The NIH-EPA Structure and Nomenclature Search System

G. W. A. MILNE*

National Institutes of Health, Bethesda, Maryland 20014

S. R. HELLER

Environmental Protection Agency, Washington, D.C. 20460

A. E. FEIN, E. F. FREES, R. G. MARQUART, J. A. MACGILL, J. A. MILLER, and D. S. SPIERS

Fein-Marquart Associates, Towson, Maryland 21212

Received July 12, 1978

The NIH-EPA Structure and Nomenclature Search Systems (SANSS) contains 110776 different compounds from 40 separate files. The structure records have been merged into a single unified file as have the nomenclature records. It is possible, using the interactive programs of the Structure and Nomenclature Search System, to retrieve all references to a particular chemical name, structure, or substructure within these 40 files of chemicals. The NIH-EPA Chemical Information System, of which the Structure and Nomenclature Search System is a central part, is available to the international scientific community via a timeshared, network computer.

INTRODUCTION

In an earlier paper in this series,¹ we described the design of the first publicly accessible version of the family of interactive programs that comprised the NIH-EPA Substructure Search System, a component of the NIH-EPA Chemical Information System.² Since then, some fundamental changes in the design of the Substructure Search System have been accomplished, and it is the purpose of this paper to describe these in detail.

Based upon programs developed originally by Feldmann,³ the Substructure Search System (SSS) permitted the searching through a file of chemical substances for a specific structure or substructure. To be made searchable by this system, a file, or list of chemical substances, was first sent to the Chemical Abstracts Service (CAS) where every compound was assigned a CAS Registry Number,⁴ using methods that have been described in detail elsewhere.⁵ Once the Registry Number was known the nomenclature associated with the substance (index names and synonyms) was extracted from the Registry maintained by CAS, and a copy of the connection table was derived from the same source. The connection table is a two-dimensional matrix which tabulates every nonhydrogen atom in the molecule, listing each of its neighbors and each of the bonds associated with the atom. The SSS basically allowed one to search through the assemble connection tables for a specific type of atom (e.g., an aromatic carbon bearing a bromine or a carbonyl carbon flanked by a fluorine) and locate all compounds that contained such an atom. Related programs within the SSS permitted one to locate all compounds having specific types of rings, molecular weights, or partial or complete molecular formulas. Every search resulted in a temporarily store file of Registry Numbers of those compounds which fulfilled the particular criterion; and, since the files could, upon command, be combined in the Boolean AND, OR, or NOT sense, the resulting system allowed searches for any chemical substance, defined precisely or otherwise. It was quite simple, for example, to locate all the tetracyclic compounds that contain just one aromatic ring substituted by two fluorines in an ortho relationship.

In June 1977, this version of the SSS was made generally available for use on a fee-for-service basis via a commercial networked computer. At that time, it was possible to search through five distinct files of chemicals, including the files of the CIS mass spectral and carbon-13 NMR spectral databases as well as the Inventory Candidate List that had been developed from CIS files for use in connection with the Toxic Substances Control Act (TSCA). Considerable general use has been made of that system, during which it became clear that the TSCA Candidate List was being searched far more that the other files.

It also became clear that a weakness of the system was that it was necessary, prior to the search, to decide which of the files one wished to search, and to search through all five files, five separate searches were necessary. Table I. Type of Search Available in SANSS

Program Effect

IDENT To search for a specific complete structure.

SUBSS To search for a specific substructure.

NPROBE To search for a complete or partial name.

FPROBE To search for specific atom-centered fragments.

RPROBE To search for specific rings or ring systems.

SPROBE To search using the CIDS keys screens.

MW To search for specific molecular weights.

MF To search for specific complete or partial molecular formulas.

RCOUNT To search for compounds having a specific number of rings.

ACCOUNT To search for compounds having a specific number of atoms.

A further problem was that it was relatively difficult to locate common compounds such as cholesterol or camphor, which have well-accepted trivial names but quite complex structures. Further, it was also impossible to search directly for an exact match to one's query structure.

These deficiencies have now been remedied and the resulting system permits searching through many files for substances identified by name, structure, substructure, or chemical "screens," as shown in Table I. The SANSS is therefore operating along precisely the lines defined in the MITRE report⁶ in connection with the Chemical Structure and Nomenclature Search System. A name search has been installed in the SSS, and there is also an IDENTITY search which will search for an exact match for the query structure. As a result, the name of the Substructure Search System has been changed to Structure and Nomenclature Search System (SANSS), which more exactly describes the capabilities of the programs.

A more substantial change is that all the previously independently searchable files, 40 in number, have been merged into a "Unified Database" thus obviating the file selection process described above. Searches are now conducted on the full Unified Database, and, as before, the answers to searches are the Registry Numbers of the retrieved compounds. In addition, however, information as to which of the 40 files contain information concerning these compounds is also available. The user can, prior to a search, specify the files in which there is interest and if this is done, substances that are only found in other files will be ignored. The fact that all searches are now conducted through the large Unified Database, rather than small, selected databases, adds slightly to the computational overhead, but as is discussed below, the increase in search cost is minor and this approach seems to represent the better of the two alternatives. Moreover, it results in a substantial reduction in storage costs. As a result of the merger, the Unified Database now contains 110776 different compounds, representing 40 different files. These changes to the original SSS are described in detail in this paper.

DISCUSSION

1. Unified Database. When a search of any sort is carried out in the SANSS, the system will locate all compounds which fulfilled the criteria specified by the user. What actually is retrieved by the SANSS is a list of the CAS Registry Numbers of these compound. If the user then seeks to examine this subset of compounds, the program will use the Registry Numbers, one-by-one, to look up the names and various other information describing the compounds. Whichever of these are requested are then presented to the user for inspection. In the early version of the system, referred to above, each file of compounds was searched independently of the others and the search resulted only in the Registry Numbers, in turn, could then be used to retrieve the names and structures of the compounds.

During 1977, it was decided to merge together all the available files and for each compound, it became necessary to retain, not only the Registry Number, but also the information as to which of the source collections contained information about that compound. Thus a substance such as benzene, which is in many files of the SANSS, will be stored once in the Unified Database, with the notation that its Registry Number is 71-43-2 and in addition that, together with its Registry Number, it is to be found in certain specific files of the SANSS. Every chemical described in the SANSS therefore has an identifier (the Registry Number) and also a sub identifier (names or numbers of the files that contain it).

With the exception of file 7 (the Merck Index), the entire Unified Database of names, CAS and other synonyms, and connection tables is in the public domain.⁷ The various component files of the Unified Database are given in Table II, which contains the name and number of each file, together with the number of substances it contains. The important advantages of this approach are that the whole SANSS can be searched with a single command and that the data associated with each unique compound need be stored only once. Prior to the merging of the component databases, there were in the SSS, 40 separate files containing a total of 173817 entries. The number of unique compounds in the system, however, was only 110776. Thus 36.35% of the stored data were redundant. Unification of the databases has made it possible to save almost all of this storage, and the resulting decrease in the cost to the end user has been considerable.

Offsetting this positive result, there has been an impact upon the actual searching through the system. The whole database of 110776 substances, rather than a component file of perhaps 10,000 substances, is being searched. Searching is based upon an inverted file concept,⁸ and so the cpu demand of a search is not changed markedly because the file is bigger. The number of retrievals to most searches is now considerably higher, however, and this places a burden upon the intersection and merge routines that are implicit in many of the searching programs. Mainly because of this, the absolute cost of carrying out a search has increased, as a result of the merging of the databases, by between 10 and 50%. If the costs are expressed, however, on a per-compound-searched basis, they can be seen to have decreased by about 50% because the number of compounds that are searched in any operation has increased greatly. The typical commercial cost of a complete structure search or a name search is comfortably under $2.00.

The merging of the databases has had a number of other, less important effects that apparent to users. At the beginning of a session, the user is informed that, unless the program is otherwise notified, all searches will be conducted through the full file. At this point, an opportunity is provided to select a smaller number of search files. In order to take advantage of this, the user must know the identifying numbers of the files in which searches are desired. Consequently, a new HELP file listing the files and their identification numbers has been added to the system. This HELP file identifies each of the 40 files that comprise the Unified Database.

Background information concerning each of the component files has also been added to the system. For example, the user who wishes more information about File 32, the NIOSH Registry of Toxic Effects of Chemical Substances, can type "HELP 32" in response to the "OPTION:" prompt, and a 554-word description of the RTECS file will then be returned to his terminal. This description includes the name, address, and telephone number of an individual at NIOSH who can provide even more detailed information.

When a search has been completed, the command SSHOW will list the compounds that were retrieved. Also, listed, at the user's option, are Registry Numbers, structural and molecular formulas, and alternative names or synonyms. It is also possible to list the component files in which each compound is represented. It is possible, at the user's discretion, to list these files by name or number, or both or neither.

Figure II. Component Files of the Unified Database

# Name Substances

1 EPA - TSCA Inventory Candidate List. 33449

2 NIH-EPA CIS Mass Spectral Data Base 25544

3 NIH-EPA CIS Carbon-13 NMR Spectral Data Base. 6505

4 EPA - Pesticides, Active Ingredients. 1444

5 EPA-OHM-TADS. 858

6 Cambridge X-Ray Crystallography Data Base. 16742

7 Merck Index. 8959

8 EPA - Pesticides, Analytical References Standards. 562

9 EPA - STORET. 234

10 EPA - Chemical Spills. 577

11 EPA - SOTDAT 572

12 NIMH - Psychotropic Drugs. 2214

13 EPA - SAROAD. 65

14 NBS - Gas Phase Proton Affinities Data Base. 514

15 CPSC - CHEMRIc File 890

16 EPA - Pesticides, Inactive Ingredients. 735

17 NBS Heats of Formation of Gaseous Ions Data Base 3156

18 National Fire Prevention Association List of Chemicals. 396

19 FDA/EPA - Pesticides Reference Standards. 613

21 International Trade Commission List of Chemicals. 9140

22 NBS - Single Crystal Data Base. 18286

25 EPA - Effluent Guidelines. 118

26 EPA - Organic Chemical Producers. 375

27 IPC - Chemical Production. 104

28 IPC - Chemical Plant. 103

29 NSF - List of Environmentally Hazardous Chemicals. 225

30 University of Tokyo - EROICA Data Base. 4492

31 PHS-149 List of Carcinogens. 4425

32 NIOSH Registry of Toxic Effects of Chemicals. 19891

33 NIOSH National Occupational Hazard Survey. 4560

35 Environmental Mutagen Information Center Data Base. 4030

36 Environmental Teratogen Information Center Data Base 3250

43 EPA - Selected Organic Air Pollutants 578

45 EPA - Section 112 of Clean Air Act. 5

58 NCTR - Potential Industrial Carcinogens and Mutagens. 91

59 EPA - List of Environmental Carcinogens. 27

66 EPA - List of Hazardous Pesticides. 22

67 EPA - Mutagenicity Studies. 25

70 CIIT - List of Candidates. 26

82 NOAA - Micro constituents of Fish and Fishery Products. 15

Total 173817

A difficulty arises when a compound is present more than once in a component file. The carbon-13 NMR file, for example, often has more than one entry for a given compound, because its spectrum may have been measured in different solvents. This does not happen in many component files, but when it does, a system of local identifiers must be used to permit distinction between the various occurrences of the same compound.

2. Nomenclature Search. The structural searching capabilities of SANSS are very powerful and appear to be the search method of choice for many chemists. A need exists, however, for a program which permits a search for substances by name or synonym, and accordingly, a nomenclature search has now been developed.

Chemical names may be single words, groups or words separated by "natural" delimiters, typically spaces, groups of words separated by special characters such as hyphens or parentheses or (in the case of many TSCA class 2 chemicals) paragraph descriptions of syntheses/processes. Names in the first two of these groups may be searched for relatively easily, and, once a definition of delimiters is established, the third group presents no particular problem.

If the program is told that the name supplied by the user is a complete name, it can search through a set of inverted files for each occurrence of exactly that character string. It is likewise a simple matter to decompose a multi word name into its component words and build and inverted file from these partial names. If the user indicates that the name supplied is a partial name, the search program simple accesses an inverted file of those name fragments for retrieval purposes. While it is unlikely that irrelevant retrieval will result from a search with the full name, such an outcome is possible with a partial name. The hits from all name searches are stored in temporary files which can be inspected by the user, who may select the compounds which were actually sought. In this sense, the name search, NPROBE, operates in just the same way as, for example, the fragment probe search, FPROBE.

An example of the use of NPROBE is shown in Figure 1. Here the user enters the name fragment SASSAFRAS. This, of course, is an utterly trivial name, but it is found as part of a name of the generically registered material, hydrogenated sassafras oil. The retrieval is reported to the user who, inspecting temporary file number 1, finds that the substance does indeed have a Registry Number, is to be found in the TSCA Inventory and the International Trade Commission List, but has no known molecular formula (W99 is used to encode this fact) or structure.

A more informative result is shown in Figure 2, where a full name, DHA, is entered by the user. This, it transpires, is an acronym for dehydroacetic acid, which is in eight of the files, has eight synonyms, including DHA, and has the structure and molecular formula shown.

As has been discussed above, it is possible to limit the retrieved information to just that pertaining to specific files. An example of this approach is shown in Figure 3. Here, the SELECT command was used to restrict retrievals to files 5, 9, 10, and 31. The first three of these are EPA files of pollutant chemicals, and the fourth is the PHS-149 List of Carcinogens. Use in NPROBE of the full name TOLUENE, followed by the SSHOW command, reveals that toluene, which has seven other names, is found in all four of the files. If, as is shown in Figure 4, retrieval is permitted from all files, then toluene is found to be in 19 files. In this example, the SETDC command has been used to limit the display to the names of the files in which toluene is found; the names, structure, and formula have been suppressed.

If the user wishes to conduct a search using part of a name, i.e., a name fragment resulting from the breaking of a name at some point other than a "natural" delimiter, this is possible, and, provided the beginning of the name or the name fragment is preserved, is not a particularly difficult task. The break in the name is denoted with a colon; thus BENZ: will retrieve benzene and benzophenone, but BENZO: will fail to retrieve benzene. Searches for a right-truncated name are conducted through the same inverted file as the partial name search described above and can have some interesting results. In figure 5, an NPROBE search is shown in which the user first establishes that a name fragment is to be entered. This is typically one word from a multi word name. Then, a right truncated fragment, DEXTROPIMAR:, of that one word, is entered. That this has been truncated is signified by the colon terminator. Two compounds are found in this search and use of the SSHOW command reveals that they are closely related to one another; the first is the naturally occurring diterpene, dextropimaric acid, and the second is the corresponding aldehyde, dextropimarinal.

The user may wish to omit specifying the beginning of a name, entering, for example, :ENZENE or ENZ:, and in this case, the search is much difficult. A file sorted on the full name is used for the full name search, but this approach would be very expensive for use with left-truncated names like :ENZENE, because a separate entry would be needed for each compound. Instead, the left-truncated name, with or without right truncation, is accepted and the program simply searches sequentially through the file of full names. The search is slow and expensive, and for this reason, the system will not permit a search for a left-truncated name through the complete Unified Database. Such searches can only be carried out within temporary files that result from some other search. In this way, pressure is applied to the user to conduct a preliminary screen before embarking on a search for the left-truncated name.

When NPROBE is invoked, the user must specify whether the search is to be for a whole or a fragment name, and then the name must be entered. If the first character is a colon, the program recognizes that a left-truncated name is being entered and asks for the number of the temporary file that is to be searched. After this is provided, the search begins. As with the Substructure Search option,¹ the program reports at regular intervals on the number of names searched and the number of hits, and the user may cleanly terminate the search at any time.

As with any other search option in the system, NPROBE stores retrieved substances in the form of a list of Registry Numbers in a temporary file. This temporary file can be logically combined with other files, or displayed, at the discretion of the user.

NIH-EPA STRUCTURE AND NOMENCLATURE SEARCH SYSTEM

Figure 1. Use of NPROBE with a name fragment.

Figure 2. Search for the full name "DHA."

Figure 3. Search for "toluene" in selected files.

3. Selection of Files from the Unified Database. The SELECT command permits users to restrict retrievals to specified files, as has been explained above. If a certain file has been selected, that fact is used by the program during a search, and substance that meet the search criteria are retrieved only if they are in the specific file. It is therefore impossible to impose a file selection after a search has been done, but there is another means of achieving the same result. If a search has been carried out through some number of files, or even through the whole Unified Database, and it is then realized that only substances in one particular file are of interest, it is possible, using the PULL command, to extract the Registry Numbers of that entire file and set them up in a temporary file. Once that is done, an INTERsect between that file and the earlier file of answers to a search will reduce the latter to only those Registry Numbers retrieved in the search that are in the file of interest.

Figure 4. Search for "toluene" in the Unified Database.

Figure 5. Search for a right-truncated name.

Figure 6. Example of RPROBE, PULL, and search for a left-truncated name.

An example of this procedure, in conjunction with NPROBE is shown in Figure 6. Here, an RPROBE search is conducted for all substances in the Unified Database having the basic steroid skeleton. This results in file 1, which contains 598 compounds. In a second command, the Merck Index file (No. 7) is "PULLed," and all 8959 substances are retained in temporary file 2. INTERsection of files 1 and 2 gives file 3, which contains 77 compounds which posses the steroid skeleton and which are also in the Merck Index.

This file can now be searched for all compounds with names ending in "...STEROL." The fragment entered, :STEROL, is a left-truncated name, and so the program asks for a temporary file to be identified. The user asks that file 3 be searched and the search begins. After the names for ten substances have been inspected, the program reports this fact and also notes the number of times a name ending is STEROL was found. Like SUBSSS, this search can be cleanly terminated at any time by the user.

A total of 27 of the 77 substances are found in this search and these are all stored in file 4. This file can be combined in a Boolean NOT sense with file 3, to create file 5, which contains the 50 substances which do not have the fragment STEROL in their name, although, since they were in file 3, they do have the steroid skeleton and appear in the Merck Index.

The first three substances in file 4 are shown in Figure 7, and the first two from file 5 are shown in Figure 8. File 4 does, in fact, contain substances such as gesterol, ergosterol, and cholesterol base H, in contrast to the compounds in file 5 whose names are devoid of the name fragment STEROL.

Figure 7. Steroids with the partial name "....sterol."

4. Identity Search. A common need of SANSS users is to find and retrieve a structure identical with the query structure. Thus, while it may be of interest to some users to invoke FPROB to find all chlorophenyl compounds, other users want simply to locate chlorobenzene itself, usually to learn its CAS Registry Number, which can be used to locate the compound in other CIS files or retrieve literature citation to the compound from Chemical Abstracts.

Figure 8. Steroids lacking the partial name "....sterol."

In the early versions of the SSS program, it was not possible to do this directly, and the indirect methods, e.g., FPROBE in conjunction with the molecular formula search, were expensive and inefficient. In view of the growing need for such an identity match, particularly in connection with registration of chemicals under the Toxic Substances Control Act, a program of this sort has been developed and incorporated into the SANSS.

When the program is invoked, it examines the query structure and asks the user to provide the correct number of hydrogen atoms for it. This is done to resolve some otherwise ambiguous situations resulting from the treatment of tautomeric bonding conditions in the structure of the compound. At this stage, the user has two options: the number of hydrogens can be entered; or the program can be asked to calculate and report back the correct number of hydrogens corresponding to the query structure. If the latter option is exercised, the program asks the user if there is any reason (such as nonneutral atoms) to doubt the program's proton count. If not, the identity search begins.

The number of protons in the query structure is first merged into the connection table corresponding to the query structure and the resulting set of numbers is hash-encoded. This hash-encoded information is used to search through an inverted file of similarly hash-encoded entries corresponding to each structure in the database. This is a very fast look-up type of search and locates the correct compound in a very short time. As a result, the identity search is probably the fastest and most economical method for searching through the full database.

An example of an IDENTITY search is shown in Figure 9. The command RING results in the generation of a six-membered ring, whose connection table is stored by the computer. This is followed by the command RING 5, which adds a five-membered ring to the structure. The rings are joined with the command ABOND 1 7, which creates a bond between atom 1 (in the six-membered ring) and atom 7 (in the five-membered ring). Atoms 7, 9, and 10 are defined as nitrogen, using the SATOM command, the bonds between atom 8 and 9 and 10 and 11 are defined as ring-double bonds, and finally, a two-atom branch at atom 8 is added. All atoms except those defined as nitrogen are assumed by the program to be carbon. The result is the query structure that is displayed in response to the command D (for display).

If the program IDENT is now invoked, it asks for a count of the number of hydrogen atoms in the structure. Rather than count them, the user responds by typing P, which causes the program to estimate the number of hydrogens at 17. This estimate is based upon the assumption that all bonds in the molecule, unless otherwise specified, are single bonds. After ascertaining that there is no reason to question this estimate, the search is accomplished and the program reports to the user that one compound has been found and has been stored in file

1. Inspection of this file, using the SSHOW option, reveals that the exact compound, hexazole, is in two files, and has the structure and names show.

Using a little under 20K words of DEC System 10 core, and identity search requires about 2 seconds of cpu time which means that a typical commercial cost for such a research is about $1.00, plus the cost of the structure generation.

Figure 9. IDENT search for hexazole.

SUMMARY

The SANSS occupies a central position in the NIH-EPA Chemical Information System, and consequently, its continuing development is in two areas. First, the SANSS Unified Database and software must be continually updated and maintained and, second, a variety of links between SANSS and other CIS components must be installed.

The design of the SANSS described here is felt to be generally satisfactory and the mechanisms for file-updating are now working well. As a result, this aspect of the system is approaching full development and now requires mainly a maintenance effort.

Links between SANSS and other CIS components have not yet been completed, however, and future work will focus upon this problem. Currently, one can learn that a particular compound is represented in the CIS Mass Spectral Search System, but retrieval of the mass spectrum can still be done only indirectly, using the CAS Registry Number. Direct retrieval of numeric data from within the SANSS is now possible with programs that are under test, and it is hoped that such capabilities can be made more generally available during the coming year.

REFERENCES AND NOTES

(1) R. F. Feldmann, G. W. A. Milne, S. R. Heller, A. Fein, J. A. Miller, and B. Koch, J. Chem. Inf. Comput. Sci., 17, 157 (1977).

(2) S. R. Heller, G. W. A. Milne, and R. J. Feldmann, Science, 195, 253 (1977).

(3) R. J. Feldmann and S. R. Heller, J. Chem. Doc., 12, 48 (1972).

(4) P. D. Dittmar, R. E. Stobaugh, and C. E. Watson, J. Chem. Inf. Comput. Sci., 16, 111 (1976).

(5) G. W. A. Milne and S. R. Heller, J. Chem. Inf. Comput. Sci., 16, 232 (1976).

(6) M. Bracken, J. Dorigan, J. Hushon, and J. Overbey, Chemical Substances Information Network, MITRE Technical Report, MTR-7558 (prepared under Contract CEQ7A010 for the Council on Environmental Quality), June 1977.

(7) Those wishing copies of any of these files should contact H. J. Bernstein, CIS Project, Brookhaven National Laboratory, Upton, N. Y. 11973; telephone (516) 345-4379 or FTS: 666-4379 or 800-645-1132.

(8) S. R. Heller, Anal. Chem., 44, 1951 (1972).