Persistence and Availability of Web Services in Computational Biology
(A) The state of all Web service addresses listed in the abstracts. We extracted the services' Web addresses from the NAR Web Server Issue abstracts and entered them into a Web browser to check for inconsistencies. We noted that, for many pages, the original published address is no longer current. The browser is either redirected transparently or a static link on the page informs the visitor of the address change (light blue). While this is a well-meant gesture, eventually, the Web server performing the redirect will be replaced or shut off and the link will appear dead. We therefore also searched for all Web services with dead links using internet search engines to determine if they had moved to a new location (dark blue). 13% of services from 2003 can be found in this way. The percentages of services that are completely unavailable are shown in the black part of the column. Total percentages for each of these measures are given in parenthesis after their description in the key. (B) Evaluation of service functionality. We show how many services are not functional even though their Web page is still available (red). This indicates that the software behind the Web pages, the actual Web service, is failing. For users, the reason for that is impossible to determine. A large percentage of services could not be evaluated under the premises of our “fair” testing possibility (cf. Methods): They do not provide example data and on top of that either require very specific file types or complex parameter settings that are not set by default (gray). Functional services make up the largest group, but not the majority (blue). Total percentages for each of these measures are given in parenthesis after their description in the key. (C) Comparing the average number of citations for available and unavailable services. Intuitively, unusable services should have a lower number of citations (black). The number of citations is not comparable among years, because older publications have had more time to be cited. Data for the years 2008 and 2009 are not shown, as these publications have not had enough time to be cited (the same trend can be observed, but it is not yet significant). The red numbers show the percentage of services from that year's issue that are unavailable as of October 2010.