Databases from scientific literature

Automatic text datamining is an important source of knowledge, with many applications in generating databases from scientific literature, such as protein-disease associations, gene expression patterns, subcellular localization, and protein-protein interactions. 

Also databases of scientific literature (such as PUBMED, MEDLINE) provide additional functionality, e.g. they can search for similar articles based on word-usage analysis. Text recognition systems are being developed that automatically extract knowledge about protein function from the abstracts of scientific articles, notably on protein-protein interactions. 

Unifying disparate information sources into a single data model is an intrinsically difficult process. Even when the same kinds of information are collected in different places, it is done so for different reasons, which are reflected in data organization. A common example is extraction of information from scientific literature, where the goal is to make a collection of a particular kind of data. Since the origins of the data are research papers, such efforts often result in document databases where the article is the primary entity and the content is represented by various attributes (e.g., lists of referenced molecules). However, a document-oriented database is rarely suitable for the specific requirements of most scientific investigations. Such efforts often invert ... 

RTECS Registry of Toxic Effects of Chemical Substances number is a unique and unchanging number used to cross-reference the RTECS database, which is a compendium of data extracted from the open scientific literature. Six types of toxicity data are included in each file (1) primary irritation, (2) mutagenic effects, (3) reproductive effects, (4) tumorigenic effects, (5) acute toxicity, and (6) other multiple dose toxicity. 

The Scientific World at http //www.thescientihcworld.com/ offers literature searching through SciBase, a collection of databases of scientific, technical, and medical research literature. SciBase currently covers more than 19 million documents published since 1965 in more than 30,000 journals. SciBase content is derived from databases created by the National Library of Medicine (MEDLINE), the British Library, BIOSIS, and PASCAL, as well as CAB ABSTRACTS. Abstracts are sometimes available free and individual articles are available for purchase. 

Increasingly over recent years, data covering a wider range of species and data from nonstandard test species and systems have been published in the scientific literature. Much of these data can be found on publically available databases such as the US EPA s ECOTOX database (http //www.epa.gov/ecotox/) and a comprehensive evaluation has recently been conducted by Giddings (Giddings... 

The need to store the wealth of data produced by large scale biological projects and to provide a robust framework for analyzing and studying PPIs is obvious. Nearly all the projects described in Chapters 2 and 3, whether produced by experimental methods, in silico predictive techniques or extracted from the scientific literature, have their data collapsed and structured in databases (see Table 1). 

The size differential between a large peptide of 100 residues and a small protein, or perhaps a protein domain, is a purely arbitrary distinction. However, both (1) the technology-driven acceleration in the rate of new peptide generation and (2) the increased size limit from 50 to 100 residues will vastly increase the number of sequences that need to be accommodated within the PRF databases. The problem is exacerbated by the fact that there is a temporal continuum to peptide production. Thus, it is impossible to know how many bona fide peptides are missing for the existing databases. Many of these sequences will have been erroneously produced and immediately discarded or simply never reported in the scientific literature. Some of these unaccounted-for peptides have no doubt been physically lost and now exist only in a virtual sense. Should peptide databases include ghosts Unfortunately, such questions are relatively trivial when one considers future implieations. The total number of all possible peptides must be finite, but what a number Moreover, it is entirely reasonable to speeu-late that the sequenees eontained within the two eurrent PRF databases are but a tiny fraetion of this total. 

Information that is put into this Database is derived from published reports of crystal structure determinations. The data extracted from the scientific literature in this way include the atomic coordinates, information on the space group, chemical connectivity, and the literature reference to each structure determination. Each compound listed in the Database is identified by a six-letter code (the REFCODE), unique to each crystal structure determination. Duplicate structures and remeasurements of the same crystal structure are identified by an additional two digits after the REFCODE. Scientific journals are scanned regularly by the Database staff for reports of crystal structure determinations, and the data are then entered into this Database. Structural data are also deposited by journals, for example. Chemical Communications, that publish articles, but do not have space for atomic coordinates. All crystallographic data reported in the literature are tested by the Database staff for internal consistency, precision, and chemical reasonableness. In... 

---