Perform release-related activities for 20.06

[AsierGonzalez]

A number of checks have to be performed on the evidence files submitted by data providers as they arrive and before the pipeline is run to see whether the agreed changes have been implemented.

[AsierGonzalez]

TEPs checked on May 20th and there are no updates

[AsierGonzalez]

EuropePMC evidence file received on May 23. It was checked and it looks good:

• 9,790,732 evidence strings (up from 9,655,854 in 20.04)
• Valid JSON
• 1,287 evidence strings annotated with COVID-19 (MONDO_0100096). However, the Hoffmann et al. Cell paper is not among them, checking with Shyama why this might be.

[AsierGonzalez]

PheWAS catalog evidence updated on 26th May:

• A few traits for which new terms were requested to be added to EFO now have a mapping, e.g. Abnormal chest sounds - see updates
• Unique association fields changed to reduce the number of duplicates (see commit):
  • phewas_string replaced by phewas_string_and_code, e.g. "Colorectal cancer" vs "Colorectal cancer [153]"
  • Variant id added (variant_id)
• 191k evidence strings in total (compared to 187k in 20.04)

[AsierGonzalez]

ChEMBL file received on 21st May:

• 434,878 evidence strings (up from 416,174 in 20.04)
• Valid JSON
• ChEMBL version in evidence strings (evidence.target2drug.provenance_type.database.version) changed from "26" to "27" as it was generated based on ChEMBL 27. New file called cttv008-27-05-2020.json.gz and uploaded to the ChEMBL Google bucket.
• It contains 761 evidence strings annptated with COVID-19 (MONDO_0100096)

[AsierGonzalez]

Baseline and differential expression files received on 12th May.

• Baseline:
  • Data processed using Ensembl 79 and Ensembl 95 genome references depending on data set. Inconsistencies observed
  • 43,663 genes (down from 50,211)
  • Bins and z-scores computed
• Differential:
  • Data from Expression Atlas release 35
  • 375,338 evidence strings received, which end in 221,633 after filtering based on activity (up from 215,921 in 20.04)
  • Filtering still performed in-house
  • Valid JSON

[AsierGonzalez]

Chemical Probes updated on 1st June:

• One new probe (NVS-MALT1) in SGC, there are 77 in total.
• 12 new probes in Chemical Probes Portal: ABBV-744, AT1, BAY-885, CM11, Eleutherobin, Glyburide, ICI-199441, NI-57, PFI-3, RO2468, T-26c, UCSF7447. There are 135 probes in total.
  • Eleutherobin is a particular case as it targets tubulin, which is an structural protein formed by the alpha- and beta-tubulins. Given that protein complexes are not considered in the platform yet, it has been decided to assign this chemical probe to all the alpha- and beta-tubulins in Open Targets, which is a total of 17. A better solution for this will be thought of in the future.
• There are still 65 chemical probes in Open Science Probes/Donated Chemical Probes portal, there not been any changes

[AsierGonzalez]

The first 20.06 pipeline run has failed due to some ChEMBL evidence strings having the evidence.drug2clinic.date_asserted in the yyyy-dd-mm format instead of yyyy-mm-dd. ChEMBL have generated a new evidence file with the correct format and OT have changed the JSON schema to capture this issue (see #1090 and PR #87). The new ChEMBL evidence file looks good:

• It has the same number of evidence strings as the previous one
• It’s valid against the new json schema version (1.6.8)
• The date of the offending 5 evidence strings that was 2016-25-11T00:00:00.000Z is now 2016-11-25T00:00:00.000Z

[AsierGonzalez]

EVA file received on 22nd May:

• 125,138 evidence strings (up from 122,219 in 20.04)
• Valid JSON
• There are still 20 evidence strings where the condition is Alhzeimer disease (AD) and they are mapped to EFO_0000249 . These were manually mapped to Early-onset autosomal dominant Alzheimer disease (Orphanet_1020) - see #104
• 60 trinucleotide repeat expansions in evidence file thanks to new variant annotation pipeline developed by EVA, which identified 85 trinucleotide repeat expansions in ClinVar. According to Kirill, the majority of missing ones are because their clinical significance is bening or not provided. However, there are about 10 variants that lack trait-to-ontology mapping and need to be checked manually. This will be revisited in the next release.

[AsierGonzalez]

Python script to generate metrics revamped and bash script included so that now all the information needed to fill the release metrics spreadsheets is created in one go and it requires very little editing. See PR #2

[AsierGonzalez]

Observations looking into invalid evidence strings in first 20.06 run ():

• Expression atlas: There are 245 invalid evidence strings because they are annotated with a disease that is not in EFO. The problematic disease terms are "pulmonary sarcoidosis" (MONDO_0001708) and "rheumatoid factor-negative juvenile idiopathic arthritis" (MONDO_0019432). Email sent to atlas to discuss the issue and potential solutions.
• Gene2Phenotype: 2 invalid evidence strings despite there has not been an update and there were none in 20.04. Both evidence strings are annotated with MONDO_0029146, which is deprecated. The gene2phenotype disease is "Saul Wilson Syndrome/Saul-Wilson Syndrome" (there are separate entries for both spellings) and Orphanet_85172 may be a potential mapping as it has a xref to OMIM:618150. It will be reviewed as part of the gene2phenotype disease mapping effort (#1077).
• PheWAS Catalog: 8,747 invalid evidence strings due to 59 disease that are not in EFO. It seems to be mixture of EFO terms in branches exclude in OT slim (e.g. "ileostomy" - EFO_0010681 or "pigmentation disease" - EFO_1000755 HP and MONDO terms not yet in EFO (e.g. "neutropenia" - HP_0001875 or "arthropathy" - MONDO_0006816) and invalid ids due to misspellings (e.g. MONDO_002144 instead of MONDO_0021440 for benign neoplasm of skin. @smnorthen is helping with the assessment of the issue.
• EVA: There are 68 invalid evidence strings either due to invalid targets or diseases.
  • Invalid targets: ENSG00000285395 invalid because it's on a non-standard chromosome. ENSG00000103489 could be used instead for target XYLT1.
  • Invalid diseases: There are 67 invalid evidence strings due to 10 invalid disease ids. Six of them are MONDO ids and 4 EFOs. The majority of MONDO ids are terms related to disease susceptibility that are not included in the OT slim EFO ("otitis media, susceptibility to" - MONDO_0008162, "invasive pneumococcal disease, recurrent isolated, 2" - MONDO_0010387, "orofacial cleft 6, susceptibility to" - MONDO_0012141, "glioma susceptibility 2" - MONDO_0013092, "glioma susceptibility 3" - MONDO_0013093) and one of them is the same obsolote term as the one in gene2phenotype ("Saul Wilson Syndrom" - MONDO_0029146). The invalid EFO ids seem to be excluded from the OT slim because they are either too general ("disease" - EFO_0000408) or not diseases ("SKMC" - EFO_0002801, "blood group--lutheran inhibitor" - EFO_0009171, "ATP binding cassette subfamily B member 6" - EFO_0009916 ) so the issue could be with the ClinVar trait or the mapping. It seems that the obsolote MONDO id and the non-disease EFOs could be fixed so I have contacted EVA.
• Genetics portal: There are 3,118 invalid evidence strings that are further divided as follows:
  • Invalid targets: 405 evidence strings, caused by 25 deprecated Ensembl ids.
  • Invalid diseases: 2,709 invalid evidence strings caused by 80 invalid ontology terms.
  • Invalid target and disease: 4 evidence strings where both the target and disease are invalid. This is caused by 4 deprecated gene ids (ENSG00000278272, ENSG00000283932, ENSG00000274267, ENSG00000270898), one HP term ("Hyperthyroidism" - HP_0000836) that is not included in EFO because there already is an equivalente EFO term (EFO_0009189) and an EFO term excluded from the OT slim-EFO ("comparative body size at age 10, self-reported" - EFO_0009819). The targets should be fixed the next time the data is updated and the diseases should be fixed by the ongoing manual disease mapping.
• ChEMBL: There are 303 invalid evidence strings that are further divided as follows:
  • Invalid targets ("unknown_uniprot_entry"): 37 invalid evidence strings caused by 3 UniProt ids (P43627, P13762, P79483) that are not included in OT. They seem to be valid targets (KIR2DL2, HLA-DRB4, HLA-DRB3) that are only annotated in non-standard chromosomes. so there is nothing that can be done.
  • Invalid diseases: 266 invalid evidence strings caused by 20 disease ids. This includes two partial URIs (/MONDO_0002254 instead of http://purl.obolibrary.org/obo/MONDO_0002254 and /MONDO_0024317 instead of http://purl.obolibrary.org/obo/MONDO_0024317), 12 HP terms not in EFO ( "Hypogonadism" - HP_0000135, "Growth hormone deficiency" - HP_0000824, "Hypoparathyroidism" - HP_0000829, "Adrenal insufficiency" - HP_0000846, "Hyperuricemia" - HP_0002149, "Gastric ulcer" - HP_0002592, "Renal angiomyolipoma" - HP_0006772, "Ocular hypertension" - HP_0007906, "Primary adrenal insufficiency" - HP_0008207, "Heparin-induced thrombocytopenia" - HP_0011874, "Myocarditis" - HP_0012819, "Macular edema" - HP_0040049), 1 HP term that has been replaced (HP_0001587 by "Premature ovarian insufficiency" - HP_0008209 which is in EFO) and 4 EFO terms not part of the slim-EFO used by OT ("digestive system disease" - EFO_0000405, "mental health" - EFO_0003935, "chronic disease" - EFO_0009714, "inflammatory disease" - EFO_0009903) and one obsolete EFO term ("influenza infection" - EFO_0001669). Some of those HP terms have equivalent terms in EFO and those could be used instead (e.g. MONDO_0002146 for "hypogonadism", EFO_0009451 for "hypoparathyroidism", EFO_0009104 for "hyperuricemia", EFO_0009454 for "gastric ulcer", EFO_1001069 for "ocular hypertension", MONDO_0015128 for "primary adrenal insufficiency", EFO_0009609 for "myocarditis"), some others have matching synonyms ("adrenal insufficiency" is a synonym of "adrenocortical insufficiency" - EFO_0009491, a synonym of the HP term "Renal angiomyolipoma" is "Kidney Angiomyolipoma" which exists in EFO as [EFO_1000312](http://www.ebi.ac.uk/efo/EFO_1000312, "macular edema" is a synonym of "macular retinal edema" - MONDO_0003005) whereas the others would need to be imported. I will discuss these issues with ChEMBL.
• UniProt: The are 78 invalid evidence strings that are further divided as follows:
  • Invalid targets ("unknown_uniprot_entry"): Unrecognised UniProt id because Q156AA - ATXN8 is not mapped to any Ensembl gene.
  • Validator error: Known issue with 7 evidence strings where the variant type is snp multiple (see #809)
  • Invalid diseases: 70 invalid evidence strings caused by 9 disease ids. 8 of them are MONDO ids and 7 of them are disease susceptibility related terms as happened with EVA invalid evidence strings ("otitis media, susceptibility to" - MONDO_0008162, "hemolytic uremic syndrome, atypical, susceptibility to, 1" - MONDO_0009335 , "invasive pneumococcal disease, recurrent isolated, 2" - MONDO_0010387, "orofacial cleft 6, susceptibility to" - MONDO_0012141, "glioma susceptibility 2" - MONDO_0013092, "glioma susceptibility 3" - MONDO_0013093, "neural tube defects, susceptibility to" - MONDO_0020705). The other one, is the obsolote MONDO term also seen in EVA and gene2phenotype ("Saul Wilson Syndrom" - MONDO_0029146). The invalid EFO id is a cell line ("GM00719" - EFO_0004922).
• Reactome: The are 932 invalid evidence strings that are further divided as follows:
  • Exception: 72 invalid evidence due to 9 UniProt ids (A6NLF2, P04908, P0C0S8, P62805, P62807, P68431, Q6FI13, Q71DI3, Q8NG57) mapping to multiple genes. It should be solved by running Miguel's UniProt fix as it is already done with UniProt, ChEMBL and EuropePMC evidence.
  • Invalid target ("unknown_uniprot_entry"): 444 invalid evidence strings caused by 28 UniProt ids that are not included in OT. It does not seem to be an easy fix for these.
  • Invalid disease: 416 invalid evidence strings caused by 8 disease ids. 4 of them are HP terms not in EFO but with equivalente terms that could be used instead ("Macular degeneration" - HP_0000608 -> EFO_0009606, "Hyperinsulinemic hypoglycemia" - HP_0000825 -> EFO_0007318, "Homocystinuria" - HP_0002156 -> MONDO_0004737, "Open angle glaucoma" - HP_0012108 -> EFO_0004190), one of them is an obsolte HP term that has an alternative term available that is already in EFO ("Glucose intolerance" - HP_0000833 -> HP_0001952), one of them is the same as a synonym of a EFO term ("Adenocarcinoma of the large intestine" - HP_0040275 -> "colorectal adenocarcinoma" - EFO_0000365), one is a HP term with no obvious equivalent term that may need to be imported to EFO ("Hyperphosphatemia" - HP_0002905) and one is an obsolete EFO term where an alternative term exists ("influenza infection" - EFO_0001669 -> "influenza" - EFO_0007328). I will talk to Reactome to see if the invalid diseases can be solved.
• Genomics England: There are 525 invalid evidence strings (336 invalid diseases, 4 invalid targets and 184 validation errors). They will be investigated further when the data is updated.
• Cancer Gene Census: There are 73 invalid evidence strings due to 5 invalid diseases. One is a MONDO term not in EFO but with an equivalent term ("meningioma (disease)" - MONDO_0016642 -> "Meningioma" - Orphanet_2495) and the other four are UBERON tissue terms that should be corrected ("tissue" - UBERON_0000479, "endometrium" - UBERON_0001295, "lymphoid tissue" - UBERON_0001744, "hematopoietic tissue" - UBERON_0012429 ) Cancer Gene Census is not providing updates at the moment, so I will talk to them about it once they start again.
• SLAPenrich: there are 13 invalid evidence strings that are further divided as follows:
  • Invalid targets: 8 invalid evidence strings caused by 2 deprecated Ensembl ids ( ENSG00000130489,ENSG00000150526).
  • Validation error: 5 invalid evidence strings because the target ids do not pass validation because they lack the Ensembl id (http://identifiers.org/ensembl/). These will be reviewed if the data is updated.
• PROGENy: 84 invalid evidence strings because they are annotated with the invalid disease EFO_0000000. This will be reviewed if the data is updated.
• PhenoDigm: The are 1,753 invalid evidence strings that are further divided as follows:
  • Invalid targets: 234 invalid evidence strings caused by 3 Ensembl ids. Two of those are genes in "non-standard" chromosomes that do not have counterparts in he main assembly (ENSG00000274847, ENSG00000277796), whereas the other one is a deprecated id (ENSG00000285258).
  • Invalid diseases: 1,518 invalid evidence strings caused by 7 diseases. 6 of those are deprecated EFO terms that have been replaced by newer terms ("adrenocortical carcinoma" - EFO_0003093, "Job's syndrome" - EFO_0003775, "acrocephalosyndactylia" - EFO_0004123, "Mobius syndrome" - EFO_1001046, "Birt-Hogg-Dube Syndrome" - EFO_1001273, "triple-A syndrome" - EFO_1001997). The last one is a term that has disappeared from EFO between 20.04 and 20.06 (Orphanet_94062)
  • Invalid target and disease: 1 invalid evidence string due to the combination of one gene in a "non-standard" chromosome (ENSG00000277796, see above) and an obsolete EFP term (EFO_0003775, see above) These will be reviewd the next time PhenoDigm is updated.
• EuropePMC: 1,194,340 invalid evidence strings that are further divided as follows:
  • Invalid targets ("unknown_uniprot_entry"): 1,099,432 invalid evidence strings due to 11,112. I have not checked all of them but probably most of them are non-human proteins.
  • Invalid diseases: 84,891 invalid diseases due to 27 diseases. All the missing diseases are in EFO but they are excluded from the slim-EFO used by OT. Some of the excluded terms look like actual diseases so I am checking it with @smnorthen and @d0choa.
  • Invalid target and disease: 10,017 invalid evidence due to 1,622 targets (probably mouse proteins) and 23 invalid diseases, all of them included in the list of 27 mentioned above. Unless there are any issues with the ontology, there will be no action taken, except for controlling that this number remains reasonably stable accross releases.

[d0choa]

Great job @AsierGonzalez. All the follow-ups make sense to me