Difference between revisions of "First-pass schedule, instructions, automation"

From WormBaseWiki
Jump to navigationJump to search
Line 35: Line 35:
  
 
=[http://www.wormbase.org/wiki/index.php/%3DAutomation_progress%3D Automation progress]=
 
=[http://www.wormbase.org/wiki/index.php/%3DAutomation_progress%3D Automation progress]=
==[http://goldturtle.caltech.edu/wcat/ Texpresso's Automation Explanation and Schedule   http://goldturtle.caltech.edu/wcat/]==
+
==[http://goldturtle.caltech.edu/wcat/ Texpresso's Automation Explanation and Schedule   http://goldturtle.caltech.edu/wcat/]==
DONE = Automated flagging is incorporated into the pipeline and comments show up on the first-pass form.
 
It does not indicate 100% precision or 100% recall.
 
 
 
{| border=2 {{table}}
 
| align="center" style="background:#f0f0f0;"|'''DONE?'''
 
| align="center" style="background:#f0f0f0;"|'''Data Type'''
 
| align="center" style="background:#f0f0f0;"|'''Who'''
 
| align="center" style="background:#f0f0f0;"|'''Notes'''
 
|-
 
|||Gene Regulation||Xiaodong/Arun||
 
|-
 
|'''YES'''||RNAi small scale||Gary/Juancarlos/Ruihua||
 
|-
 
|'''YES'''||antibody||Wen/Juancarlos||ready for postgres population 4/8/09
 
|-
 
|||allele extraction||Jolene/Arun||
 
|-
 
|||human disease||Ranjana/Juancarlos||
 
|-
 
|||chemicals/drugs||Karen/Michael/Juancarlos||
 
|}
 
  
 
----
 
----

Revision as of 20:51, 6 May 2009

[back]

First-pass data types explained

Link to an explanation of the data types as they existed on the old curator firstpass form

Texpresso/Author/Curator interim form

Link to an explanation of the data types on the new curator form that combines data flagging from all three sources.

Genetics author first pass form

First-Pass Rotation

First-Pass Curator Two-week period Landmarks
Karen 2/16-3/1
Ranjana 3/2-3/15
Raymond 3/16-3/29
Jolene 3/30-4/12 New forms active
Gary 4/13-4/26 Author forms sent out
Xiaodong 4/27-5/10 Author feedback coming in
Erich 5/11-5/24
Wen 5/25- 6/7
Kimberly 6/8-6/21

Automation progress

Texpresso's Automation Explanation and Schedule http://goldturtle.caltech.edu/wcat/


First-Pass a paper

Access papers on the WBPaperEditor page

Pick a paper and access the curation form

  • Go to http://tazendra.caltech.edu/~postgres/cgi-bin/wbpaper_editor.cgi
  • Choose your name
  • Scroll down the page and select "Not Curated Plus Textpresso!" The body of these papers have passed through Textpresso and should report all found datatypes in the automated pipeline in the appropriate fields.
  • Scroll/Page down to a paper and select "Curate!"


Alternatively

  • Access the curation.cgi from the WBPaperID page itself
  • Select the WBPaperID from left column to take you to the paper page-- ONLY SELECT PAPERS FROM WBPaper0030000 AND LATER
  • Select first-pass curate

Note: the paper pdf can be accessed from the paper page along with supplemental materials.

Either action takes you to the curation.cgi (SEE BELOW)

New firstpass curation.cgi

Old firstpass page curation.cgi

  • At the top of the page are links to:
    • The tazendra site map (other forms)
    • Documentation for the set up, paths from, and changes to the curation.cgi form
    • Guidelines for the form which include some summary information about the fields and features, written by Raymond in 2001 and needs some updating.
  • For each data type you can check the box or enter text:
    • Check the box = a '"yes" is entered in to the field
    • Enter text = the text is recorded
  • E-mail is set for default send if there is a "yes" (from a check) or text in the data type flag box.
  • When done, you can see the preview of the submission, by selecting "Preview!"
    • If acceptable, Select "New Entry!" and flags will be sent to data curators.

Saving data: If you want to come back to the paper later, you can choose to 'Save!' the data. If you do this the next time you 'Load!' the data, the form will NOT send flags to curators unless you specify it to do so. You can restore the default 'send all to curators that have data type info' through the buttons at the top right of the form.

Adding new gene paper connections

You can add gene paper connections through the WBPaper editor page. You can search directly for the paper on this page or you can access the paper by hitting "DISPLAY ALL!" and choosing the paper from the left column link. You will be taken to a summary page for the paper, which at the bottom of the page gives you an area to confirm or add new genes associated with that paper.


Other Tasks

Who When Task Goal
Jolene current with Ruihua: create curation textpresso webpage workspace automating newmutant sentence identification/extraction for phenotype curation.
Wen DONE make foreign language tag for Journals/Articles Clear FP list of non-English language articles, which can't be curated
Andrei DONE Analyze author fill-out form To provide a summary of what works and what not and to seek improvements of the procedure.
Andrei DONE Work out the correspondence of fields between the author and curator forms.
Juancarlos DONE Waiting for consensus on what to put in fields for author and curator for FP.
Raymond suggests : To have a first pass form for curators that shows author's 
submission for curator's approval (e.g. a tick) for that information to be sent 
(along with whatever information a curator puts in) for data extraction. If the 
first-pass curator dis-approves author's input (by not ticking), then the author's 
input will not be further processed or sent but it will be nevertheless stored as 
is in the database. For the next phase, results from automated first-pass via 
textpresso could be treated similarly as that of the author's. The ultimate goal 
is to maximize the number of fields that need no curator approval. Sources of 
first-pass curation should be clearly distinguished by a person ID (textpresso 
will be assigned one).
Juancarlos is okay with this, but while I'm leaning towards assigning the author 
response to a PersonID, if this is not going to be used for evidence anywhere, 
the corresponding email is possibly the more accurate evidence since the receiver 
may pass it on to someone who is not the WBPerson that email is assigned to.  In 
that case Textpresso wouldn't need a person ID.  I'm still leaning toward using 
IDs though, I'm just not sure it reflects the right things if we ever want that 
in WB or something like that.
Juancarlos also needs to know how curators want to enter data.  For any given 
Paper-Field, would curators want to be able to enter unlimited entries, and make 
then invalid to delete ?  Would you prefer the current system where there's only 
a single box where everyone mushes in all data ?  Would you care about the history 
of deleted things ?|  
Juancarlos DONE Resolve duplicates There are many papers on the firstpass list that are already firstpassed. Most of these papers are duplicates and have two WBPaperID assignments. Is there a way to resolve this?.
Juancarlos DONE Assign Publication Type 'Review' to all papers that are annotated as 'review' in the Comments section of the FP curation form Remove these papers from the corpus that textpresso searches for data type patterns.
Arun DONE Increase scans for transgene objects to occur daily and change pipeline to run on new papers only
Juancarlos DONE Sort papers based on first pass checkout list based on whether or not they have been passed through Textpresso Curators can now focus on papers whose data fields have already be filled in by textpresso.

Unassigned tasks and comments that need more discussion

  • We do have a record of data type curation for each paper, is there some way of combining the first pass curation with the curation status form?

-The curation status form gets flagged data from the first pass form (don't think I understand the question) -- Juancarlos

  • How does the false positive work?

- the words ``false positive get appended (or prepended, I forget and can't find an example) into the text field -- Juancarlos

  • Curator preference for FP remarks, can people deal with the not getting detailed notes about their data type and where it exists in the paper or should this be a mandatory part of first-pass?


  • Discrepancy between FP papers and total papers curated. For some data types, curators get to the paper before the FP curator, it would be good to know that a curator already touched it. Is there a way to mark the paper in the FP list as curated for a specific data type but not others?
"I have curated some papers that did not go through first pass, mostly Expr_pattern papers published before 2001. 
Is it possible that you check current WormBase, find all the paper with Expr_pattern data and flag these papers 
in postgres as Expr_pattern "yes"?" WC

"Jolene asked this question last week when she firstpassed an old paper and found that the Expr_patterns she flagged were already curated." JC 
"Sure, please scp a .ace dump with the Expr_pattern and Papers into the
 acedb account on tazendra and I'll mark those as ``yes (the papers
 themselves won't show as already being curated, but will have that
 data like the old system worked with textpresso transgene)" JC
"Would it be possible to do this for all other data types that have been curated from papers that have not gone through first-pass? 
There are a number of papers that were curated for phenotypes but were not first-passed." KY
 "Yes.  Likewise generate a .ace file with the data you want, scp it to
 the acedb account on tazendra, and let me know which postgres tables
 to populate with that data." JC


- I thought on the checkout section they showed as ``RNAi only or something like that. -- Juancarlos


Automating data type flagging

COMPLEX constructions

  • Gene Regulation

Basically, Arun got training sets from Andrei's first-pass flagged papers. Two training sets include: positive and negative flagged gene-regulation papers. Each set contains about a thousand papers. Then machine will produce index matrices (SVM, support vector machine, more specifically) according to two training sets. Arun can use some representative words/phrases appearing frequently in gene-regulation to fine tune his SVM. A new whole set of papers will be tested and we will see how this works.

  • Expression

SIMPLE patterns

  • RNAi via pattern recognition
  • Antibody via pattern recognition
  • Transgene

Pipeline

  • Pattern Design - Work with Juancarlos and or Michael
  • Pattern is run on PDFs loaded into postgres.
  • Curator gets list of papers that were not picked up by Andrei during first pass to evaluate possible outcomes
    • are they false positives - refine pattern
    • were they missed by Andrei - keep pattern
    • papers are not true articles and are 'no curatable' - omit from evaluation
  • Once refined, implement auto population of first pass form for curation
  • Pattern is run on PDFs loaded into postgres daily.
  • Scripts populate first pass form with results from pattern searches.

FP curator comments for St.Louis and Sanger structure correction data type

kjy 19:43, 9 February 2009 (EST)