Biomolecular Relationships Discovered from Biological Labyrinth and Lost in Ocean of Literature: Community Efforts Can Rescue Until Automated Artificial Intelligence Takes Over

Abstract

Many brilliant minds are at work to decipher the biological labyrinth and as a result immense amount of information about biological entities and their relationships is getting accumulated in the form of published literature (Hunter and Cohen, 2006). To cater the needs of a researcher, many tools are designed to perform tasks of Named Entity Recognition (NER), Information Retrieval (IR), and Information Extraction (IE) viz. A Combined Clinical Concept Annotator (Kang et al., 2012), BANNER (Leaman and Gonzalez, 2008), Biblio-MetReS (Usie et al., 2014), BioTextQuest+ (Papanikolaou et al., 2014), BIOSMILE Web Search (Dai et al., 2008), E3Miner (Lee et al., 2008), EBIMed (Rebholz-Schuhmann et al., 2007), eFIP (Arighi et al., 2011), FACTA+ (Tsuruoka et al., 2008), GNSuite1, iHOP (Hoffmann and Valencia, 2004), MyMiner (Salgado et al., 2012), RLIMS-P(Hu et al., 2005), Anni (Jelier et al., 2008), CoPub (Frijters et al., 2008), MedScan (Novichkova et al., 2003), PPInterFinder (Raja et al., 2012), pGenN (Ding et al., 2015), SciMiner (Hur et al., 2009), BIGNER (Li et al., 2009), hybrid named entity tagger (Raja et al., 2014), and more such tools can be obtained from BIONLP resource2 and in detail analysis of many NLP tools is given by Krallinger et al. (2008) and Fleuren and Alkema (2015). Table ​Table11 gives an informational and statistical insight into some of these literature mining tools, shedding light on their efficiency translated by statistical parameters viz. F-score, recall, and precision. Many tools are domain specific like kinase family specific but still calls for human intervention for exactitude and thus limit their usage. Moreover, the data output formats are sometimes too vague as name highlighting; to be put to use for bigger literature searches. Table 1 Informational (viz. data used, parameters for evaluation and working platform) and statistical (viz. f-value, recall and precision) insights for a few literature mining tools with their brief description and links to the tools' home page. The naming ambiguity in scientific literature is one of the major concerns for NER and sentence structure for IR and IE. Presently, NER tools need to maintain a comprehensive dictionary of all names, aliases and web-repository specific IDs or have their AI (Artificial Intelligence) defined algorithms trained on many test data sets. Many such dictionaries are available but the list is ever-increasing and so is the training data set. This results into investing more money, time and effort in obtaining a comprehensive list of names, aliases and IDs. A very comprehensive work on NLP can be found on BioNLP3. The availability of manpower or intellect is huge but there is acute scarcity of funds (Bourne et al., 2015), so we have to device optimized approaches to take care of the issues discussed in subsequent section.