Report PanelApp data findings to Genomics England

[tskir]

Follow-up issue from https://github.com/opentargets/platform/issues/1636.

Final set of suggestions will need to be compiled after EFO mapping is implemented as well.

[tskir]

List of some data normalisation issues (updated 2022-09-15)

• Multiple formats used to specify OMIM identifiers:
  • 103580
  • MIM 615553
  • MIM# 615249 (tab character used as a separator)
  • OMIM 610549
  • OMIM #612073
  • OMIM:612186
  • OMIM: 617236
  • OMIM #617767
  • OMIM# 618658
• Multiple formats for specifying other ontologies:
  • ORPHA90636
  • Orphanet:453499
  • OrphaNet: ORPHA404454
• Multiple formats used when both the string and the OMIM identifier are present:
  • #107480:Townes-Brocks syndrome
  • #601869: Deafness, autosomal recessive 15
  • Cenani-Lenz syndactyly syndrome 212780
  • Cenani-Lenz syndactyly syndrome, 212780
  • Centronuclear myopathy 5 ( 615959)
  • Cerebellar ataxia, areflexia, pes cavus, optic atrophy and sensorineural hearing loss (CAPOS, #601338)
  • {Amyotrophic lateral sclerosis, susceptibility to, 13}, 183190
• Typos and formatting errors in specific disease names
  • "Usher syndrome, type 1D, 601067;Deafness, autosomal recessive; 12, 601386;Usher syndrome, type 1D/F digenic, 601067" — Disease name is using a semicolon, which breaks phenotype splitting
  • Combined oxidative phosphorylation deficiency 6, 300816Cowchock syndrome, 310490 — Two diseases are glued to each other
  • 17,20-lyase deficiency, isolated 202110 and 17-alpha-hydroxylase/17,20-lyase deficiency 202110 — Two diseases in one, separated by “and”
• Additional unstructured information present in phenotype strings
  • Publication references: Meier-Gorlin syndrome with craniosynostosis (from PMID 27374770)
  • Disease synonyms: Jervell and Lange-Nielsen syndrome [Congenital sensorineural hearing loss, Prolonged QT interval on EKG, SyncopeTorsades de pointes, Sudden cardiac death, Caused by mutation in the potassium voltage-gated channel, KQT-like subfamily, member 1 gene (KCNQ1), Caused by mutation in the potassium voltage-gated channel, Isk-related subfamily, member 1 gene (KCNE1)]
  • Disease specifiers: 22q11.2 deletion syndrome, Orphanet:567 (includes developmental delay)
• Phenotype names starting with or containing ? with unclear meaning:
  • 616553 ?Dyskeratosis congenita 6 and 7
  • 3- ?-hydroxysterol ?5-oxidoreductase/isomerase deficiency (Disorders of bile acid biosynthesis)
• Very different ways of describing the ontologies:
  • Epileptic encephalopathy, intellectual disability, no OMIM# yet
• Inconsistent separators (; with and without leading/trailing spaces)

Different ways of specifying the publications in the API

98(6):1193-2The list of everything I could find 07. doi: 10.1016/j.ajhg.2016.05.004. PubMed PMID: 27259053, PubMed Central PMCID: PMC4908191.
doi:10.​1007/​s12265-016-9673-5
DOI: https://doi.org/10.1016/j.xhgg.2021.100033
https://doi.org/10.1101/797787
https://doi-org.ezproxy.library.qmul.ac.uk/10.1093/brain/awaa085
PMID: 26933893
PMID: 27078007 (full text not available to confirm findings).
Aldahmesh (2012) Genet Med 14(12):955-962, PMID: 22935719
12702164
15985586 (two siblings)
16060907 (Camilot et al., 2005 report subclinical hypothyroid subjects with heterozygous substitutions
25674101 - review from the same authors as PMID:23972370

2194867118000911 — but displayed as two separate ones: https://panelapp.genomicsengland.co.uk/panels/81/gene/FAM20C/#!details

[tskir]

Investigation of 2022 data update (2020-09-28 → 2022-08-04)

Spreadsheet with data and the metrics is available here: https://docs.google.com/spreadsheets/d/1VBykrN6iyEqBuGJgOJYNOYefINe3dmJ2NjEKCuXmHcE/edit#gid=654888619. The following report contains some exerpts & analysis.

The files being compared are:

1. Old: All_genes_20200928-1959.tsv (2020-09-28, MD5 024bbad3685a0a9797e63314e6e7c77a)
2. New: All_genes_20220804-1350_green_amber_red_public_v1plus_no_superpanels.xlsx (2022-08-04, MD5 6a52bece16f49891f8b9aa7135d0e476).

Compared to the old file, the new file has significantly fewer:

• Total number of records: 60772 → 31103
• Total number of distinct panel IDs: 326 → 297
• Average number of genes per panel: 186 → 105.

The complete list of the 29 panels which are missing from the new file: Panel ID	Panel name	Number of genes
8	Refuted genes	5
14	Multiple bowel polyps	14
28	Congenital neutropaenia	17
32	Kyphoscoliotic Ehlers-Danlos syndrome	3
58	Ehlers-Danlos syndrome type 3	55
64	ClinGen Gene Validity Curations	47
67	Epileptic encephalopathy	183
121	A- or hypo-gammaglobulinaemia	28
124	Combined B and T cell defect	24
135	Dilated Cardiomyopathy (DCM)	75
137	Familial colon cancer	26
160	Genetic Epilepsies with Febrile Seizures Plus (GEFS+)	6
161	Epilepsy Plus	142
170	SCID	25
203	Agranulocytosis	2
204	Bilateral microtia	46
210	ClinGen_Familial thoracic aortic aneurysm and aortic dissection	53
240	Familial Genetic Generalised Epilepsies	25
252	Familial Focal Epilepsies	10
268	Meiges disease	14
289	Multiple Tumours	129
297	Bardet-Biedl Syndrome	22
399	Additional findings health related	14
412	Gene therapy clinical trials	21
657	Autism	735
720	Groopman et al 2019 - Genes with diagnostic variants	66
745	CHARGE syndrome	1
880	Nephrolithiasis and Nephrocalcinosis_KidGen_VCGS	30
928	Viral resistance	24

The metrics which have not significantly changed between the versions:

• Distribution of genes by status: green ~60%, amber ~10%, red ~30% in both datasets
• Total count of unique gene symbols: 5541 → 5438

Finally, our parser successfully runs on the new file and generates the evidence, which valid against the schema. However, in light of several important-looking panels missing, I do not recommend migration to it right away.

[tskir]

Followed up with Eleanor regarding data normalisation

Re-ran the parser on the new data (All_genes_20220804-1350_green_amber_red_public_v1plus_no_superpanels.tsv) with debug tables, made sure that the preprocessing regular expressions are still holding up. As far as I can see, everything still gets processed correctly. The debug tables can be found here, sheets "Phenotypes" and "PMIDs".

There are no more actions on our side. Closing this issue.