aboutsummaryrefslogtreecommitdiff
path: root/doc
diff options
context:
space:
mode:
Diffstat (limited to 'doc')
-rw-r--r--doc/database.org159
1 files changed, 156 insertions, 3 deletions
diff --git a/doc/database.org b/doc/database.org
index d4c04848..df34e7ed 100644
--- a/doc/database.org
+++ b/doc/database.org
@@ -207,10 +207,18 @@ Metadata
?
-** Chr_Length
+** Chr_Length (/cross/BXD.json)
Default mm9, column for mm8
+select * from Chr_Length;
+
+| Name | SpeciesId | OrderId | Length | Length_mm8 |
+| 1 | 1 | 1 | 197195432 | 197069962 |
+| 2 | 1 | 2 | 181748087 | 181976762 |
+
+Table should be merged with
+
** Dataset_mbat
Menu for BXD (linkouts)
@@ -275,10 +283,19 @@ Wiki info (nightly updated from NCBI)
XRef should be foreign keys
-** Geno
+** Geno (genotype/marker/'marker'.json)
SNP or marker info
+INFO:base.trait:.sql: retrieve_info:
+ select Geno.Chr, Geno.Mb from Geno, Species
+ where Species.Name = 'mouse' and
+ Geno.Name = 'rs3693478' and
+ Geno.SpeciesId = Species.Id
+
+| Id | SpeciesId | Name | Marker_Name | Chr | Mb | Sequence | Source | chr_num | Source2 | Comments | used_by_geno_file | Mb_mm8 | Chr_mm8 |
+| 1 | 1 | 01.001.695 | 01.001.695 | 1 | 4.678288 | GCCCTGCCCACCTCAGAGCAAGCTGCCACCCAGGAGTCCGTGTTTCAGGAGATGTGTGAGGAGGGCCTGCTGGAGGAGTGTGATGGTGAGGATGAGGCAGGCCGTGCCGCG[T/C]AGCCAGAGGCTGGTGATGGGACCACCGAGATCTCACCCACTGGTGCTGCTGATCCTGAGAAGAGGATGGAGAAGAAGACGGAGCAGCAGCACACCGGCGGCGGGAGAAAGCTGCTCGTAAGCTGCTCGTAAGCTACGGGTGCAGCAGGCTGCACTTAGGGCAGCCCGGCTTCAGCACCAAGAACTCTTCAGGCTGCATGGGATCAAGGCCCAGGTGGCCCGAAGGCTGGCAGAACTCGCACACGGGAGGGAGCAGCAGCGCATACAGCGACTGGCAGAGGCTGACAAGCCCCGAAGGCTGGGACGACTCAAGTACCAGGCTCCTGACATTGATGTGCAGCTCAGCTCTGAGCTGTCTGGCCCACTCAGGACACTGAAACCAGAAGGTCACATTCTCCAAGACAGGTTCAAGAGCTTCCAGAAGAGAAATATGATTGAGCCCCGAGAACGAGCCAAGTTCAAGCGCAAATAAAAAATGAAGTTGGTGGAGAAGCGGGCCTACCATGAGATTCAGTTGTAGCTGTGCAGATGTCGGAGCCCCGCCCCTCAATAAAGTTCTGTGACAAAAAAAAAAAAAAAAAAAGAAGAAGAAGAAGAAAAGGAAAAAAAAGAAGAAAAAGAAAAAAAAAGAAAAAAGAAAAAGAAAACACATCACTTGGCAAAACTCCATAGACTCTATGTGATTCATGTTTCAAACATGCACCTA | GNF_SNP | 1 | GNF | NULL | NULL | 4.678288 | 1 |
+
** GenoCode
Belongs to someone else
@@ -311,7 +328,7 @@ Heritability for probeset(?)
Homology, not used much
-** InbredSet
+** InbredSet (/cross/BXD.info)
Group in menu
@@ -490,6 +507,36 @@ select count(*) from ProbeSet limit 5;
2 rows in set (0.00 sec)
+** ProbeSetXRef (phenotypes/dataset_name.json)
+
+For every probe set (read dataset measuring point):
+
+select * from ProbeSetXRef;
+
+| ProbeSetFreezeId | ProbeSetId | DataId | Locus_old | LRS_old | pValue_old | mean | se | Locus | LRS | pValue | additive | h2 |
+| 112 | 123528 | 23439389 | NULL | NULL | NULL | 6.7460707070707 | NULL | rs6239372 | 10.9675593568894 | 0.567 | 0.0448545966228878 | NULL |
+| 112 | 123527 | 23439388 | NULL | NULL | NULL | 6.19416161616162 | NULL | rs13476936 | 10.9075670392762 | 0.567 | -0.0358456732993988 | NULL |
+
+where ProbeSetFreezeId is the dataset (experiment). ProbesetId refers
+to the probe set information (measuring point). DataId points to the
+data point. The othe values are used for search.
+
+It is used in search thus:
+
+SELECT distinct ProbeSet.Name as TNAME,
+ ProbeSetXRef.Mean as TMEAN, ProbeSetXRef.LRS as TLRS,
+ ProbeSetXRef.PVALUE as TPVALUE, ProbeSet.Chr_num as TCHR_NUM,
+ ProbeSet.Mb as TMB, ProbeSet.Symbol as TSYMBOL,
+ ProbeSet.name_num as TNAME_NUM
+FROM ProbeSetXRef, ProbeSet
+WHERE ProbeSet.Id = ProbeSetXRef.ProbeSetId
+ and ProbeSetXRef.ProbeSetFreezeId = 112
+ ORDER BY ProbeSet.symbol ASC limit 5;
+
+| TNAME | TMEAN | TLRS | TPVALUE | TCHR_NUM | TMB | TSYMBOL | TNAME_NUM |
+| 1445618_at | 7.05679797979798 | 13.5417452764616 | 0.17 | 8 | 75.077895 | NULL | 1445618 |
+| 1452452_at | 7.232 | 30.4944361132252 | 0.0000609756097560421 | 12 | 12.6694 | NULL | 1452452 |
+
** ProbeSetData
Probedata - main molecular data. Probesets, metabolome,
@@ -691,6 +738,19 @@ show indexes from ProbeSetFreeze;
| ProbeSetFreeze | 1 | NameIndex | 1 | Name2 | A | 2 | NULL | NULL | | BTREE | | |
+----------------+------------+-----------+--------------+-------------+-----------+-------------+----------+--------+------+------------+---------+---------------+
+** ProbeSetSE
+
+ select * from ProbeSetSE limit 5;
++--------+----------+----------+
+| DataId | StrainId | error |
++--------+----------+----------+
+| 1 | 1 | 0.681091 |
+| 1 | 2 | 0.361151 |
+| 1 | 3 | 0.364342 |
+| 1 | 4 | 0.827588 |
+| 1 | 5 | 0.303492 |
++--------+----------+----------+
+
** Publication and publishdata (all pheno)
Phenotype pubs
@@ -794,6 +854,18 @@ INFO:db.call:.sql: __init__:
INFO:db.call:.sql: ('BXD', 1)
+The actual search is
+
+SELECT distinct ProbeSet.Name as TNAME, 0 as thistable,
+ ProbeSetXRef.Mean as TMEAN, ProbeSetXRef.LRS as TLRS,
+ ProbeSetXRef.PVALUE as TPVALUE, ProbeSet.Chr_num as TCHR_NUM,
+ ProbeSet.Mb as TMB, ProbeSet.Symbol as TSYMBOL,
+ ProbeSet.name_num as TNAME_NUM
+FROM ProbeSetXRef, ProbeSet
+WHERE ProbeSet.Id = ProbeSetXRef.ProbeSetId
+ and ProbeSetXRef.ProbeSetFreezeId = 112
+ ORDER BY ProbeSet.symbol ASC limit 5;
+
INFO:base.species:.sql: __init__:
Select
Chr_Length.Name, Chr_Length.OrderId, Length from Chr_Length, InbredSet
@@ -876,6 +948,87 @@ INFO:base.data_set:.sql: get_trait_info:
** Fetch phenotypes
+*** Through the trait page
+
+When hitting the trait page, e.g.
+
+curl "http://localhost:5003/show_trait?trait_id=1443823_s_aet=HC_M2_0606_P"
+
+First the BXD's are queried with
+
+DEBUG:base.data_set:.get_samplelist: Sample list: : ['BXD1',
+ 'BXD2',
+ 'BXD5',
+ ...
+
+main probeset info (trait) is retrieved with
+
+SELECT ProbeSet.name, ProbeSet.symbol, ProbeSet.description, ProbeSet.probe_target_description, ProbeSet.chr, ProbeSet.mb, ProbeSet.alias, ProbeSet.geneid, ProbeSet.genbankid, ProbeSet.unigeneid, ProbeSet.omim, ProbeSet.refseq_transcriptid, ProbeSet.blatseq, ProbeSet.targetseq, ProbeSet.chipid, ProbeSet.comments, ProbeSet.strand_probe, ProbeSet.strand_gene, ProbeSet.probe_set_target_region, ProbeSet.probe_set_specificity, ProbeSet.probe_set_blat_score, ProbeSet.probe_set_blat_mb_start, ProbeSet.probe_set_blat_mb_end, ProbeSet.probe_set_strand, ProbeSet.probe_set_note_by_rw, ProbeSet.flag
+ FROM ProbeSet, ProbeSetFreeze, ProbeSetXRef
+ WHERE
+ ProbeSetXRef.ProbeSetFreezeId = ProbeSetFreeze.Id AND
+ ProbeSetXRef.ProbeSetId = ProbeSet.Id AND
+ ProbeSetFreeze.Name = 'HC_M2_0606_P' AND
+ ProbeSet.Name = '1443823_s_at'
+
+Followed by
+
+INFO:base.trait:.sql: retrieve_info:
+ SELECT
+ ProbeSetXRef.Locus, ProbeSetXRef.LRS, ProbeSetXRef.pValue, ProbeSetXRef.mean, ProbeSetXRef.additive
+ FROM
+ ProbeSetXRef, ProbeSet
+ WHERE
+ ProbeSetXRef.ProbeSetId = ProbeSet.Id AND
+ ProbeSet.Name = "1443823_s_at" AND
+ ProbeSetXRef.ProbeSetFreezeId =112
+
+| Locus | LRS | pValue | mean | additive |
+| NES13033186 | 35.466324074542 | 0.00000900000000003676 | 15.0551313131313 | -0.16750405405405402 |
+
+Then the interesting bit, the sample data is fetched with
+
+INFO:base.data_set:.sql: retrieve_sample_data:
+ SELECT
+ Strain.Name, ProbeSetData.value, ProbeSetSE.error, ProbeSetData.Id, Strain.Name2
+ FROM
+ (ProbeSetData, ProbeSetFreeze, Strain, ProbeSet, ProbeSetXRef)
+ left join ProbeSetSE on
+ (ProbeSetSE.DataId = ProbeSetData.Id AND ProbeSetSE.StrainId = ProbeSetData.StrainId)
+ WHERE
+ ProbeSet.Name = '1443823_s_at' AND ProbeSetXRef.ProbeSetId = ProbeSet.Id AND
+ ProbeSetXRef.ProbeSetFreezeId = ProbeSetFreeze.Id AND
+ ProbeSetFreeze.Name = 'HC_M2_0606_P' AND
+ ProbeSetXRef.DataId = ProbeSetData.Id AND
+ ProbeSetData.StrainId = Strain.Id
+ Order BY
+ Strain.Name
+
+| Name | value | error | Id | Name2 |
+| 129S1/SvImJ | 14.552 | NULL | 23422417 | 129S1/SvImJ |
+| A/J | 14.34 | NULL | 23422417 | A/J |
+| AKR/J | 14.338 | NULL | 23422417 | AKR/J |
+| B6D2F1 | 15.251 | NULL | 23422417 | B6D2F1 |
+| BALB/cByJ | 14.164 | NULL | 23422417 | BALB/cByJ |
+| BALB/cJ | 14.563 | NULL | 23422417 | BALB/cJ |
+| BXD1 | 15.198 | NULL | 23422417 | BXD1 |
+| BXD11 | 15.084 | NULL | 23422417 | BXD11 |
+| BXD12 | 15.192 | NULL | 23422417 | BXD12 |
+ etc.
+
+Then some repeated queries and this fetchest the nearest SNP
+
+INFO:wqflask.show_trait.show_trait:.sql: get_nearest_marker:
+ SELECT Geno.Name FROM Geno, GenoXRef, GenoFreeze WHERE Geno.Chr = '1'
+ AND GenoXRef.GenoId = Geno.Id AND GenoFreeze.Id =
+ GenoXRef.GenoFreezeId AND GenoFreeze.Name = 'BXDGeno' ORDER BY ABS(
+ Geno.Mb - 173.149434) limit 1;
+
+| Name |
+| NES13033186 |
+
+*** Digging deeper
+
To get at phenotype data ProbeSetData is the main table (almost all
important molecular assay data is in this table including probe set
data, RNA-seq data, proteomic data, and metabolomic data. 2.5 billion