String search

Whereas it is unwarranted to dwell on any small differences in the absolute numbers of hits obtained from alphanumeric string searches, it should be possible to make qualitative comparisons of the data and to note important trends. If one assumes that the number of false hits and missed hits of the kinds just mentioned will be roughly comparable from one search to another, then the number of hits representing actual uses of a program or a force field will be useful on a relative basis. 

In contrast to the character string searches that were done on the full articles in the CJACS and CJWILEY databases, a different method of counting was used with MMCC Results. The newsletter usually indicates which of the... 

A search facility was provided using the CROSSBOW Bit and String search program, which is sufficient for small files but clumsy to use. However our primary aim was to obtain clean and useful WLN files, ready for conversion to a graphics system in due course, so the extra expense of connection table based searching seemed unjustified. 

The Chemical Information Database contains general information about a chemical entity or drug formulation. Originally, all this information was contained in just a MACCS database. The difficulty with keeping all the data only in MACCS is that there is multiple information about a chemical entity. One of the most obvious examples of this is the data associated with each synthesis (batch) of the compound. There is also multiple information such as synonyms for the name of the compound that also must be stored. Since MACCS is not relational, the best one can do with such data in MACCS is to string data into fields (i.e., MACCS datatypes) and then do a string search of this field when trying to retrieve the information. 

Since a canonical WLN provides a unique representation of a chemical structure, WLN has been used for compound registration. Of the major indexing and retrieval tools the Index Chemicus Registry System (ICRS) and the printed Chemical Substructure Index of the Institute for Scientific Information (ISI) are based upon WLN. The CROSSBOW (Computerized Retrieval of Oiganic Structures Based on Wiswesser) system, as its name suggests, supports structure input via WLN and the interrogation of WLN files by string search of the notations. 

The string search, i.e. a search character-by-character is of particular value to the searcher, because it allows one to search for stop-words as well. The command demands intense operations from the host, thus it can only be entered when referring to a preceding search step with not too many matches. 

QUESTEL is the only host, indicating review articles in the field Document Type. Japanese patent numbers are only searchable with the Gregorian year (Fig. 173). Like ORBIT, the QUESTEL command language contains a string search feature (Sect. 8.1.4). 

String search search for a character sequence including blanks and special characters, which costs a lot of CPU time,... 

The CROSSBOW chemical search system (1-5) is a multilevel one. Search of the bit screens is first carried out and this quickly and cheaply reduces the file to ten per cent (or less) of its original size. There will almost certainly be many false drops but most of these can usually be removed by string search of the WLNs and/or molecular formulae and/or reference numbers. String searching is slower and more expensive than bit search. Connection table generation and atom-by-atom search of the connection tables (the third search level) are still slower and even more expensive, but the atom-by-atom search program is a very powerful tool which is used in about 80% of all searches. The CROSSBOW connection tables for the hits from any search are finally used as input to a structure display program. 

Bit and string searching is an interactive process. The paramaters are input on a VDU, the hit count is displayed on-line and, if necessary, the search parameters can be modified and the search repeated. At the end of a session, search hit files are merged as required and connection table generation, atom-by-atom search and structure display are run batchwise. 

An average bit and string search takes about five minutes (searching 190,000 compounds). The CROSSBOW connection table generation program handles 1500-2000 compounds a minute, and the atom-by-atom program searches 600 compounds a minute. Well over 90% of notations are amenable to connection table generation and over 90% of the compounds on the database can be structurally displayed directly from the connection table. 

A bit search is specified using combinations of AND, OR and NOT logic, and also nested AND-within-OR logic, as required. String search allows the same logical operators, but in addition there are ignore and followed-by facilities and other specialised syntax to simplify the specification of several alternative strings. 

Figure 8 shows a typical bit and string search. In the bits search, fragments 50 and 137 must both be present. 

Atom-by-atom search is resorted to if the string search of WLNs and/or molecular formulae and/or reference numbers is not specific enough. This will occur if, for example, the substructure is very branched or may be partially or wholly embedded in a ring. An atom-by-atom search may also be needed if there is a specific positional relationship between two groups and that relationship cannot be accurately defined by WLN sub-strings. The structure in Figure 8 falls into the last category. 

---