About AMENDA

Introduction

AMENDA (Automatic Mining of ENzyme DAta) is a supplement to BRENDA that is available to the public since BRENDA release 5.2 (December 2005). It contains a large amount of enzyme data which are automatically extracted from more than 26 million PubMed references (titles and abstracts).

As a subset of FRENDA, AMENDA currently covers enzyme-specific information on their occurences in organisms (more than 348,000 records, compared to about 82,000 records in BRENDA), (sub)cellular localizations (more than 77,000 records, compared to about 33,000 records in BRENDA) and source tissues (roughly 310,000 records, compared to about 90,000 records in BRENDA).

More than 26 million PubMed reference - titles and when available abstracts - (status: July 2016) are analyzed in a cooccurence based textmining approach. AMENDA uses a dictionary-based approach for detecting named entities in titles and abstracts. Search terms for enzyme names, organism names, localization, and sources and tissues are compiled from public sources.
References with enzyme hits are searched for organism names (scientific names and synonyms).
References with organism hits are searched for tissue terms and localization terms. Tissue names and synonyms, as well as localization terms and synonyms are matched against reference title, abstract, and MeSH terms.

The text mining approach described above was tested on a manually annotated text corpus of 1000 randomly selected references (listed in BRENDA as containing tissue information). A precision of approximately 87% was achieved (this means there were 13% false positives among the results).
The text mining results were classified into 4 reliability categories depending on the occurrence of search terms in title and/or abstract and/or MeSH terms. This classification is provided with the commentaries in the AMENDA database.

The AMENDA supplement is freely accessible for academic users. Commercial use requires a license - See copyright notice