Substance-specific information the user. detailed information about each one. Que- (usually limited to substance name and Usage of the ChemFinder WebServer ries that match a single substance display synonyms, structure, formula, and molec- is currently showing an approximately the detailed information page (Fig. 7) di- ular weight) is shown at the top of the 200% yearly growth rate, with strong sea- rectly. Fulfilling its role as a search engine page. This information is not intended to sonal variations (Fig. 8). As of May , 1998, INTERNET - WHERE IS CHEMISTRY GOING? 663
CHIMIA 52 (1998) Nr. II (November)
over 75000 individual substances were [8] See the Index to Allergy/Asthma Medica- indexed from some 300 sites. Each sub- tions at http://www.cmh.edu/Allergy/Drugs stances had an average of2.4links to other land the Australian Guide to Medications at sites on the Internet containing additional http://www.avm.com.au/agtm/agtm.html. information about the substance. Although for example. we have been unable to quantify the ways [9] See the AIDS Drug Assistance Program at that most people are using the ChemFind- http://www .aids infonyc.org/network/ ac- er WebServer, an examination of the most ces/index.html, for example. [10] See the Fisher Scientific site athttp://www. popular searches suggests that much of its fisherl.com, for example. use isin locating general information about [II] See the Callery Chemical Company at http:/ common substances. /www.callery.com (physical properties), Spectrum Laboratories at http://www.spec- lab.com (usage, analytical methods and Future Prospects environmental fate), and Penn Bioorganics at http://www.pennbio.com (structures, The ChemFinder WebServer repre- melting and boiling points) as representa- sents a new approach to providing chem- tive examples. ical data, directed not at the trained infor- [12] See The Coral Group site at http://www. mation specialist but at the lay users that coralchem.com, for example. [13] See the Hoechst site at http://www .hoechst. now represent the majority of traffic on the comandtheDiscovery Chemical, Inc. site Internet. Further developments of the at http://www.inc.comluserslkkeller.html. ChemFinder WebServer will likely result for example. in larger data sets, but will probably not [14] See the EPA Restricted Use Products re- reach the comprehensiveness achieved by port at http://www .epa.gov /docs/RestProd/ the Chemical Abstracts Service Registry and the Alaskan Administrative Code at File (17 million entries) or the Rei/stein http://www.touchngo.com/lglcntr/akstats/ database (6 million entries). AAC.htm, for example. Additional effort will also likely be [15] See a list of Australian approved food addi- placed on further enhancements to the tiyes at http://www.hawkesbury.uws.ed.au/ user interface, search engine, and normal- -skurrayg/code.txt, for example. [16] See the Helsinki Convention of 1992 at ization routines. All work will be directed http://www.helcom.fi. for example. to meeting what is the goal of all search [17] http://www.nhgri.nih.gov/HGP/ engines and directories: the rapid, easy, [18] http://nysaes.comell.edu/pheronetl and accurate location of information. [19] http://www.pdb.bnl.gov [20] F.c. Bernstein, T.F. Koetzle, GJ.B. Willi- ams, E.F. Meyer, Jr., M.D. Brice, J.R. Rod- gers, O. Kennard, T. Shimanouchi,M. Tasu- Received: September 11,1998 mi, J. Mol. Bioi. 1977,112,535. [21] 'What's New on Klotho' (see http://www. ibc. wustl.edu/moirai/klotho/whats_new. [I] S. Lawrence, c.L. Giles, Science 1998, html). 280,98. [22] 'The WWW Chemical Structures Data- [2] M. Kloster, 'The Web Robots Pages' (http: base' (seehttp://www2.ccc.uni-erlangen.de //i nfo. webcrawler.com/mak/projects/ro- services/web mol. htm I). bots/robots.html). [23] W.G. Mallard, P.J. Lindstrom, Eds., 'NIST [3] RelevantKnowledge, Inc., 'RelevantKnowl- Chemistry WebBook, NIST Standard Ref- edge Releases March's Top 25 Web Prop- erence Database Number 69' ,March 1998, erties and Domains by Unique Visitors' National Institute of Standards and Tech- (http://www.relevantknowledge.comlPress nology, Gaithersburg MD, 20899 (http:// /release/04 _08_98.html). webbook.nist.gov). [4] Chemical Abstracts Service, 'Check Digit Verification of CAS Registry Numbers' [24] D. Weininger, J. Chern. In! Comput. Sci. 1998,28,31. (http://www.cas.org/EO/checkdig.html) . [25] National Center for Supercomputing Ap- [5] J.S. Brecher, 'Chemical errors found on plications, 'The Common Gateway Inter- WWW sites' (http://chemfinder.camsoft. face' (http://hoohoo.ncsa.uiuc.edu/cgi/) com/errorsfound.shtm) . and R.B. Denny, 'Windows COl 1.3a [6] See the Internet HazDat Site Contaminant Interface' (http://www.dc3.com/wsdocs/ Query at http://atsdrl.atsdr. cdc.gov:8080/ 32demo/windows-cgi.html). gsql/sitecontam.script and The Coalition Opposed to PCB Ash in Monroe County, Indiana, at http://copa.org, for example. [7] See the Berkeley Carcinogenic Potency Project at http://potency.berkeley.edu/ cpdb.html and the Contact Derrnatitis Home Page at http://www.mc.vanderbilt.edu/ vumcdeptlderrnlcontactlindex. html, for ex - ample.