Opened 10 years ago
Closed 9 years ago
#1941 closed task (fixed)
Store experimental factor values as part experiments
Reported by: | Nicklas Nordborg | Owned by: | Nicklas Nordborg |
---|---|---|---|
Priority: | major | Milestone: | BASE 3.6 |
Component: | core | Version: | |
Keywords: | Cc: |
Description
The current implementation of experimental factor values rely on annotations (primary or inherited) on the raw bioassays. There are some drawbacks with this:
- Bioassays that are part of multiple experiments need to inherit annotations for experimental factors for both experiments. This can impractical since it require WRITE permission on the raw bioassays and it becomes difficult to share data with users that you don't want to give that permission to.
- Changing the annotation values are immediately reflected in the experimental factors. This might be undesirable if the old values have been used in the analysis already.
So, we need a solution that can store experimental factor values as part of the experiment instead of as part of the raw bioassays.
- The values should be copies of the annotations that was used to create them.
- A link to the originating annotation should be stored, but it should not block changing or removing the original annotation.
- It should be possible to manually change the experimental factor values.
- It should be possible to display the current "synchronization" status and re-sync with the original annotation.
- More...?
Downstream tools that currently find experimental factors by loading annotations from the raw bioassays need to be updated.
Change History (43)
comment:1 by , 9 years ago
Owner: | changed from | to
---|---|
Status: | new → assigned |
comment:2 by , 9 years ago
Summary: | Store exerimental factor values as part experiments → Store experimental factor values as part experiments |
---|
comment:3 by , 9 years ago
(In [6908]) References #1941: Store experimental factor values as part experiments
Added RootRawBioAssayData
class. A database schema update is moving over existing raw bioassays to the new table and then drops the old table.
Some old mappings have been kept and this change doesn't introduce any incompatibility in the public API, but the new class is still only used for the link and not for annotations/experimental factors.
comment:4 by , 9 years ago
(In [6909]) References #1941: Store experimental factor values as part experiments
Introduced RootRawBioAssay
and use this item for inheriting experimental factors. There is still only inheritance by linking to the original value. Cloning of values remains to be implemented.
Downstream tools such as Experiment explorer, the plot tool, etc. have been updated to use the new items when searching for experimental factor values.
comment:5 by , 9 years ago
(In [6910]) References #1941: Store experimental factor values as part experiments
Removing support for inheriting annotation sets. Annotations must now be inherited one by one but utility methods exists that work on all current annotations in an annotation set. But it will no longer be possible to automatically inherit new annotations from an item by inheriting the full annotation set.
The update will transform all current inherited annotation sets to inherited annotations.
The snapshot cache version number was increased to 4 to make sure no old snaphots remains.
comment:6 by , 9 years ago
comment:7 by , 9 years ago
comment:8 by , 9 years ago
(In [6914]) References #1941: Store experimental factor values as part experiments
Experimental factor annotations for an experiment which are found as either primary or inherited annotations on the raw bioassays in the experiment are inherited to the root raw bioassay items. This means that existing experiments will continue to have access to the experimental factors as before.
comment:9 by , 9 years ago
comment:10 by , 9 years ago
comment:11 by , 9 years ago
(In [6917]) References #1941: Store experimental factor values as part experiments
Started to redesign the AnnotationData
class so that it can hold both primary/inherited/cloned annotations. The source
column is an enum indicating the source of the annotation 0=primary, 1=inherited, 2=cloned. All annotations should be associated with an annotation set and an annotation type. The primary and cloned annotation have values and possible a unit and last updated date. Inherited annotations and cloned annotations have a link to the original annotation.
If the original annotation is removed, the link is should be nullified (but values are kept) for cloned annotations, but removed by cascade for inherited annotations. This is not yet implemented.
The update script should move existing inherited annotations to the new schema, but much of the existing code in the core is currently not functional and has been commented out while waiting to be fixed.
comment:12 by , 9 years ago
comment:13 by , 9 years ago
comment:14 by , 9 years ago
(In [6920]) References #1941: Store experimental factor values as part experiments
Fixes some more places in the core API that used the old "InheritedAnnotations" table.
comment:15 by , 9 years ago
(In [6921]) References #1941: Store experimental factor values as part experiments
Started to implement support for cloning by adding a second boolean parameter to AnnotationSet.inheritAnnotation()
.
Existing code seems to not be affected much but behaves as if the annotation was inherited by linking.
The TestAnnotation
test code has some new test cases that are preliminary so far. Some things may need to change when implementing more realistic scenarios in the web client.
comment:16 by , 9 years ago
(In [6922]) References #1941: Store experimental factor values as part experiments
Lots of changes in AnnotationSnapshot
to make it possible to use with cloned annotations. Almost all existing methods in this class has been deprecated and replaced with new variants. The deprecate method should behave as before for primary and inherited annotations. Cloned annotations will get some information from the inherited annotation and some from the cloned information. This might confuse existing code that is not prepared to handle cloned annotations.
The "Annotations" tab in the web gui has been updated to display the correct information but it is not yet possible to edit or otherwise manage cloned annotations.
comment:17 by , 9 years ago
comment:18 by , 9 years ago
comment:19 by , 9 years ago
comment:20 by , 9 years ago
comment:21 by , 9 years ago
(In [6937]) References #1941: Store experimental factor values as part experiments
Removing the annotation on a parent item will nullify the link from cloned annotations instead of removing the clones.
Other changes are related to fixing various NullPointerExceptions and other problems resulting from having "inherited" annotations without a parent item.
But more work is needed on this since the cloned annotations now end up in a state that is almost impossible to manage via the gui. The "batch inherit annotations" can be used to remove or update cloned annotation.
comment:22 by , 9 years ago
(In [6938]) References #1941: Store experimental factor values as part experiments
Started with redesigning the annotations edit dialog. The idea is to include both primary and inherited/cloned annotations in the "Annotations" tab and manage all types of annotations there. The "Inherited annotations" tab will be removed.
So far, the new implementation will list all types of annotations. PRIMARY and CLONED annotations are editable, INHERITED annotations only display some information. Support for multi-valued annotations is not implemeted.
comment:23 by , 9 years ago
(In [6939]) References #1941: Store experimental factor values as part experiments
Updated "Item overview" functionality to become compatible with cloned annotations. There may still be some issues with suggested fixes related to inherited annotations since it is not yet clear how those will be managed.
comment:24 by , 9 years ago
comment:25 by , 9 years ago
comment:26 by , 9 years ago
comment:27 by , 9 years ago
comment:28 by , 9 years ago
comment:29 by , 9 years ago
comment:30 by , 9 years ago
comment:31 by , 9 years ago
comment:32 by , 9 years ago
(In [6958]) References #1941: Store experimental factor values as part experiments
Updated test program to set experimental factor values on root raw bioassays instead of raw bioassays.
Noted that the "Item overview validation" now complains about missing experimental factors, probably because it has not been updated and still check the raw bioassays and not the root raw bioassays.
comment:33 by , 9 years ago
(In [6959]) References #1941: Store experimental factor values as part experiments
The "Item overview" functionality now load "root raw bioassays" instead of "raw bioassays" when going up from the experiment level. This change is needed so that experimental factor values can be found. The loading then skips the actual raw bioassay item. This change causes lots of other loaders/validators to fail since the are expecting or looking for a raw bioassay. Those have been updated so that they can work with either a root raw bioassay or a raw bioassay.
comment:34 by , 9 years ago
comment:35 by , 9 years ago
comment:36 by , 9 years ago
comment:37 by , 9 years ago
comment:38 by , 9 years ago
comment:39 by , 9 years ago
comment:40 by , 9 years ago
comment:41 by , 9 years ago
comment:42 by , 9 years ago
(In [6976]) References #1941: Store experimental factor values as part experiments
When updating a MySQL installation it is not possible to drop indexes without also dropping foreign keys that use the same columns. After dropping the index, the foreign keys need to be re-created. In schema version 124 we need to drop the unique index on annotationset_id+annotationtype_id in the Annotations table since cloning annotations means that there can be multiple annotations for the same annotation type.
When updating a MySQL installation it is not possible to drop a NOT NULL constraint on a column. Instead the column definition must be modified as if a new column was created without a NOT NULL constraint. This requires that we know the data type of the column. Since the SQL to do this is so different from how PostgreSQL does this, we put this in the mysql-queries.xml file.
comment:43 by , 9 years ago
Resolution: | → fixed |
---|---|
Status: | assigned → closed |
The current idea is to replace the Experiment<->Raw bioassay link which is currently a simple many-to-many relation with a new item (
RootRawBioAssay?
). The item should beAnnotatable
and the annotation set should be used for storing the experimental factor values. This solves the first problem with inheriting annotations to the raw bioassay since they are now inherited to the root raw bioassay instead.Cloning annotations need additional support. We need to add rules so that the annotations targeted for other item types can be stored on the root raw bioassay. There is already similar functionality for plug-ins.
Additional functionality is also needed for the link back to the originating annotation. In the suggested design this link need to go from
Annotation
to the originatingAnnotation
. The drawback with this is that the link is lost if the original annotation is lost and then re-created. It might be better to store this as a "soft" link using type+id to the originating item instead.This change will introduce incompatibilities in the API and underlying database that affect down-stream analysis tools that use annotations or other information from the raw bioassay level.