aboutsummaryrefslogtreecommitdiff
path: root/bh20seqanalyzer/main.py
diff options
context:
space:
mode:
authorPjotr Prins2020-04-09 15:34:06 -0500
committerPjotr Prins2020-04-09 15:34:06 -0500
commit0670ac0644c1e7366952e254bdee2db62e673275 (patch)
treeaed056a1ca4208cf30993da3e96bbb4fb08dbe52 /bh20seqanalyzer/main.py
parent146cf2f5d1be9a5dd9d6cd65ce9c760853d014f8 (diff)
parentdbe094a150d6c969b3d69f112b3538e6a87a74a2 (diff)
downloadbh20-seq-resource-0670ac0644c1e7366952e254bdee2db62e673275.tar.gz
bh20-seq-resource-0670ac0644c1e7366952e254bdee2db62e673275.tar.lz
bh20-seq-resource-0670ac0644c1e7366952e254bdee2db62e673275.zip
Merge branch 'master' of github.com:arvados/bh20-seq-resource
Diffstat (limited to 'bh20seqanalyzer/main.py')
-rw-r--r--bh20seqanalyzer/main.py187
1 files changed, 166 insertions, 21 deletions
diff --git a/bh20seqanalyzer/main.py b/bh20seqanalyzer/main.py
index 23e58e9..1a8965b 100644
--- a/bh20seqanalyzer/main.py
+++ b/bh20seqanalyzer/main.py
@@ -1,29 +1,73 @@
import argparse
import arvados
+import arvados.collection
import time
import subprocess
import tempfile
import json
+import logging
+import ruamel.yaml
+from bh20sequploader.qc_metadata import qc_metadata
-def start_analysis(api, collection, analysis_project, workflow_uuid):
+logging.basicConfig(format="[%(asctime)s] %(levelname)s %(message)s", datefmt="%Y-%m-%d %H:%M:%S",
+ level=logging.INFO)
+logging.getLogger("googleapiclient.discovery").setLevel(logging.WARN)
+
+def validate_upload(api, collection, validated_project,
+ fastq_project, fastq_workflow_uuid):
+ col = arvados.collection.Collection(collection["uuid"])
+
+ # validate the collection here. Check metadata, etc.
+ valid = True
+
+ if "metadata.yaml" not in col:
+ logging.warn("Upload '%s' missing metadata.yaml", collection["name"])
+ valid = False
+ else:
+ metadata_content = ruamel.yaml.round_trip_load(col.open("metadata.yaml"))
+ #valid = qc_metadata(metadata_content) and valid
+ if not valid:
+ logging.warn("Failed metadata qc")
+
+ if valid:
+ if "sequence.fasta" not in col:
+ if "reads.fastq" in col:
+ start_fastq_to_fasta(api, collection, fastq_project, fastq_workflow_uuid)
+ return False
+ else:
+ valid = False
+ logging.warn("Upload '%s' missing sequence.fasta", collection["name"])
+
+ dup = api.collections().list(filters=[["owner_uuid", "=", validated_project],
+ ["portable_data_hash", "=", col.portable_data_hash()]]).execute()
+ if dup["items"]:
+ # This exact collection has been uploaded before.
+ valid = False
+ logging.warn("Upload '%s' is duplicate" % collection["name"])
+
+ if valid:
+ logging.info("Added '%s' to validated sequences" % collection["name"])
+ # Move it to the "validated" project to be included in the next analysis
+ api.collections().update(uuid=collection["uuid"], body={
+ "owner_uuid": validated_project,
+ "name": "%s (%s)" % (collection["name"], time.asctime(time.gmtime()))}).execute()
+ else:
+ # It is invalid, delete it.
+ logging.warn("Deleting '%s'" % collection["name"])
+ api.collections().delete(uuid=collection["uuid"]).execute()
+
+ return valid
+
+
+def run_workflow(api, parent_project, workflow_uuid, name, inputobj):
project = api.groups().create(body={
"group_class": "project",
- "name": "Analysis of %s" % collection["name"],
- "owner_uuid": analysis_project,
+ "name": name,
+ "owner_uuid": parent_project,
}, ensure_unique_name=True).execute()
with tempfile.NamedTemporaryFile() as tmp:
- inputobj = json.dumps({
- "sequence": {
- "class": "File",
- "location": "keep:%s/sequence.fasta" % collection["portable_data_hash"]
- },
- "metadata": {
- "class": "File",
- "location": "keep:%s/metadata.jsonld" % collection["portable_data_hash"]
- }
- }, indent=2)
- tmp.write(inputobj.encode('utf-8'))
+ tmp.write(json.dumps(inputobj, indent=2).encode('utf-8'))
tmp.flush()
cmd = ["arvados-cwl-runner",
"--submit",
@@ -32,24 +76,125 @@ def start_analysis(api, collection, analysis_project, workflow_uuid):
"--project-uuid=%s" % project["uuid"],
"arvwf:%s" % workflow_uuid,
tmp.name]
- print("Running %s" % ' '.join(cmd))
+ logging.info("Running %s" % ' '.join(cmd))
comp = subprocess.run(cmd, capture_output=True)
if comp.returncode != 0:
- print(comp.stderr.decode('utf-8'))
+ logging.error(comp.stderr.decode('utf-8'))
+
+ return project
+
+
+def start_fastq_to_fasta(api, collection,
+ analysis_project,
+ fastq_workflow_uuid):
+ newproject = run_workflow(api, analysis_project, fastq_workflow_uuid, "FASTQ to FASTA", {
+ "fastq_forward": {
+ "class": "File",
+ "location": "keep:%s/reads.fastq" % collection["portable_data_hash"]
+ },
+ "metadata": {
+ "class": "File",
+ "location": "keep:%s/metadata.yaml" % collection["portable_data_hash"]
+ },
+ "ref_fasta": {
+ "class": "File",
+ "location": "keep:ffef6a3b77e5e04f8f62a7b6f67264d1+556/SARS-CoV2-NC_045512.2.fasta"
+ }
+ })
+ api.collections().update(uuid=collection["uuid"],
+ body={"owner_uuid": newproject["uuid"]}).execute()
+
+def start_pangenome_analysis(api,
+ analysis_project,
+ pangenome_workflow_uuid,
+ validated_project):
+ validated = arvados.util.list_all(api.collections().list, filters=[["owner_uuid", "=", validated_project]])
+ inputobj = {
+ "inputReads": []
+ }
+ for v in validated:
+ inputobj["inputReads"].append({
+ "class": "File",
+ "location": "keep:%s/sequence.fasta" % v["portable_data_hash"]
+ })
+ run_workflow(api, analysis_project, pangenome_workflow_uuid, "Pangenome analysis", inputobj)
+
+
+def get_workflow_output_from_project(api, uuid):
+ cr = api.container_requests().list(filters=[['owner_uuid', '=', uuid],
+ ["requesting_container_uuid", "=", None]]).execute()
+ if cr["items"] and cr["items"][0]["output_uuid"]:
+ return cr["items"][0]
else:
- api.collections().update(uuid=collection["uuid"], body={"owner_uuid": project['uuid']}).execute()
+ return None
+
+
+def copy_most_recent_result(api, analysis_project, latest_result_uuid):
+ most_recent_analysis = api.groups().list(filters=[['owner_uuid', '=', analysis_project]],
+ order="created_at desc", limit=1).execute()
+ for m in most_recent_analysis["items"]:
+ wf = get_workflow_output_from_project(api, m["uuid"])
+ if wf:
+ src = api.collections().get(uuid=wf["output_uuid"]).execute()
+ dst = api.collections().get(uuid=latest_result_uuid).execute()
+ if src["portable_data_hash"] != dst["portable_data_hash"]:
+ logging.info("Copying latest result from '%s' to %s", m["name"], latest_result_uuid)
+ api.collections().update(uuid=latest_result_uuid,
+ body={"manifest_text": src["manifest_text"],
+ "description": "Result from %s %s" % (m["name"], wf["uuid"])}).execute()
+ break
+
+
+def move_fastq_to_fasta_results(api, analysis_project, uploader_project):
+ projects = api.groups().list(filters=[['owner_uuid', '=', analysis_project],
+ ["properties.moved_output", "!=", True]],
+ order="created_at desc",).execute()
+ for p in projects["items"]:
+ wf = get_workflow_output_from_project(api, p["uuid"])
+ if wf:
+ logging.info("Moving completed fastq2fasta result %s back to uploader project", wf["output_uuid"])
+ api.collections().update(uuid=wf["output_uuid"],
+ body={"owner_uuid": uploader_project}).execute()
+ p["properties"]["moved_output"] = True
+ api.groups().update(uuid=p["uuid"], body={"properties": p["properties"]}).execute()
+
def main():
parser = argparse.ArgumentParser(description='Analyze collections uploaded to a project')
parser.add_argument('--uploader-project', type=str, default='lugli-j7d0g-n5clictpuvwk8aa', help='')
- parser.add_argument('--analysis-project', type=str, default='lugli-j7d0g-y4k4uswcqi3ku56', help='')
- parser.add_argument('--workflow-uuid', type=str, default='lugli-7fd4e-mqfu9y3ofnpnho1', help='')
+ parser.add_argument('--pangenome-analysis-project', type=str, default='lugli-j7d0g-y4k4uswcqi3ku56', help='')
+ parser.add_argument('--fastq-project', type=str, default='lugli-j7d0g-xcjxp4oox2u1w8u', help='')
+ parser.add_argument('--validated-project', type=str, default='lugli-j7d0g-5ct8p1i1wrgyjvp', help='')
+
+ parser.add_argument('--pangenome-workflow-uuid', type=str, default='lugli-7fd4e-mqfu9y3ofnpnho1', help='')
+ parser.add_argument('--fastq-workflow-uuid', type=str, default='lugli-7fd4e-2zp9q4jo5xpif9y', help='')
+
+ parser.add_argument('--latest-result-collection', type=str, default='lugli-4zz18-z513nlpqm03hpca', help='')
args = parser.parse_args()
api = arvados.api()
+ logging.info("Starting up, monitoring %s for uploads" % (args.uploader_project))
+
while True:
+ move_fastq_to_fasta_results(api, args.fastq_project, args.uploader_project)
+
new_collections = api.collections().list(filters=[['owner_uuid', '=', args.uploader_project]]).execute()
+ at_least_one_new_valid_seq = False
for c in new_collections["items"]:
- start_analysis(api, c, args.analysis_project, args.workflow_uuid)
- time.sleep(10)
+ at_least_one_new_valid_seq = validate_upload(api, c,
+ args.validated_project,
+ args.fastq_project,
+ args.fastq_workflow_uuid) or at_least_one_new_valid_seq
+
+ if at_least_one_new_valid_seq:
+ start_pangenome_analysis(api,
+ args.pangenome_analysis_project,
+ args.pangenome_workflow_uuid,
+ args.validated_project)
+
+ copy_most_recent_result(api,
+ args.pangenome_analysis_project,
+ args.latest_result_collection)
+
+ time.sleep(15)