A few observations resulting from me annotating many papers via BioNotate system. <br><br>1. Double Statements. <br> There are papers reporting on association of a certain variant with double phenotype, lile<br>"We studied associations of rs4646 with autism and bipolar disorders. Our findings suggest that<br>
it is positively associated with bipolar, and not associated with autism".<br> In this case it is impossible to summarize the findings with our current multi-choice, what needs to happen is<br> a possibility for a curator to mark up "autism" only as a phenotype and respective choice of summary,<br>
then go back to the same paper and mark up "bipolar" as phenotype and choose different summary. <br> Curator's instructions should change accordingly. <br><br>2. UNRELATED terms.<br> There are often phenotypes, genes and variants which are NOT related to the statement in question. <br>
"We investigated rs4646 and rs13345345 as well as variants of gene p53 to common flu and headaches". <br> Instructions should ask curators not to label any phenotypes. genes or variants which are not directly<br>
related to the statement supported by the summary choice. One other goal here is to learn synonyms from<br>assuming that all genes and phenotypes marked as such in a single abstract are related. <br><br>3. quoted support<br>
I think it will be crucially important for the next round of automation to have what we call "predicate support"<br>labeled. For example "our findings support that rs4646 is strongly associated with the risk of breast cancer". <br>
I'd label "is strongly associated with" as predicate support. Once the sentence is parsed by automatic parser<br>and both rs4646 and "breast cancer" are identified as objects of the relation, one would be able to train the <br>
NLP system using these mark up instances. <br> So, to this end in my annotations I have been using the "Sporadic observation" mark-up instead of "predicate support"<br>but it'd be good to include this in instructions and provide a special button. <br>
<br>4. a bug.<br> Using Chrome on Mac OSX, if I am trying to add new paper via PMID , type in the number and instead of clicking "Add" button just push <Enter> I get sent to Bionotate page with the abstract of the last paper on the page. <br>
<br><br>5. Double negations. Drugs. <br> There are cases when a variant is associated with the "improved reaction to anti-cancer drug X". One could be marking these up as "positively associated with phenotype" in conjunction with marking up "improved reaction to anti-cancer drug X" as a phenotype. Not sure what's a good way to handle this. Also we might want to mark up medications and think about "mediated relationships" where a variant mediates relationship between two other fluents. <br>
<br><br>-Leon <br><br><br>