Literature_NN1 mining_NN1 speed_NN1 reading_VVG In_II 2002_MC ,_, when_RRQ he_PPHS1 began_VVD to_TO make_VVI the_AT transition_NN1 from_II basic_JJ cell_NN1 biology_NN1 to_TO research_VVI into_II Alzheime_NP1 's_GE disease_NN1 ,_, Virgil_NP1 Muresan_NP1 found_VVD himself_PPX1 all_RR21 but_RR22 overwhelmed_VVN by_II the_AT sheer_JJ volume_NN1 of_IO literature_NN1 on_II the_AT disease_NN1 ._. 
He_PPHS1 and_CC his_APPGE wife_NN1 ,_, Zoia_NP1 ,_, both_RR now_RT at_II the_AT University_NN1 of_IO Medicine_NN1 and_CC Dentistry_NN1 of_IO New_NP1 Jersey_NP1 in_II Newark_NP1 ,_, were_VBDR hoping_VVG to_TO test_VVI an_AT1 idea_NN1 that_CST they_PPHS2 had_VHD developed_VVN about_II the_AT formation_NN1 of_IO the_AT protein_NN1 plaques_NN2 in_II the_AT brains_NN2 of_IO people_NN with_IW Alzheimer_NP1 's_GE disease_NN1 ._. 
But_CCB ,_, as_CSA newcomers_NN2 to_II the_AT field_NN1 ,_, they_PPHS2 were_VBDR finding_VVG it_PPH1 almost_RR impossible_JJ to_TO figure_VVI out_RP whether_CSW their_APPGE hypothesis_NN1 was_VBDZ consistent_JJ with_IW existing_JJ publications_NN2 ._. 
"_" It_PPH1 's_VBZ really_RR difficult_JJ to_TO be_VBI up_JJ31 to_JJ32 date_JJ33 with_IW so_RG much_RR being_VBG published_VVN ,_, "_" says_VVZ Virgil_NP1 Muresan_NP1 ._. 
And_CC it_PPH1 's_VBZ a_AT1 challenge_NN1 that_CST is_VBZ increasingly_RR facing_VVG researchers_NN2 in_II every_AT1 field_NN1 ._. 
The_AT 19_MC million_NNO citations_NN2 and_CC abstracts_NN2 covered_VVN by_II the_AT US_NP1 National_JJ Library_NN1 of_IO Medicine_NN1 's_GE PubMed_JJ search_NN1 engine_NN1 include_VV0 nearly_RR 830,000_MC articles_NN2 published_VVN in_II 2009_MC ,_, up_RP from_II some_DD 814,000_MC in_II 2008_MC and_CC around_RG 772,000_MC in_II 2007_MC ._. 
The_AT Muresans_NN2 ,_, however_RR ,_, were_VBDR able_JK to_TO make_VVI use_NN1 of_IO Semantic_JJ Web_NN1 Applications_NN2 in_II Neuromedicine_NP1 (_( SWAN_NP1 )_) ,_, one_MC1 of_IO a_AT1 new_JJ generation_NN1 of_IO online_JJ tools_NN2 designed_VVN to_TO help_VVI researchers_NN2 zero_VVI in_RP on_II the_AT papers_NN2 most_RGT relevant_JJ to_II their_APPGE interests_NN2 ,_, uncover_VV0 connections_NN2 and_CC gaps_NN2 that_CST might_VM not_XX otherwise_RR be_VBI obvious_JJ ,_, and_CC test_NN1 and_CC generate_VV0 new_JJ hypotheses_NN2 ._. 
The_AT use_NN1 of_IO computers_NN2 to_TO help_VVI researchers_NN2 drink_VVI from_II the_AT literature_NN1 firehose_NN1 dates_VVZ back_RP to_II the_AT early_JJ 1960s_MC2 and_CC the_AT first_MD experiments_NN2 with_IW techniques_NN2 such_II21 as_II22 keyword_NN1 searching_VVG ._. 
More_RGR recent_JJ efforts_NN2 include_VV0 the_AT striking_JJ 'maps_NN2 of_IO science'_NN1 that_CST cluster_VV0 papers_NN2 together_RL on_II the_AT basis_NN1 of_IO how_RGQ often_RR they_PPHS2 cite_VV0 one_PPX121 another_PPX122 ,_, or_CC by_II similarities_NN2 in_II the_AT frequencies_NN2 of_IO certain_JJ keywords_NN2 ._. 
As_RG fascinating_JJ as_CSA these_DD2 maps_NN2 can_VM be_VBI ,_, however_RR ,_, they_PPHS2 do_VD0 n't_XX get_VVI at_II the_AT semantics_NN1 of_IO the_AT papers_NN2 --_NN1 the_AT fact_NN1 that_CST they_PPHS2 are_VBR talking_VVG about_II specific_JJ entities_NN2 such_II21 as_II22 genes_NN2 and_CC proteins_NN2 ,_, and_CC making_VVG assertions_NN2 about_II those_DD2 entities_NN2 (_( such_II21 as_II22 gene_NN1 X_ZZ1 regulates_VVZ gene_NN1 Y_ZZ1 )_) ._. 
Deeper_RRR meaning_VVG The_AT goal_NN1 of_IO these_DD2 tools_NN2 is_VBZ to_TO help_VVI researchers_NN2 analyze_VVI and_CC integrate_VVI the_AT literature_NN1 more_RGR efficiently_RR than_CSN they_PPHS2 can_VM do_VDI through_II their_APPGE own_DA reading_NN1 ,_, to_TO hone_VVI in_RP on_II the_AT most_RGT fruitful_JJ experiments_NN2 to_TO do_VDI and_CC to_TO make_VVI new_JJ predictions_NN2 of_IO gene_NN1 functions_NN2 ,_, say_VV0 ,_, or_CC drug_NN1 side_NN1 effects_NN2 ._. 
The_AT first_MD step_NN1 towards_II that_DD1 goal_NN1 is_VBZ for_IF the_AT text-_NN1 or_CC semantic-mining_JJ tool_NN1 to_TO recognize_VVI key_JJ terms_NN2 ,_, or_CC entities_NN2 ,_, such_II21 as_II22 genes_NN2 and_CC proteins_NN2 ._. 
For_REX21 example_REX22 ,_, academic_JJ publisher_NN1 Elsevier_NN1 ,_, headquartered_VVD in_II Amsterdam_NP1 ,_, has_VHZ piloted_VVN Reflect_VV0 in_II two_MC recent_JJ online_JJ issues_NN2 of_IO its_APPGE journal_NN1 Cell_NN1 ._. 
The_AT technology_NN1 was_VBDZ developed_VVN at_II the_AT European_JJ Molecular_JJ Biology_NN1 Laboratory_NN1 in_II Heidelberg_NP1 ,_, Germany_NP1 ,_, and_CC won_VVD Elsevier_NP1 's_GE Grand_JJ Challenge_NN1 2009_MC competition_NN1 for_IF new_JJ tools_NN2 that_CST improve_VV0 the_AT communication_NN1 and_CC use_NN1 of_IO scientific_JJ information_NN1 ._. 
Reflect_VV0 automatically_RR recognizes_VVZ and_CC highlights_VVZ the_AT names_NN2 of_IO genes_NN2 ,_, proteins_NN2 and_CC small_JJ molecules_NN2 in_II the_AT Cell_NN1 articles_NN2 ._. 
Users_NN2 clicking_VVG on_II a_AT1 highlighted_JJ term_NN1 will_VM see_VVI a_AT1 pop-up_NN1 box_NN1 containing_VVG information_NN1 related_VVN to_II that_DD1 term_NN1 ,_, such_II21 as_II22 sequence_NN1 data_NN and_CC molecular_JJ structures_NN2 ,_, along_II21 with_II22 links_NN2 to_II the_AT sources_NN2 of_IO the_AT data_NN ._. 
Reflect_VV0 obtains_VVZ this_DD1 information_NN1 from_II its_APPGE dictionary_NN1 of_IO millions_NNO2 of_IO proteins_NN2 and_CC small_JJ molecules_NN2 ._. 
Such_DA 'entity_NN1 recognition'_NN1 can_VM be_VBI done_VDN fairly_RR accurately_RR by_II many_DA2 mining_NN1 tools_NN2 today_RT ._. 
But_CCB other_JJ tools_NN2 take_VV0 on_RP the_AT tougher_JJR challenge_NN1 of_IO recognizing_VVG relationships_NN2 between_II the_AT entities_NN2 ._. 
Researchers_NN2 from_II Leiden_NP1 University_NN1 and_CC Erasmus_NP1 University_NN1 in_II Rotterdam_NP1 ,_, both_RR in_II the_AT Netherlands_NP1 ,_, have_VH0 developed_VVN software_NN1 called_VVN Peregrine_NN1 ,_, and_CC used_VVN it_PPH1 to_TO predict_VVI an_AT1 undocumented_JJ interaction_NN1 between_II two_MC proteins_NN2 :_: calpain_VV0 3_MC ,_, which_DDQ when_CS mutated_JJ causes_NN2 a_AT1 type_NN1 of_IO muscular_JJ dystrophy_NN1 ,_, and_CC parvalbumin_NN1 B_ZZ1 ,_, which_DDQ is_VBZ found_VVN mainly_RR in_II skeletal_JJ muscle_NN1 ._. 
Their_APPGE analysis_NN1 found_VVD that_CST these_DD2 proteins_NN2 frequently_RR co-occurred_VVD in_II the_AT literature_NN1 with_IW other_JJ key_JJ terms_NN2 ._. 
Experiments_NN2 then_RT validated_VVD that_CST the_AT two_MC proteins_NN2 do_VD0 indeed_RR physically_RR interact_VVI (_( H._NP1 H._NP1 van_NP1 Haagen_NP1 et_RA21 al_RA22 ._. 
PLoS_NP2 One_MC1 4_MC ,_, e7894_FO ;_; 2009_MC )_) ._. 
Development_NN1 role_NN1 At_II the_AT University_NN1 of_IO Colorado_NP1 in_II Denver_NP1 ,_, bioinformatician_NN1 Lawrence_NP1 Hunter_NP1 and_CC his_APPGE research_NN1 group_NN1 have_VH0 developed_VVN a_AT1 tool_NN1 called_VVN the_AT Hanalyzer_NP1 (_( short_JJ for_IF 'high-throughput_JJ analyzer'_NN1 )_) ,_, and_CC have_VH0 used_VVN it_PPH1 to_TO predict_VVI the_AT role_NN1 of_IO four_MC genes_NN2 in_II mouse_NN1 craniofacial_JJ development_NN1 ._. 
They_PPHS2 gathered_VVD gene-expression_JJ data_NN from_II three_MC facial_JJ tissues_NN2 in_II developing_JJ mice_NN2 and_CC generated_VVD a_AT1 'data_NN1 network'_NN1 showing_VVG which_DDQ genes_NN2 were_VBDR active_JJ together_RL at_II what_DDQ stage_NN1 of_IO development_NN1 ,_, and_CC in_II which_DDQ tissues_NN2 ._. 
The_AT team_NN1 also_RR mined_VVD relevant_JJ abstracts_NN2 and_CC molecular_JJ databases_NN2 for_IF information_NN1 about_II those_DD2 genes_NN2 and_CC used_VVD this_DD1 to_TO create_VVI a_AT1 'knowledge_NN1 network'_NN1 ._. 
Using_VVG both_DB2 networks_NN2 ,_, the_AT researchers_NN2 homed_VVD in_RP on_II a_AT1 group_NN1 of_IO 20_MC genes_NN2 that_CST were_VBDR upregulated_VVN at_II the_AT same_DA time_NNT1 ,_, first_MD in_II the_AT mandible_NN1 (_( lower_JJR jaw_NN1 area_NN1 )_) and_CC then_RT ,_, about_RG 36_MC hours_NNT2 later_RRR ,_, in_II the_AT maxilla_NN1 (_( upper_JJ jaw_NN1 )_) ._. 
A_AT1 closer_JJR look_NN1 at_II the_AT knowledge_NN1 network_NN1 suggested_VVD that_CST these_DD2 genes_NN2 were_VBDR involved_JJ in_II tongue_NN1 development_NN1 ,_, because_CS the_AT tongue_NN1 is_VBZ the_AT largest_JJT muscle_NN1 group_NN1 in_II the_AT head_NN1 and_CC is_VBZ in_II the_AT mandible_NN1 ._. 
Further_JJR analysis_NN1 led_VVD them_PPHO2 to_II four_MC other_JJ genes_NN2 that_CST had_VHD not_XX been_VBN previously_RR linked_VVN to_II craniofacial_JJ muscle_NN1 development_NN1 but_CCB that_DD1 were_VBDR active_JJ in_II the_AT same_DA area_NN1 at_II the_AT same_DA time_NNT1 ._. 
Subsequent_JJ experiments_NN2 confirmed_VVD that_CST these_DD2 genes_NN2 were_VBDR also_RR involved_JJ in_II tongue_NN1 development_NN1 (_( S._NP1 M._NN1 Leach_NP1 et_RA21 al_RA22 ._. 
PLoS_NP2 Comput_VV0 ._. 
Biol._NP1 5_MC ,_, e1000215_FO ;_; 2009_MC )_) ._. 
"_" Somebody_PN1 staring_VVG at_II the_AT data_NN or_CC using_VVG existing_JJ tools_NN2 would_VM never_RR come_VVI up_RP with_IW this_DD1 hypothesis_NN1 ._. "_" 
Lawrence_NP1 Hunter_NP1 "_" I_PPIS1 do_VD0 n't_XX see_VVI that_CST there_EX is_VBZ any_DD way_NN1 that_CST somebody_PN1 staring_VVG at_II the_AT data_NN or_CC using_VVG existing_JJ tools_NN2 would_VM have_VHI ever_RR come_VVN up_RP with_IW this_DD1 hypothesis_NN1 ,_, "_" says_VVZ Hunter_NP1 ._. 
Although_CS extracting_VVG entities_NN2 and_CC the_AT relationships_NN2 between_II them_PPHO2 is_VBZ a_AT1 common_JJ approach_NN1 for_IF literature-mining_JJ tools_NN2 ,_, it_PPH1 is_VBZ not_XX enough_RR to_TO pull_VVI out_RP the_AT full_JJ meaning_NN1 of_IO research_NN1 papers_NN2 ,_, says_VVZ Anita_NP1 de_NP1 Waard_NP1 ,_, a_AT1 researcher_NN1 of_IO disruptive_JJ technologies_NN2 at_II Elsevier_JJR Labs_NN2 in_II Amsterdam_NP1 ._. 
Scientific_JJ articles_NN2 typically_RR lay_VVD out_RP a_AT1 set_NN1 of_IO core_NN1 claims_NN2 ,_, together_RL with_IW the_AT empirical_JJ evidence_NN1 that_CST supports_VVZ them_PPHO2 ,_, and_CC then_RT use_VV0 those_DD2 claims_NN2 to_TO argue_VVI for_IF a_AT1 conclusion_NN1 or_CC hypothesis_NN1 ._. 
"_" Generally_RR that_DD1 's_VBZ where_RRQ the_AT real_JJ ,_, interesting_JJ science_NN1 is_VBZ ,_, "_" de_NP1 Waard_NP1 says_VVZ ._. 
Capturing_VVG the_AT higher-level_JJ argument_NN1 is_VBZ an_AT1 even_RR more_RGR difficult_JJ task_NN1 for_IF a_AT1 computer_NN1 ,_, but_CCB a_AT1 small_JJ number_NN1 of_IO groups_NN2 ,_, such_II21 as_II22 the_AT SWAN_NN1 group_NN1 ,_, are_VBR trying_VVG to_TO do_VDI so_RR ._. 
The_AT SWAN_NN1 website_NN1 ,_, which_DDQ opened_VVD to_II the_AT public_NN1 in_II May_NPM1 2009_MC ,_, was_VBDZ developed_VVN by_II two_MC Boston-based_JJ groups_NN2 ,_, the_AT Massachusetts_NP1 General_JJ Hospital_NN1 and_CC the_AT Alzheimer_NP1 Research_NN1 Forum_NN1 ,_, a_AT1 community_NN1 and_CC news_NN1 website_NN1 for_IF Alzheimer_NP1 's_GE researchers_NN2 ._. 
For_IF each_DD1 hypothesis_NN1 in_II the_AT system_NN1 ,_, SWAN_NP1 shows_VVZ the_AT factual_JJ claims_NN2 that_CST support_VV0 it_PPH1 ,_, plus_II links_NN2 to_II the_AT papers_NN2 supporting_VVG each_DD1 claim_NN1 ._. 
Because_CS claims_NN2 from_II the_AT various_JJ hypotheses_NN2 are_VBR linked_VVN together_RL in_II a_AT1 network_NN1 ,_, a_AT1 user_NN1 can_VM browse_VVI from_II one_PN1 to_II the_AT next_MD and_CC see_VV0 the_AT connections_NN2 between_II them_PPHO2 ._. 
The_AT visualization_NN1 tool_NN1 uses_VVZ a_AT1 red_JJ icon_NN1 to_TO show_VVI when_RRQ two_MC claims_NN2 conflict_NN1 and_CC a_AT1 green_JJ icon_NN1 to_TO show_VVI when_RRQ they_PPHS2 're_VBR consistent_JJ ,_, allowing_VVG the_AT user_NN1 to_TO see_VVI at_II a_AT1 glance_NN1 which_DDQ hypotheses_NN2 are_VBR controversial_JJ and_CC which_DDQ are_VBR well_RR supported_VVN by_II the_AT literature_NN1 (_( see_VV0 graphics_NN ,_, above_RL )_) ._. 
At_II the_AT moment_NN1 ,_, this_DD1 information_NN1 is_VBZ unlikely_JJ to_TO surprise_VVI experts_NN2 in_II Alzheimer_NP1 's_GE disease_NN1 ._. 
In_II its_APPGE current_JJ stage_NN1 of_IO development_NN1 ,_, SWAN_NP1 may_VM be_VBI more_RGR useful_JJ for_IF newcomers_NN2 trying_VVG to_TO get_VVI up_RP to_TO speed_VVI on_II the_AT subject_NN1 ._. 
Beneficiaries_NN2 could_VM include_VVI more_RGR established_JJ scientists_NN2 such_II21 as_II22 the_AT Muresans_NN2 who_PNQS want_VV0 to_TO move_VVI into_II a_AT1 different_JJ field_NN1 ,_, or_CC researchers_NN2 with_IW a_AT1 pharmaceutical_JJ or_CC biotech_VV0 company_NN1 who_PNQS have_VH0 just_RR been_VBN put_VVN on_II an_AT1 Alzheimer_NP1 's_GE disease_NN1 project_NN1 ._. 
Building_NN1 up_II SWAN_NP1 also_RR has_VHZ scalability_NN1 issues_NN2 ._. 
The_AT vast_JJ majority_NN1 of_IO the_AT hypotheses_NN2 ,_, claims_NN2 and_CC literature_NN1 links_VVZ in_II SWAN_NP1 have_VH0 been_VBN annotated_VVN and_CC entered_VVN by_II the_AT site_NN1 's_GE curator_NN1 ,_, Gwen_NP1 Wong_NP1 ,_, with_IW the_AT help_NN1 of_IO authors_NN2 ._. 
This_DD1 curation_NN1 is_VBZ a_AT1 painstaking_JJ process_NN1 that_CST has_VHZ so_RG far_RR produced_VVN only_RR 1,933_MC claims_NN2 and_CC 47_MC fully_RR annotated_VVN hypotheses_NN2 ._. 
But_CCB the_AT intent_NN1 is_VBZ for_IF these_DD2 early_JJ hand-curation_JJ efforts_NN2 to_TO set_VVI a_AT1 'gold_NN1 standard'_NN1 for_IF how_RRQ the_AT SWAN_NN1 knowledge_NN1 base_NN1 should_VM be_VBI built_VVN by_II the_AT community_NN1 as_II a_AT1 whole_NN1 ._. 
The_AT SWAN_NN1 developers_NN2 plan_VV0 to_TO improve_VVI the_AT user_NN1 interface_NN1 to_TO encourage_VVI scientists_NN2 to_TO submit_VVI their_APPGE own_DA hypotheses_NN2 ,_, post_NN1 comments_NN2 and_CC even_RR do_VD0 some_DD of_IO the_AT curation_NN1 themselves_PPX2 ._. 
The_AT need_NN1 for_IF some_DD level_NN1 of_IO manual_JJ curation_NN1 is_VBZ common_JJ to_II the_AT various_JJ literature_NN1 tools_NN2 ,_, and_CC limits_VVZ their_APPGE scalability_NN1 ._. 
The_AT SWAN_NN1 team_NN1 is_VBZ working_VVG to_TO automate_VVI parts_NN2 of_IO the_AT curation_NN1 process_NN1 ,_, such_II21 as_II22 extracting_VVG gene_NN1 names_NN2 ._. 
Elsewhere_RL ,_, de_NP1 Waard_NP1 and_CC other_JJ researchers_NN2 are_VBR investigating_VVG ways_NN2 of_IO automatically_RR recognizing_VVG hypotheses_NN2 --_NN1 for_REX21 example_REX22 ,_, by_II looking_VVG for_IF specific_JJ word_NN1 patterns_NN2 ._. 
For_IF most_DAT of_IO these_DD2 tools_NN2 ,_, however_RR ,_, curation_NN1 is_VBZ unlikely_JJ to_TO become_VVI fully_RR automated_JJ ._. 
"_" Literature_NN1 mining_NN1 is_VBZ hard_JJ to_TO do_VDI in_II a_AT1 way_NN1 that_CST is_VBZ both_RR high_JJ scale_NN1 and_CC high_JJ accuracy_NN1 ,_, "_" says_VVZ John_NP1 Wilbanks_NP1 ,_, director_NN1 of_IO Science_NN1 Commons_NP ,_, a_AT1 data-sharing_JJ initiative_NN1 in_II Cambridge_NP1 ,_, Massachusetts_NP1 ._. 
Developers_NN2 say_VV0 a_AT1 more_RGR likely_JJ solution_NN1 ,_, at_RR21 least_RR22 in_II the_AT short_JJ term_NN1 ,_, is_VBZ that_DD1 papers_NN2 will_VM have_VHI to_TO be_VBI curated_VVN and_CC annotated_VVN through_II some_DD combination_NN1 of_IO automated_JJ tools_NN2 ,_, professional_JJ curators_NN2 and_CC the_AT papers_NN2 '_GE authors_NN2 ,_, who_PNQS might_VM ,_, for_REX21 example_REX22 ,_, be_VBI prevailed_VVN on_II21 to_II22 write_VV0 their_APPGE abstracts_NN2 in_II a_AT1 more_RGR structured_JJ machine-readable_JJ form_NN1 ._. 
The_AT right_JJ people_NN Are_VBR authors_NN2 willing_JJ to_TO add_VVI to_II the_AT already_RR arduous_JJ task_NN1 of_IO writing_VVG an_AT1 article_NN1 ?_? 
And_CC are_VBR authors_NN2 even_RR the_AT best_JJT people_NN to_TO do_VDI this_DD1 job_NN1 ?_? 
The_AT journal_NN1 FEBS_NPM2 Letters_NN2 experimented_VVN in_II 2009_MC with_IW structured_JJ digital_JJ abstracts_NN2 to_TO see_VVI how_RRQ authors_NN2 would_VM respond_VVI and_CC perform_VVI in_II shaping_VVG their_APPGE own_DA machine-readable_JJ abstracts_NN2 ._. 
The_AT results_NN2 were_VBDR not_XX encouraging_JJ ._. 
Authors_NN2 presented_VVD their_APPGE abstracts_NN2 about_II protein_NN1 --_NN1 protein_NN1 interactions_NN2 as_CSA structured_JJ paragraphs_NN2 describing_VVG entities_NN2 ,_, the_AT relationships_NN2 between_II the_AT entities_NN2 and_CC their_APPGE methods_NN2 using_VVG specific_JJ ,_, simple_JJ vocabularies_NN2 (_( for_REX21 example_REX22 ,_, 'protein_NN1 A_ZZ1 interacts_VVZ with_IW protein_NN1 B'_NP1 )_) ._. 
But_CCB the_AT curators_NN2 of_IO a_AT1 protein_NN1 database_NN1 did_VDD n't_XX accept_VVI them_PPHO2 ,_, says_VVZ de_NP1 Waard_NP1 ._. 
"_" Authors_NN2 are_VBR not_XX the_AT right_JJ people_NN to_TO validate_VVI their_APPGE own_DA claims_NN2 ,_, "_" she_PPHS1 says_VVZ ._. 
The_AT community_NN1 --_NN1 referees_NN2 ,_, editors_NN2 ,_, curators_NN2 ,_, readers_NN2 at_II large_JJ --_NN1 is_VBZ still_RR needed_VVN ._. 
This_DD1 could_VM be_VBI a_AT1 business_NN1 opportunity_NN1 for_IF the_AT publishers_NN2 ,_, says_VVZ Wilbanks_NP1 :_: they_PPHS2 could_VM curate_NN1 and_CC mark_VV0 up_RP their_APPGE publications_NN2 for_IF text_NN1 and_CC semantic_JJ mining_NN1 and_CC provide_VV0 that_CST as_II a_AT1 value-added_JJ service_NN1 ._. 
"_" There_EX 's_VBZ a_AT1 lot_NN1 of_IO business_NN1 out_RP there_RL for_IF the_AT publishers_NN2 ,_, but_CCB it_PPH1 's_VBZ not_XX the_AT same_DA business_NN1 ,_, "_" says_VVZ Allen_NP1 Renear_NP1 ,_, associate_JJ dean_NN1 for_IF research_NN1 at_II the_AT Graduate_NN1 School_NN1 of_IO Library_NN1 and_CC Information_NN1 Science_NN1 at_II the_AT University_NN1 of_IO Illinois_NP1 at_II Urbana-Champaign_NP1 ._. 
"_" If_CS they_PPHS2 keep_VV0 making_VVG PDFs_NP2 ,_, that_DD1 's_VBZ not_XX going_VVG to_II work_NN1 for_IF them_PPHO2 ._. 
They_PPHS2 have_VH0 to_TO get_VVI into_II more_DAR of_IO the_AT semantic_JJ side_NN1 of_IO this_DD1 ._. "_" 
Perhaps_RR the_AT largest_JJT challenge_NN1 is_VBZ getting_VVG scientists_NN2 to_TO use_VVI these_DD2 tools_NN2 ._. 
It_PPH1 will_VM be_VBI up_II21 to_II22 the_AT developers_NN2 to_TO demonstrate_VVI the_AT benefits_NN2 and_CC make_VVI their_APPGE wares_NN2 easy_JJ to_TO use_VVI ._. 
That_DD1 's_VBZ going_VVGK to_TO be_VBI difficult_JJ ,_, says_VVZ Hunter_NP1 ._. 
Academic_JJ informaticians_NN2 are_VBR rewarded_VVN more_RRR for_IF coming_VVG up_RP with_IW new_JJ algorithms_NN2 ,_, and_CC less_RRR for_IF making_VVG their_APPGE programs_NN2 usable_JJ and_CC widely_RR adoptable_JJ by_II biomedical_JJ scientists_NN2 ,_, he_PPHS1 says_VVZ ._. 
Only_RR a_AT1 few_DA2 tools_NN2 are_VBR being_VBG developed_VVN by_II companies_NN2 for_IF more_RGR widespread_JJ use_NN1 ._. 
Major_JJ issues_NN2 that_CST all_DB technology_NN1 developers_NN2 will_VM need_VVI to_TO tackle_VVI are_VBR transparency_NN1 ,_, provenance_NN1 and_CC trust_NN1 ._. 
Scientists_NN2 wo_VM n't_XX trust_VVI what_DDQ a_AT1 computer_NN1 is_VBZ suggesting_VVG in_II31 terms_II32 of_II33 new_JJ connections_NN2 or_CC hypotheses_NN2 if_CS they_PPHS2 do_VD0 n't_XX know_VVI how_RRQ the_AT results_NN2 were_VBDR generated_VVN and_CC what_DDQ the_AT primary_JJ sources_NN2 were_VBDR ._. 
"_" We_PPIS2 as_II informaticians_NN2 are_VBR going_VVGK to_TO have_VHI to_TO take_VVI on_II these_DD2 more_DAR user-driven_NN1 and_CC less_RGR technology-driven_JJ problems_NN2 ,_, "_" says_VVZ Hunter_NP1 ._. 
Even_CS21 if_CS22 researchers_NN2 do_VD0 start_VVI to_TO trust_VVI the_AT new_JJ tools_NN2 ,_, it_PPH1 's_VBZ not_XX clear_JJ how_RGQ much_DA1 of_IO their_APPGE reading_NN1 they_PPHS2 will_VM delegate_VVI ._. 
"_" As_CSA reading_NN1 becomes_VVZ more_RGR effective_JJ ,_, "_" says_VVZ Renear_NP1 ,_, "_" some_DD people_NN have_VH0 speculated_VVN that_CST we_PPIS2 wo_VM n't_XX do_VDI as_CSA much_RR because_CS we_PPIS2 'll_VM get_VVI done_VDN what_DDQ we_PPIS2 need_VV0 to_TO do_VDI sooner_RRR ._. "_" 
Or_CC ,_, he_PPHS1 says_VVZ ,_, "_" it_PPH1 may_VM be_VBI that_CST we_PPIS2 'll_VM do_VDI more_DAR reading_NN1 because_CS it_PPH1 's_VBZ more_RGR valuable_JJ ._. 
Which_DDQ one_PN1 is_VBZ true_JJ is_VBZ actually_RR an_AT1 empirical_JJ question_NN1 ._. "_" 
Analyzing_VVG articles_NN2 in_II new_JJ ways_NN2 leads_VVZ to_II the_AT larger_JJR question_NN1 of_IO whether_CSW the_AT articles_NN2 themselves_PPX2 should_VM change_VVI in_II structure_NN1 ._. 
If_CS an_AT1 article_NN1 is_VBZ to_TO be_VBI boiled_VVN down_RP into_II machine-readable_JJ bits_NN2 ,_, why_RRQ bother_VVI writing_VVG whole_JJ articles_NN2 in_II the_AT first_MD place_NN1 ?_? 
Why_RRQ do_VD0 n't_XX researchers_NN2 just_RR deal_VVI with_IW statements_NN2 and_CC facts_NN2 and_CC distribute_VVI and_CC mash_VVI them_PPHO2 up_RP to_TO generate_VVI hypotheses_NN2 and_CC knowledge_NN1 ?_? 
"_" Human_JJ commentary_NN1 and_CC insight_NN1 are_VBR still_RR extraordinarily_RR valuable_JJ ,_, "_" says_VVZ Martone_NN1 ._. 
"_" Those_DD2 insights_NN2 do_VD0 n't_XX immediately_RR fall_VVI out_II21 of_II22 data_NN without_IW human_JJ ingenuity_NN1 ._. 
So_RR you_PPY need_VV0 to_TO be_VBI able_JK to_TO communicate_VVI that_DD1 and_CC that_CST generally_RR means_VVZ building_VVG an_AT1 argument_NN1 and_CC a_AT1 set_NN1 of_IO supporting_JJ claims_NN2 ._. 
These_DD2 things_NN2 are_VBR not_XX going_VVGK to_TO go_VVI away_RL any_DD time_NNT1 soon_RR ._. "_" 
