update modeling to reflect qualifier changes for chemical -> gene associations #512

andrewsu · 2022-10-04T21:15:16Z

This is a new issue to track our team's implementation of this issue on qualifiers for chemical-gene associations: NCATSTranslator/ReasonerAPI#348

From that issue:

The schema changes are here: https://github.com/NCATSTranslator/ReasonerAPI/pull/330/files
and there are some examples here: https://github.com/NCATSTranslator/ReasonerAPI/tree/master/examples/Message

According to the progress tracker, the target completion date for this issue is 2022-10-25

related/broader issue on constraints: #482

andrewsu · 2022-10-05T17:04:02Z

Steps

update SmartAPI annotations (Target completion as of 10/11 is 11/1)
- Minimal solution: find/replace examples of "targets", "upregulates", etc.
- TODO: @colleenXu to find/create a single example of an API edge that needs to be changed; check a modified version into a branch so we can test using API overrides
make sure BTE passes qualifiers (including qualified predicates, subject/object qualifiers) through in TRAPI results (Target completion as of 10/11 is 11/15)
- need to make sure we handle edge reversal/inversion to match query direction
BTE respects qualifiers in queries (Target completion as of 10/11 is 11/15)
- ~~check on reverse operations (if we have A - [upregulates] -> B, need to be able to query B - [upregulated_by] -> A)~~ (SmartAPI annotations are explicitly declared both ways)
- confirm whether Biolink 3.0 contains a hierarchy of qualifier_values (e.g., localization -> [tissue_localization, subcellular_localization])
  - EDIT: the existence of a hierarchy in the qualifier values is confirmed. For example, see https://github.com/biolink/biolink-model/blob/master/biolink-model.yaml#L11682 for gene_or_gene_product_or_chemical_entity_aspect_enums that show, for example, abundance has subclasses degradation, synthesis, expression, etc.
  - AS: support of hierarchy doesn't need to be in first implementation

andrewsu · 2022-10-07T15:21:01Z

Additional notes from my conversation with Sierra and Matt B.:

New predicate hierarchy with deprecated predicates removed can be viewed at http://tree-viz-biolink.herokuapp.com/predicates/deprecated-predicates
Current scope limited to associations related to chemicals (chemical gene interaction association and chemical affects gene association) to assess technical feasibility
The intent of the revised model is to express two statements in a single record. For example, consider the example at https://github.com/NCATSTranslator/ReasonerAPI/blob/master/examples/Message/examples_and_rules.md#object-qualifiers

subject: Bisphenol A
predicate: affects 
qualified_predicate: causes
object: ESR1
object_aspect_qualifier: degradation
object_direction_qualifier: decreased

One interpretation of this record is the simple SPO triple -- Bisphenol A affects ESR1 -- and one is the fully qualified triple -- Bisphenol A causes decreased degradation of ESR1

Full documentation document at https://docs.google.com/document/d/14UwR4g9ckGvX2NhFHWo5XOp1muCFlXUU16pvgTGiiUs/edit#heading=h.cxntx0nr2fjg

colleenXu · 2022-10-10T19:18:44Z

Helpful links:

Migration guide for Biolink 3.0 https://github.com/biolink/biolink-model/blob/master/Migration_3.0_Guide.md
Predicate transformation table: https://github.com/biolink/biolink-model/blob/master/images/predicate_transformation.pdf
A TRAPI example https://github.com/biolink/biolink-model/blob/master/guidelines/TRAPI-examples.md. More examples in the ReasonerAPI repo https://github.com/NCATSTranslator/ReasonerAPI/blob/master/examples/Message/examples_and_rules.md and in this repo https://github.com/NCATSTranslator/Evidence-Provenance-Confidence-Working-Group/blob/master/qualifier_example_data_and_queries.md

Update on Step 1

Note that the format (aka case) of qualifier values is still unclear. Internal Slack link.

Noting that qualifier-refactor discussions haven't happened with Multiomics Provider for their APIs:

Multiomics BigGIM-DrugResponse KP API
Multiomics ClinicalTrials KP
Multiomics EHR Risk KP API
Multiomics Wellness KP API

Their APIs didn't come up as I was doing my find-replace for all the deprecated predicates.

And there's discussion with Text-Mining Provider + Translator Data Modeling group on what predicates/qualifiers to use for "chemical-regulates-gene" relationships (internal Slack link). Depending on how the conversation goes, we may want to revise those operations for Text-Mining's API and other APIs that have "regulates" relationships

colleenXu · 2022-11-04T05:21:01Z

News:

Translator data modeling group confirmed that "regulates" should only be used in specific cases
- many gene-gene relationships will count. Many chemical-gene relationships will be "affects" instead.
- Quoting Mike Bada: "regulators as biologically evolved things, particularly (endogenous) genes/gene products, as we've defined regulation as biologically evolved things affecting other things for their proper functioning"
- Internal Slack post links here, here
Translator data modeling group confirmed (DM with Sierra Moxon) that qualifier-values should be snake_case (and all "enum" values)
Informed Multiomics + Text-Mining Providers of the qualifier-refactor situation, how BioThings-APIs + BTE-as-TRAPI-wrapper works. Planning to leave these teams to update their x-bte annotation for the qualifiers refactor
Text-Mining Provider has their own branch for doing qualifier-refactor work. Jackson has connected the dev instance to this branch's yaml
- They've opened a PR for an update ("regulates" vs "affects") and I've provided initial feedback

X-bte annotation updates:

semmeddb was updated: replace "regulates"-predicates + qualifiers with "affects"-predicates + qualifiers.
DGIdb and semmeddb have been updated to use snake_case for qualifier-values

Update 2022-11-10: modified the code to generate semmeddb operations with the qualifier/biolink3 mappings we're currently doing.

colleenXu · 2022-11-05T05:36:14Z

@tokebe Added a commit to query-handler (add-qualifiers branch) to adjust templates for the predicate changes biothings/bte_trapi_query_graph_handler@3a0ca1a. I haven't tested this change yet.

I'm also not sure about adding qualifier-constraints

tokebe · 2022-11-07T14:16:51Z

Once I have the implementation, I see no reason that you couldn't use qualifier-constraints. You can add them now if you want, since they won't break anything.

tokebe · 2022-11-07T18:40:08Z

@colleenXu I've pushed an experimental implementation of qualifier constrain pre-filtering. Below is the rundown of the behavior:

The following pre-filtering is only applied if an operation doesn't have the bte-trapi tag, which is added by BTE when TRAPI ingesting operations. If the operation has the bte-trapi tag, it is kept (because the qualifier constraints will be passed in as part of the TRAPI subquery)
For each operation, if at least one of the qualifier constraint sets has all of its qualifiers present/matching in the operation, the operation is kept.
If the given qualifier constraint set all have matches in the operation, but the operation has additional qualifiers, the operation is kept
All operations 'kept' are subject to subsequent filtering passes, such as predicate, input/output IDs, etc.

If this behavior looks good, could you run a few tests to make sure the implementation is behaving as expected (I've run some preliminary tests and it looks fine)? I'll be working on ensuring everything's set for adding in biolink 3.

colleenXu · 2022-11-07T19:36:10Z

@tokebe I found this set of implementation rules on qualifier_constraints https://github.com/NCATSTranslator/ReasonerAPI/blob/v1.3.0/ImplementationRules.md#qedgequalifier_constraints

tokebe · 2022-11-07T19:43:34Z

I'm aware of this set of rules. My implementation follows these; however these rules are very broad and do not cover the specifics of qualifier-constraint matching for a KP, which we must follow for our non-TRAPI KPs.

These implementation rules only specify "is compatible with", which I've interpreted both to the best of my ability and with my understanding from previous conversations. I'm asking if you could look over the rundown I wrote and confirm that that specific set of cases looks fine.

Biolink-3 implementation seems relatively straightforward with only some test fixes. @andrewsu does it make sense to leave hierarchy traversal for qualifiers (and qualifier-constraint pre-filtering) for another issue?

andrewsu · 2022-11-07T19:59:37Z

@andrewsu does it make sense to leave hierarchy traversal for qualifiers (and qualifier-constraint pre-filtering) for another issue?

Yes, I think that makes sense. This issue is already getting a little bit unwieldy, so breaking things down to a series of more focused issues seems like a good plan...

tokebe · 2022-11-07T22:06:47Z

@colleenXu I'm rolling the biolink 3.0.3 rollout into the qualifiers issue given that they're both very related and the biolink 3 rollout won't require a lot of work.

What's left (some may be put in separate issues):

additional testing for qualifier constraints
write tests for qualifier support
implement biolink 3 qualifier hierarchy traversal for qualifier constraints -> new issue

Do we want to somehow add qualifiers to our /meta_knowledge_graph endpoint? There hasn't been any guidance on this from Translator that I'm aware of but I imagine it's something that will be expected eventually...

colleenXu · 2022-11-07T22:58:04Z

@tokebe I think it makes sense to roll the #514 and this qualifier issue into the same set of branches/PRs.

I plan to write up how I manually test qualifier-constraints - which may help for writing tests

I think the qualifier-value hierarchy traversal would definitely be a separate issue.

I don't think we need to add qualifiers to our /meta_knowledge_graph endpoint for this issue because we don't know the format yet (from TRAPI) and we're not leading the effort there...

colleenXu · 2022-11-09T05:56:50Z

Feedback Part 1

Update 2022-11-15: with this code fix, the behavior with 1 set of 1 constraint looks mostly fixed. There is still an issue with qualified_predicate (described in the later comment).

Qualifier constraints doesn't seem to work: compare the first and second queries below

Testing setup

add-qualifier branches for

biolink-model

api-response-transform

bte-trapi (probably optional - I remember making this branch)

call-apis

query_graph_handler

smartapi-kg

checkout bte-trapi's dev branch's src/config/smartapi_overrides.json file. It has all the API overrides to use biolink3/qualifier-updated SmartAPI annotations

run API_OVERRIDE=true npm run smartapi_sync to get all the API overrides

First set of testing

I used DGIdb's x-bte annotation that has qualifiers as a source. And I ran my queries restricted to using DGIdb only (link has v1/smartapi/e3edd325c76f2992a111b43a907a4870/query).
First: as a baseline - no qualifier_constraints
query
{
    "message": {
        "query_graph": {
            "nodes": {
                "n0": {
                    "ids":["NCBIGene:9132"],
                    "categories":["biolink:Gene"]
                },
                "n1": {
                    "categories":["biolink:SmallMolecule"]
               }
            },
            "edges": {
                "e1": {
                    "subject": "n0",
                    "object": "n1"
                }
            }
        }
    }
}
KCNQ4.txt

Currently, I get 9 results with 11 Edges:

1 result Celecoxib has only 1 Edge that doesn't have qualifiers (predicate is interacts_with)

5 results have Edges with qualifiers decreased activity molecular_channel_blockage: Guanidine hydrochloride, Nerispirdine, 4-Aminopyridine, TEDISAMIL, Bepridil

1 result Linopirdine has Edge with qualifiers decreased activity_or_abundance inhibition

1 result Flindokalner has Edge with qualifiers increased activity activation

1 result Retigabine has 3 Edges (one with interacts_with, one with the decreased activity molecular_channel_blockage qualifiers and one with the increased activity activation).
Second: 1 set of 1 constraint
query
{
    "message": {
        "query_graph": {
            "nodes": {
                "n0": {
                    "ids":["NCBIGene:9132"],
                    "categories":["biolink:Gene"]
                },
                "n1": {
                    "categories":["biolink:SmallMolecule"]
               }
            },
            "edges": {
                "e1": {
                    "subject": "n0",
                    "object": "n1",
                    "qualifier_constraints": [
                        {
                            "qualifier_set": [
                                {
                                    "qualifier_type_id": "biolink:subject_direction_qualifier",
                                    "qualifier_value": "increased"
                                }
                            ]
                        }
                    ]
                }
            }
        }
    }
}
Compared to the first query, I'd expect to get only 2 of the results, Flindokalner and Retigabine, since both had edges with the qualifiers increased activity activation.

Instead, I'm getting 8 results (all the results with "affected_by" as the predicate).

colleenXu · 2022-11-11T05:36:39Z

Feedback Part 2

Update 2022-11-15: with this code fix, the behavior looks fixed. Yay!

The "reversing" of qualifiers to match the QEdge direction isn't working as-expected. I'm running my queries restricted to using DGIdb only (link has v1/smartapi/e3edd325c76f2992a111b43a907a4870/query).

If I run a query like Gene -(affected_by)-> SmallMolecule X, I expect BTE to "reverse" the edge for execution (so it can start with the SmallMolecule X ID).

BTE will then run the sub-query SmallMolecule X -(affects)-> Gene. I know DGIdb's x-bte annotation will have the "causes" qualified_predicate and "object_*" qualifiers in the sub-query responses.
query to test correct reversal
{
    "message": {
        "query_graph": {
            "nodes": {
                "n0": {
                    "ids":["CHEMBL.COMPOUND:CHEMBL730"],
                    "categories":["biolink:SmallMolecule"],
                    "name": "NITROGLYCERIN"
                },
                "n1": {
                    "categories":["biolink:Gene"]
               }
            },
            "edges": {
                "e1": {
                    "subject": "n1",
                    "object": "n0",
                    "predicates": ["biolink:affected_by"]
                }
            }
        }
    }
}
Afterwards, BTE should reverse the predicate (affects) and the qualifiers for the Edges, so the Edges in the response have the predicate "affected_by" (same direction as QEdge), "caused_by" as the qualified_predicate, and "subject_*" qualifiers.

However, the current code doesn't correctly reverse the qualifiers. It does correctly reverse the predicate.

Here's the response I get to the query:
example_reverse.txt

colleenXu · 2022-11-11T21:59:28Z

Feedback Part 3

Update 2022-11-15: with this code fix, (2) seems to be working. (1) isn't fully working - there are still issues with qualified_predicate (listed in a later comment)

I deleted a post and edited headings for the two posts above, to make these main concerns clearer. I'm also organizing the various things that have come up, in this post.

Here are some smaller features we want after recent discussions:

when qualifier_type_id is qualified_predicate, the qualifier_value should have a biolink-prefix ("biolink:causes" or "biolink:caused_by"). This is for qualifiers in Edges and qualifier_constraints in QEdges.

@tokebe 2022-11-11 said we can do this in code, so we don't have to add the biolink-prefix to x-bte annotation (similar to how we treat predicates + node categories)

query-validation:

UPDATE 2022-11-14 6pm PT, after discussion with Andrew: this is optional. We don't have to worry about query-validation because we should expect queries to the guidelines set by Data-Modeling/TRAPI groups.

"within 1 qualifier_set, each object must have a unique value for the qualifier_type_id field". This is what I think point 1i of this doc means.

If there are objects in a set with the same qualifier_type_id, I guess we'd want BTE to not execute further and return an informative error (there are two qualifier_constraints in the same set that have the same qualifier_type_id; please modify the query so each qualifier_type_id in a set is unique)

UPDATE 2022-11-14 6pm PT, after discussion with Andrew:

Don't worry about TRAPI KPs and how they'd handle "multiple curies" + "qualifier constraints":

TRAPI KPs should fail gracefully if they can't handle this, and BTE should be able to handle failures from sub-queries to TRAPI KPs

We're also asking for clarification...
According to the TRAPI 1.3.0 spec, qualifier_constraints are not supposed to be on a QEdge when the QNodes of that specific QEdge have multiple curies. BTE's implementation will likely handle multiple curies + qualifier_constraints without issues...but this may have consequences for how we do TRAPI-API sub-querying with qualifier_constraints:

Do we have to send IDs one-at-a-time when that hop has qualifier_constraints?

Do we have to send the query without qualifier_constraints, and then do post-filtering with the qualifier_constraints?

Some notes:

Right now, qualifier_value values that are enums from biolink-model DO NOT need the biolink-prefix
Theoretically you could query with qualifier_constraints and without specifying a predicate. I haven't seen any requirement forbidding this. BTE's implementation will likely be able to handle this.
According to the TRAPI 1.3.0 spec, qualifier_constraints are not supposed to be on a QEdge in the following situations. However, @tokebe and I think BTE can actually handle these. We don't see "implementation rules" on how to respond to these situations (like stopping execution / error messages).
- that specific QEdge has multiple predicates specified.
  - This isn't an issue for TRAPI-API sub-querying because we only query 1 operation at a time (which has only 1 predicate)
- the QNodes of that specific QEdge have multiple categories specified.
  - This situation comes up regularly due to the difference between the categories returned by NodeNormalizer and the categories specified on the QNodes (by the user).
  - And this isn't an issue for TRAPI-API sub-querying because we only query 1 operation at a time (which has only 1 category for subject and 1 for object)

Future issues from this "requirements" doc:

qualifier_value's value can be an enum from the biolink-model, and we're expected to "expand" qualifier_constraint terms using the enum hierarchy (for example, querying for activity_or_abundance should bring back Edges that have expression)
qualifier_value's value can also be an ontology-term (curie) and we're expected to "expand" qualifier_constraint terms using the ontology...
- It's not clear if only specific ontologies can be the value, for specific qualifier_type_ids...

tokebe · 2022-11-15T18:21:59Z

@colleenXu I've pushed fixes to each respective package.

Fixed constraints. For some reason one change required to pass constraints into the filter criteria didn't get committed
Fixed reversing. Reversed records weren't building using the newly reversed qualifiers.
Fixed biolink: prefix missing
Added validation for duplicate qualifier types within a set.

colleenXu · 2022-11-16T05:40:50Z

Feedback round 2

EDIT: done retesting. From my testing POV, there were only 2 small things with qualified_predicate, which are probably related and easily fixed. Once these are addressed, this issue (Steps 1-3) may be complete...

Qualified predicates

Note: Testing setup is the same as above (aka using only DGIdb at the moment).)

Qualified_predicate in a qualifier_constraint is not working when the value (predicate) includes the biolink-prefix (see example query below), which is the format we're supposed to follow. Currently, BTE finds no matching metaKG edges.
I notice that if I provide the qualified predicate without the biolink-prefix, it works.
ATP_1.txt

correct qualified_predicate-format example

{
    "message": {
        "query_graph": {
            "nodes": {
                "n0": {
                    "ids":["CHEMBL.COMPOUND:CHEMBL14249"],
                    "categories":["biolink:SmallMolecule"],
                    "name": "ATP"
                },
                "n1": {
                    "categories":["biolink:Gene"]
               }
            },
            "edges": {
                "e1": {
                    "subject": "n0",
                    "object": "n1",
                    "qualifier_constraints": [
                        {
                            "qualifier_set": [
                                {
                                    "qualifier_type_id": "biolink:qualified_predicate",
                                    "qualifier_value": "biolink:causes"
                                }
                            ]
                        }
                    ]
                }
            }
        }
    }
}

BTE isn't finding metaKG edges when I test the qualified_predicate constraint with a QEdge that'll need reversal, even when the predicate biolink-prefix is removed:

"reversed QEdge" version of the query above

Look carefully at what the QEdge subject and object are (Gene -> ATP) and the qualified predicate.

{
    "message": {
        "query_graph": {
            "nodes": {
                "n0": {
                    "ids":["CHEMBL.COMPOUND:CHEMBL14249"],
                    "categories":["biolink:SmallMolecule"],
                    "name": "ATP"
                },
                "n1": {
                    "categories":["biolink:Gene"]
               }
            },
            "edges": {
                "e1": {
                    "subject": "n1",
                    "object": "n0",
                    "qualifier_constraints": [
                        {
                            "qualifier_set": [
                                {
                                    "qualifier_type_id": "biolink:qualified_predicate",
                                    "qualifier_value": "biolink:caused_by"
                                }
                            ]
                        }
                    ]
                }
            }
        }
    }
}

Like the query above, there should be 15 results when the metaKG edges are found.

Correction #1: Having two constraints in the same set is working.

It's expected to work as "AND" (both conditions must be met on one metaKG edge).

compare query with 1 constraint vs 2 constraints

First, the query w/ 1 constraint has 3 results. One of the results, TRPM4, has two edges - one for "activity" and one for "activity_or_abundance".
ATP_decreased.txt

1 constraint

{
    "message": {
        "query_graph": {
            "nodes": {
                "n0": {
                    "ids":["CHEMBL.COMPOUND:CHEMBL14249"],
                    "categories":["biolink:SmallMolecule"],
                    "name": "ATP"
                },
                "n1": {
                    "categories":["biolink:Gene"]
               }
            },
            "edges": {
                "e1": {
                    "subject": "n0",
                    "object": "n1",
                    "qualifier_constraints": [
                        {
                            "qualifier_set": [
                                {
                                    "qualifier_type_id": "biolink:object_direction_qualifier",
                                    "qualifier_value": "decreased"
                                }
                            ]
                        }
                    ]
                }
            }
        }
    }
}

Compare it to the response with 2 constraints (AND). While it also has 3 results, TRPM4, only has 1 edge - the one for "activity" to match the 2nd constraint.
ATP_decreased_activity.txt

1 constraint

{
    "message": {
        "query_graph": {
            "nodes": {
                "n0": {
                    "ids":["CHEMBL.COMPOUND:CHEMBL14249"],
                    "categories":["biolink:SmallMolecule"],
                    "name": "ATP"
                },
                "n1": {
                    "categories":["biolink:Gene"]
               }
            },
            "edges": {
                "e1": {
                    "subject": "n0",
                    "object": "n1",
                    "qualifier_constraints": [
                        {
                            "qualifier_set": [
                                {
                                    "qualifier_type_id": "biolink:object_direction_qualifier",
                                    "qualifier_value": "decreased"
                                },
                                {
                                    "qualifier_type_id": "biolink:object_aspect_qualifier",
                                    "qualifier_value": "activity"
                                }
                            ]
                        }
                    ]
                }
            }
        }
    }
}

Correction #2: Having 2 sets of constraints (1 in each) is working.

It's expected to work as "OR" (at least one condition must be met on one metaKG edge).

example with 2 sets of constraints

{
    "message": {
        "query_graph": {
            "nodes": {
                "n0": {
                    "ids":["CHEMBL.COMPOUND:CHEMBL14249"],
                    "categories":["biolink:SmallMolecule"],
                    "name": "ATP"
                },
                "n1": {
                    "categories":["biolink:Gene"]
               }
            },
            "edges": {
                "e1": {
                    "subject": "n0",
                    "object": "n1",
                    "qualifier_constraints": [
                        {
                            "qualifier_set": [
                                {
                                    "qualifier_type_id": "biolink:object_direction_qualifier",
                                    "qualifier_value": "increased"
                                }
                            ]
                        },
                        {
                            "qualifier_set": [
                                {
                                    "qualifier_type_id": "biolink:object_direction_qualifier",
                                    "qualifier_value": "decreased"
                                }
                            ]
                        }
                    ]
                }
            }
        }
    }
}

This will return 15 results - some with increased and some with decreased qualifiers.

Useful basic queries for qualifier-constraints:

KCNQ4 NCBIGene:9132 -(affected_by)-> chem
IGF1R NCBIGene:3480 -(affected_by)-> chem
RYR2 NCBIGene:6262 -(affected_by)-> chem
ATP CHEMBL14249 -(affects)-> Gene
Quetiapine CHEMBL716 -(affects)-> Gene

Each of these has at least 3 kinds of qualifier-combos (increased activity, decreased activity, decreased activity_or_abundance)

tokebe · 2022-12-07T20:13:21Z

@colleenXu I pushed fixes for these, currently working on fixing any broken tests (most failing tests on the query_graph_handler are due to changes causing qEdge hashes to change).

Edit: there are still a few more minor fixes to be made, will update when everything's ready for another round of testing.

colleenXu · 2022-12-09T04:31:55Z

Tested the latest code for qualified_predicate handling. Now it works, whether the value has the biolink prefix or not.

Did some quick testing of other queries with qualified predicates -> these still work as well.

tokebe · 2022-12-09T14:25:45Z

Alright, I think I can leave it to a future validation issue for not supporting non-prefixed values, if we care about that.

colleenXu · 2022-12-09T20:11:15Z

Don't think we care about it :P

tokebe · 2023-03-28T20:33:06Z

Deployed to prod 🚀

tokebe assigned tokebe and colleenXu Oct 10, 2022

colleenXu mentioned this issue Oct 12, 2022

Respect canonical predicate direction #476

Closed

tokebe mentioned this issue Dec 7, 2022

Support qualifiers in creative query template matching #534

Closed

colleenXu mentioned this issue Mar 15, 2023

investigate TRAPI validation errors showing in ARAX UI #587

Closed

tokebe closed this as completed Mar 28, 2023

colleenXu mentioned this issue Apr 5, 2023

phase 3: add qualifier info from TRAPI KPs to general-ARA /meta_knowledge_graph response #609

Closed

update modeling to reflect qualifier changes for chemical -> gene associations #512

update modeling to reflect qualifier changes for chemical -> gene associations #512

Comments

andrewsu commented Oct 4, 2022

andrewsu commented Oct 5, 2022 • edited Loading

andrewsu commented Oct 7, 2022 • edited Loading

colleenXu commented Oct 10, 2022

colleenXu commented Oct 10, 2022 • edited Loading

colleenXu commented Oct 11, 2022

tokebe commented Oct 14, 2022 • edited Loading

andrewsu commented Oct 17, 2022

colleenXu commented Oct 19, 2022

colleenXu commented Oct 19, 2022

tokebe commented Oct 20, 2022

tokebe commented Oct 20, 2022

colleenXu commented Oct 20, 2022

colleenXu commented Oct 27, 2022 • edited Loading

colleenXu commented Nov 2, 2022 • edited Loading

Update on Step 1

colleenXu commented Nov 4, 2022 • edited Loading

colleenXu commented Nov 5, 2022

tokebe commented Nov 7, 2022

tokebe commented Nov 7, 2022 • edited Loading

colleenXu commented Nov 7, 2022

tokebe commented Nov 7, 2022

andrewsu commented Nov 7, 2022

tokebe commented Nov 7, 2022 • edited Loading

colleenXu commented Nov 7, 2022

colleenXu commented Nov 9, 2022 • edited Loading

Feedback Part 1

Testing setup

First set of testing

colleenXu commented Nov 11, 2022 • edited Loading

Feedback Part 2

colleenXu commented Nov 11, 2022 • edited Loading

Feedback Part 3

tokebe commented Nov 15, 2022

colleenXu commented Nov 16, 2022 • edited Loading

Feedback round 2

Qualified predicates

Correction #1: Having two constraints in the same set is working.

Correction #2: Having 2 sets of constraints (1 in each) is working.

tokebe commented Dec 7, 2022 • edited Loading

colleenXu commented Dec 9, 2022 • edited Loading

tokebe commented Dec 9, 2022

colleenXu commented Dec 9, 2022

tokebe commented Mar 28, 2023

andrewsu commented Oct 5, 2022 •

edited

Loading

andrewsu commented Oct 7, 2022 •

edited

Loading

colleenXu commented Oct 10, 2022 •

edited

Loading

tokebe commented Oct 14, 2022 •

edited

Loading

colleenXu commented Oct 27, 2022 •

edited

Loading

colleenXu commented Nov 2, 2022 •

edited

Loading

colleenXu commented Nov 4, 2022 •

edited

Loading

tokebe commented Nov 7, 2022 •

edited

Loading

tokebe commented Nov 7, 2022 •

edited

Loading

colleenXu commented Nov 9, 2022 •

edited

Loading

colleenXu commented Nov 11, 2022 •

edited

Loading

colleenXu commented Nov 11, 2022 •

edited

Loading

colleenXu commented Nov 16, 2022 •

edited

Loading

tokebe commented Dec 7, 2022 •

edited

Loading

colleenXu commented Dec 9, 2022 •

edited

Loading