A b str a ct G e n o m e -wi d e ass o ci ati o n st u di es ( G W A S) h a v e i m pr o v e d o ur u n d erst a n di n g of dis e as e a eti ol o g y b y i d e ntif yi n g g e n eti c v ari a nts ass o ci at e d wit h c o m pl e x h u m a n tr aits a n d dis e as e p h e n ot y p es. T h e G W A S C e ntr al r es o ur c e e n a bl es br o a d a n d c o n v e ni e nt i nt e gr ati v e vis u alis ati o n of, a n d a c c ess t o, s u m m ar y -l e v el G W A S d at a. G W A S r e p ort e d i n t h e bi o m e di c al lit er at ur e ar e m a n u all y c ur at e d a n d i m p ort e d i nt o t h e d at a b as e. T o a c c el er at e t h e i m p ort pr o c ess, at a ti m e w h e n t h e si z e of G W A S ar e i n cr e asi n g , w e ar e usi n g n at ur al l a n g u a g e pr o c essi n g t o st a n d ar dis e p u bli c ati o n f ull t e xt a n d t a bl es, a n d t e xt mi ni n g t o e xtr a ct G W A S d at a. W e ar e als o d e v el o pi n g a n a n n ot at e d f ull -t e xt G W A S c or p us t o e v al u at e t h e a c c ur a c y of t h e t e xt mi ni n g al g orit h m . K e y w or d s 1 G e n o m e -wi d e ass o ci ati o n st u di es , G W A S C e ntr al, t e xt mi ni n g, bi o-o nt ol o gi es 1. I ntr o d u cti o n H e alt h r es e ar c h is a d v a n c e d t hr o u g h g e n o m e -wi d e ass o ci ati o n st u di es ( G W A S) w hi c h pr o vi d e a d e e p er u n d erst a n di n g of dis e as e a eti ol o g y b y d et e cti n g ass o ci ati o ns b et w e e n g e n eti c m ar k ers a n d dis e as e p h e n ot y p es i n p o p ul ati o n s a m pl es. G W A S C e ntr al ( w w w. g w as c e ntr al. or g) is a c o m pr e h e nsi v e c oll e cti o n of s u m m ar y -l e v el G W A S d at a [ 1]. G e n eti c m ar k er d at a ar e st a n d ar dis e d wit h d b S N P i d e ntifi ers a n d p h e n ot y p es ar e a n n ot at e d usi n g M e di c al S u bj e ct H e a di n gs ( M e S H) an d H u m a n P h e n ot y p e O nt ol o g y ( H P O) t er ms. B uil di n g o n t h e ri c h s e m a nti c p h e n ot y p e a n n ot ati o n l a y er, a c or e s u bs et of G W A S d at a is m a d e a v ail a bl e as R D F n a n o p u bli c ati o ns [ 2]. R e c e ntl y, t h er e h as b e e n a l ar g e i n cr e as e i n t h e a m o u nt of d at a r e p ort e d b y i n di vi d u al st u di es t h at i n v esti g at e g e n eti c ass o ci ati o ns wit h li pi d o m e, m et a b ol o m e a n d mi cr o bi o m e p h e n ot y p es. T his , i n t ur n, r e q uir es eff orts fr o m bi o c ur at ors t o i nt er pr et a n d e xtr a ct i n cr e as e d a m o u nts of ass o ci ati o n d at a fr o m p u bli c ati o ns. As t h e si z e of G W A S i ncr e as e, r e q uiri n g e v er l ar g er d at as ets t o b e e xtr a ct e d fr o m t h e lit er at ur e a n d i m p ort e d i nt o d at a b as es, t h er e is a n e e d f or n e w bi oi nf or m ati cs c a p a biliti es t o s u p p ort s c al a bl e d at a c ur ati o n. W e h a v e d e v el o p e d a n at ur al l a n g u a g e pr o c essi n g ( N L P) t o ol kit t o e xtr a ct G W A S e ntiti es a n d r el ati o ns fr o m t h e s ci e ntifi c lit er at ur e. T h e t o ol kit li n ks p h e n ot y p e e ntiti es wit h bi o-o nt ol o g y t er ms a n d us es t h e t e xt mi ni n g Bi o C st a n d ar d f or m at f or s h ari n g t e xt d o c u m e nts a n d a n n ot ati o ns. T h e t o ol kit c a n b e us e d t o pr o c ess n e w ' u ns e e n' p u bli c ati o ns as w ell as p u bli c ati o ns t h at h a v e pr e vi o usl y b e e n s e e n a n d c ur at e d. E ntiti es t h at h a v e pr e vi o usl y b e e n m a n u all y e xtr a ct e d b y d at a b as e c ur at ors c a n b e m a p p e d b a c k o n t o p u bli c ati o ns t o g e n er at e n e w a n n ot at e d c or p or a i n t h e Bi o C st a n d ar d. H er e, w e pr es e nt A ut o C O R P us f or st a n d ar disi n g p u bli c ati o n t e xts a n d t a bl es, G W A S -T a g g er f or a n n ot ati n g c ur at e d G W A S p u bli c ati o ns, a n d G W A S -Mi n er f or e xtr a cti n g G W A S e ntiti es a n d r el ati o ns fr o m u ns e e n p u bli c ati o ns. 2. I m pl e m e nt ati o n of a n at ur al l a n g u a g e pr o c e s si n g t o ol kit 2. 1. A ut o -C O R P u s : p u bli c ati o n t e xt a n d t a bl e st a n d ar di s ati o n
converted to the BioC JSON format with each publication section (e.g., introduction, methods, results) annotated using the Information Artifact Ontology (IAO). A directed graph formed from s2e1c,t8i4o9n headers from 2,441 open access publications is used by a section prediction algorithm to assign the IAO classification. Since BioC does not provide native support for tabular data, the publication tables are converted to a BioC compliant ta-bJlSeON structure. Allabbreviations declared withinapublication are converted to JaSON output that relateesach abbreviation with the full definitioTn.he abbreviation collection supports text mining tasks such as named entity recogni(tNioEnR) by including abbreviations unique to individual publications that are not contained withi-nonbtoi ologies. 2.2.
The GWAS-Tagger application searches BioC fu-tlelxt and tabl-eJSON files for GWAS entities and relations that have previously been extreadct from a publication because of, for example, manual biocuration or text mining. Entities and relations are written to the BioC an-JdSOtaNblefiles using the BioC annotation format. The following GWAS entities, and relations between them, can be provided to GWAS-Tagger to find within publications: • Genetic marker given as a dbSNP identifier. • Phenotype given as a MeSHor HPO term. Synonyms and variations of text such as hyphenation, roman numeral usage and plurals are used if an exact term match is not found. • P-value association of the genetic marker and phenotype given as a number with optional exponential notation. 2.3.
GWAS-Miner is a hybri-dbased NLP applicationthat usesthe SciSpaCy data model for procedures such as tokenisation, pa-rotf-sentence tagging, dependency parsing, together witrhules for NERusing a combination of bio-ntologies, vocabularies and pattern matchingB. ioC full-text files are processed to identify key sentences which include relevdaanta. Sentencestructure and contexatre analysed using a dependency tree alongside the use of shortest dependency path to isolate relationGsWhipAsS.-Miner is designed to be easily updated with new ontology data as soon as ontologies are updated.
We have used GWA-STagger to annotate 700 publications that have been curated by GWAS Central. We aer manually evaluating these annotations, and editing them where necessary, to build a gold standard ful-ltext GWAS corpus. The corpus will be used to assess the precision, recall and F1 score of GWASM-iner. The supplementary materials of GWAS publicatiocnans contain large amounts of GWAS data, so we will extenAduto-CORPus to process supplementary files. Converting heterogenous supplementary formats to the BioC standard will optimise the text and table contents for text mining. Finally, we will incorporaGteWAS-Miner into the GWAS Central data import pipeline.