diff options
author | Frederick Muriuki Muriithi | 2024-09-27 11:43:48 -0500 |
---|---|---|
committer | Frederick Muriuki Muriithi | 2024-09-27 11:43:48 -0500 |
commit | f04e340323d0e289aaa379ab1a8d82033da73e62 (patch) | |
tree | be44d5f39b9dfbc4d78cf2f296bc5f6ec8c2830d /uploader/phenotypes | |
parent | 0729e38e2f5bbc5ab23153adfed3d35ee59dc3d5 (diff) | |
download | gn-uploader-f04e340323d0e289aaa379ab1a8d82033da73e62.tar.gz |
Show some details for a phenotype dataset.
Diffstat (limited to 'uploader/phenotypes')
-rw-r--r-- | uploader/phenotypes/models.py | 48 | ||||
-rw-r--r-- | uploader/phenotypes/views.py | 47 |
2 files changed, 85 insertions, 10 deletions
diff --git a/uploader/phenotypes/models.py b/uploader/phenotypes/models.py index 4ef674f..c3b6dfb 100644 --- a/uploader/phenotypes/models.py +++ b/uploader/phenotypes/models.py @@ -22,6 +22,54 @@ def datasets_by_population( return tuple(dict(row) for row in cursor.fetchall()) +def dataset_by_id(conn: mdb.Connection, + species_id: int, + population_id: int, + dataset_id: int) -> dict: + """Fetch dataset details by identifier""" + with conn.cursor(cursorclass=DictCursor) as cursor: + cursor.execute( + "SELECT s.SpeciesId, pf.* FROM Species AS s " + "INNER JOIN InbredSet AS iset ON s.Id=iset.SpeciesId " + "INNER JOIN PublishFreeze AS pf ON iset.Id=pf.InbredSetId " + "WHERE s.Id=%s AND iset.Id=%s AND pf.Id=%s", + (species_id, population_id, dataset_id)) + return dict(cursor.fetchone()) + + +def phenotypes_count(conn: mdb.Connection, + population_id: int, + dataset_id: int) -> int: + """Count the number of phenotypes in the dataset.""" + with conn.cursor(cursorclass=DictCursor) as cursor: + cursor.execute( + "SELECT COUNT(*) AS total_phenos FROM Phenotype AS pheno " + "INNER JOIN PublishXRef AS pxr ON pheno.Id=pxr.PhenotypeId " + "INNER JOIN PublishFreeze AS pf ON pxr.InbredSetId=pf.InbredSetId " + "WHERE pxr.InbredSetId=%s AND pf.Id=%s", + (population_id, dataset_id)) + return int(cursor.fetchone()["total_phenos"]) + + +def dataset_phenotypes(conn: mdb.Connection, + population_id: int, + dataset_id: int, + offset: int = 0, + limit: Optional[int] = None) -> tuple[dict, ...]: + """Fetch the actual phenotypes.""" + _query = ( + "SELECT pheno.*, pxr.Id, ist.InbredSetCode FROM Phenotype AS pheno " + "INNER JOIN PublishXRef AS pxr ON pheno.Id=pxr.PhenotypeId " + "INNER JOIN PublishFreeze AS pf ON pxr.InbredSetId=pf.InbredSetId " + "INNER JOIN InbredSet AS ist ON pf.InbredSetId=ist.Id " + "WHERE pxr.InbredSetId=%s AND pf.Id=%s") + ( + f" LIMIT {limit} OFFSET {offset}" if bool(limit) else "") + with conn.cursor(cursorclass=DictCursor) as cursor: + cursor.execute(_query, (population_id, dataset_id)) + debug_query(cursor) + return tuple(dict(row) for row in cursor.fetchall()) + + def phenotypes_data(conn: mdb.Connection, population_id: int, dataset_id: int, diff --git a/uploader/phenotypes/views.py b/uploader/phenotypes/views.py index 88cb89c..4565844 100644 --- a/uploader/phenotypes/views.py +++ b/uploader/phenotypes/views.py @@ -1,20 +1,24 @@ """Views handling ('classical') phenotypes.""" -from flask import (request, +from flask import (flash, + request, url_for, redirect, Blueprint, render_template, current_app as app) -from uploader.datautils import order_by_family from uploader.authorisation import require_login from uploader.db_utils import database_connection from uploader.species.models import all_species, species_by_id from uploader.request_checks import with_species, with_population +from uploader.datautils import safe_int, order_by_family, enumerate_sequence from uploader.population.models import (populations_by_species, population_by_species_and_id) -from .models import datasets_by_population, phenotypes_data +from .models import (dataset_by_id, + phenotypes_count, + dataset_phenotypes, + datasets_by_population) phenotypesbp = Blueprint("phenotypes", __name__) @@ -40,7 +44,7 @@ def index(): methods=["GET"]) @require_login @with_species(redirect_uri="species.populations.phenotypes.index") -def select_population(species: dict, **kwargs): +def select_population(species: dict, **kwargs):# pylint: disable=[unused-argument] """Select the population for your phenotypes.""" with database_connection(app.config["SQL_URI"]) as conn: if not bool(request.args.get("population_id")): @@ -58,7 +62,7 @@ def select_population(species: dict, **kwargs): flash("No such population found!", "alert-danger") return redirect(url_for( "species.populations.phenotypes.select_population", - species_id=species_id)) + species_id=species["SpeciesId"])) return redirect(url_for("species.populations.phenotypes.list_datasets", species_id=species["SpeciesId"], @@ -72,7 +76,7 @@ def select_population(species: dict, **kwargs): @require_login @with_population(species_redirect_uri="species.populations.phenotypes.index", redirect_uri="species.populations.phenotypes.select_population") -def list_datasets(species: int, population: int, **kwargs): +def list_datasets(species: dict, population: dict, **kwargs):# pylint: disable=[unused-argument] """List available phenotype datasets.""" with database_connection(app.config["SQL_URI"]) as conn: return render_template("phenotypes/list-datasets.html", @@ -92,9 +96,32 @@ def list_datasets(species: int, population: int, **kwargs): @require_login @with_population(species_redirect_uri="species.populations.phenotypes.index", redirect_uri="species.populations.phenotypes.select_population") -def view_dataset(species: int, population: int, dataset_id: int, **kwargs): +def view_dataset(# pylint: disable=[unused-argument] + species: dict, population: dict, dataset_id: int, **kwargs): """View a specific dataset""" with database_connection(app.config["SQL_URI"]) as conn: - from flask import jsonify - return jsonify(phenotypes_data( - conn, population["Id"], dataset_id, offset=0, limit=20)) + dataset = dataset_by_id( + conn, species["SpeciesId"], population["Id"], dataset_id) + if not bool(dataset): + flash("Could not find such a phenotype dataset!", "alert-danger") + return redirect(url_for( + "species.populations.phenotypes.list_datasets", + species_id=species["SpeciesId"], + population_id=population["Id"])) + + start_at = max(safe_int(request.args.get("start_at") or 0), 0) + count = int(request.args.get("count") or 20) + return render_template("phenotypes/view-dataset.html", + species=species, + population=population, + dataset={"Id": dataset_id}, + phenotype_count=phenotypes_count( + conn, population["Id"], dataset_id), + phenotypes=enumerate_sequence( + dataset_phenotypes(conn, + population["Id"], + dataset_id, + offset=start_at, + limit=count), + start=start_at+1), + activelink="view-dataset") |