Update mariadb-database-architecture.gmi

Updated information on data upload to gn2 production stage

1 files changed, 258 insertions, 18 deletions

diff --git a/topics/database/mariadb-database-architecture.gmi b/topics/database/mariadb-database-architecture.gmi
index b200dbd..5c9b0c5 100644
--- a/topics/database/mariadb-database-architecture.gmi
+++ b/topics/database/mariadb-database-architecture.gmi
@@ -534,32 +534,30 @@ select * from ProbeSetSE limit 5;
 
 # More information
 
-For the other tables, you may check the GN2/doc/database.org document (that was the starting point for this document).
+For the other tables, you may check the GN2/doc/database.org document (the starting point for this document).
 
-## Additional Information
+# Contributions regarding data upload to the GeneNetwork webserver 
+* Ideas shared by the GeneNetwork team to facilitate the process of uploading data to production 
 
-* This includes all contributions from the gn2 team on making sure the process of uploading data to gn2 production is a tremendous success, enhancing the process of uploading using Fred's upload system, and most importantly, making sure that the dataset uploaded is valid and can be tested and used by investigators while in gn2 webserver.
-
-### Arthur's contributions
-
-#### On the quality check and integrity of the data to be uploaded to gn2
+## Quality check and integrity of the data to be uploaded to gn2
 
 * A note to add (from Arthur): Some datasets have the following identifiers: ProbeSet IDs {chr_3020701, chr_3020851, etc}. This is not an acceptable way to name the probeset IDs. So, the data provider needs to understand what format is needed for gn2 to accept the ProbeSet IDs in their dataset
 * Also, for the annotation file, among other important columns, it is crucial that there are descriptions, aliases, and location columns. And the formatting should be exactly as found in the public repositories such as NCBI, Ensembl, etc. For instance, for description: `X-linked Kx blood group related 4`, and Aliases: ` XRG4; Gm210; mKIAA1889` as in
 => https://www.ncbi.nlm.nih.gov/gene/497097
 
-#### On the valid ProbeSetIDs
+## Valid ProbeSetIDs
 
-* He suggested that the official ProbeSetIDs would be the one from the vendor. This would also constitute the platform used to generate data {Novogene-specific platform}, for instance; `NovaSeqPE150` for the MBD UTHSC mice seq dataset
+* The official ProbeSetIDs would be the one from the vendor. This would also constitute the platform used to generate data {Novogene-specific platform}, for instance; `NovaSeqPE150` for the MBD UTHSC mice seq dataset
 * NB; in this case, if the vendor does not provide the official names as expected, we can use the platform + the numbering order of the file to generate probeset IDs. For instance; `NseqPE150_000001 to NseqPE150_432694` for samples 1 to 432694
+* Avoid IDs with meaning, e.g. =chr1_3020701= → Chromosome 1 at 3020701 base pairs. Prefer IDs with no meaning
 
-### Rob's contributions
-
-#### On the importance of having unique identifiers within a platform
+## The importance of having unique identifiers within a platform
 
 * Unique identifiers solve the hurdles that come with having duplicate genes. So, the QA tools in place should ensure the uploaded dataset adheres to the requirements mentioned
-* However, newer RNA-seq data sets generated by sequencing do not usually have an official vendor identifier. The identifier is usually based on the NCBI mRNA model (NM_XXXXXX) that was used to evaluate an expression and on the sequence that is involved, usually the start and stop nucleotide positions based on a specific genome assembly or just a suffix to make sure it is unique. In this case, you are looking at mRNA assays for a single transcript, but different parts of the transcript that have different genome coordinates. We now typically use ENSEMBL identifiers. Here is the mouse version of the sonic hedgehog gene as an example: `ENSMUST00000002708` or `ENSMUSG00000002633` sources should be fine. The important thing is to know the provenance of the ID—who is in charge of that ID type?
-* When a mRNA assay is super precise (one exon only or a part of the 5' UTR), then we should use exon identifiers from ENSEMBL probably. Ideally, we should enter the first and last 100 nt of the sequence in GeneNetwork for verification and  alignment. We did this religiously for arrays, but have started to get lazy now. The sequence is the ultimate identifier
+* However, newer RNA-seq data sets generated by sequencing do not usually have an official vendor identifier. The identifier is usually based on the NCBI mRNA model (NM_XXXXXX) that was used to evaluate an expression and on the sequence that is involved, usually the start and stop nucleotide positions based on a specific genome assembly or just a suffix to make sure it is unique. In this case, you are looking at mRNA assays for a single transcript, but different parts of the transcript that have different genome coordinates. We now typically use ENSEMBL identifiers.
+* The mouse version of the sonic hedgehog gene as an example: `ENSMUST00000002708` or `ENSMUSG00000002633` sources should be fine. The important thing is to know the provenance of the ID—who is in charge of that ID type?
+* When a mRNA assay is super precise (one exon only or a part of the 5' UTR), then we should use exon identifiers from ENSEMBL probably. 
+* Ideally, we should enter the sequence's first and last 100 nt in GeneNetwork for verification and  alignment. We did this religiously for arrays, but have started to get lazy now. The sequence is the ultimate identifier
 * For methylation arrays and CpG assays, we can use this format `cg14050475` as seen in MBD UTHSC Ben's data
 * For metabolites like isoleucine—the ID we have been using is the mass-to-charge (MZ) ratio such as `130.0874220_MZ`
 * For protein and peptide identifiers we have used the official Protein ID followed by an underscore character and then some or all of the sequence. This is then followed by another underscore and a number. Evan to confirm, but the suffix number is the charge state if I remember correctly
@@ -577,12 +575,254 @@ abcb10_q9ji39_t312
 * The above is just the gene symbol then the protein ID and not so sure what t311 and t312 mean
 * Ideally these IDs are explained to some extent when they embed some information
 
-### Zach's contributions
 
-#### Regarding the BXD individuals as far as GeneNetwork is concerned
 
-* Basically groups (represented by the InbredSet tables) are primarily defined by their list of samples/strains (represented by the Strain tables). When we create a new group, it's because we have data with a distinct set of samples/strains from any existing groups. So when we receive data for BXD individuals, as far as the database is concerned they are a completely separate group (since the list of samples is new/distinct from any other existing groups). We can choose to also enter it as part of the "generic" BXD group (by converting it to strain means/SEs using the strain of each individual, assuming it's provided like in the files Arthur was showing us). This same logic could apply to other groups as well - we could choose to make one group the "strain mean" group for another set of groups that contain sample data for individuals. But the database doesn't reflect the relationship between these groups*
+## BXD individuals
+
+* Basically groups (represented by the InbredSet tables) are primarily defined by their list of samples/strains (represented by the Strain tables). When we create a new group, it's because we have data with a distinct set of samples/strains from any existing groups. 
+* So when we receive data for BXD individuals, as far as the database is concerned they are a completely separate group (since the list of samples is new/distinct from any other existing groups). We can choose to also enter it as part of the "generic" BXD group (by converting it to strain means/SEs using the strain of each individual, assuming it's provided like in the files Arthur was showing us). 
+* This same logic could apply to other groups as well - we could choose to make one group the "strain mean" group for another set of groups that contain sample data for individuals. But the database doesn't reflect the relationship between these groups*
 * As far as the database is concerned, there is no distinction between strain means and individual sample data - they're all rows in the ProbeSetData/PublishData tables. The only difference is that strain mean data will probably also have an SE value in the ProbeSetSE/PublishSE tables and/or an N (number of individuals per strain) value in the NStrain table
 * As for what this means for the uploader - I think it depends on whether Rob/Arthur/etc wants to give users the ability to simultaneously upload both strain mean and individual data. For example, if someone uploads some BXD individuals' data, do we want the uploader to both create a new group for this (or add to an existing BXD individuals group) and calculate the strain means/SE and enter it into the "main" BXD group? My personal feeling is that it's probably best to postpone that for later and only upload the data with the specific set of samples indicated in the file since it would insert some extra complexity to the uploading process that could always be added later (since the user would need to select "the group the strains are from" as a separate option)
-* In general Fred and I are probably the best people to ask about database stuff, while Arthur (and maybe also Fred?) is the best reference for the actual files that are submitted (which I'm not that familiar with). Arthur is familiar with the database structure, but not as familiar with the way the code interacts with it
 * The relationship is sorta captured in the CaseAttribute and CaseAttributeXRefNew tables (which contain sample metadata), but only in the form of the metadata that is sometimes displayed as extra columns in the trait page table - this data isn't used in any queries/analyses currently (outside of some JS filters run on the table itself) and isn't that important as part of the uploading process (or at least can be postponed)
+
+## Individual Datasets and Derivatives datasets in gn2 
+* Individual dataset reflects the actual data provided or submitted by the investigator (user). Derivative datasets include the processed information from the individual dataset, as in the case of the average datasets. 
+* An example of an individual dataset would look something like; (MBD dataset) 
+```
+#+begin_example
+sample, strain, Sex, Age,…
+FEB0001,BXD48a,M,63,…
+FEB0002,BXD48a,M,15,…
+FEB0003,BXD48a,F,22,…
+FEB0004,BXD16,M,39,…
+FEB0005,BXD16,F,14,…
+⋮
+#+end_example
+```
+* The strain column above has repetitive values. Each value has a one-to-many relationship with values on sample column. From this dataset, there can be several derivatives. For example; 
+- Sex-based categories 
+- Average data (3 sample values averaged to one strain value) 
+- Standard error table computed for the averages 
+
+## Saving data to database 
+* Strain table schema 
+```
+#+begin_src sql
+  MariaDB [db_webqtl]> DESC Strain;
+  +-----------+----------------------+------+-----+---------+----------------+
+  | Field     | Type                 | Null | Key | Default | Extra          |
+  +-----------+----------------------+------+-----+---------+----------------+
+  | Id        | int(20)              | NO   | PRI | NULL    | auto_increment |
+  | Name      | varchar(100)         | YES  | MUL | NULL    |                |
+  | Name2     | varchar(100)         | YES  |     | NULL    |                |
+  | SpeciesId | smallint(5) unsigned | NO   |     | 0       |                |
+  | Symbol    | varchar(20)          | YES  | MUL | NULL    |                |
+  | Alias     | varchar(255)         | YES  |     | NULL    |                |
+  +-----------+----------------------+------+-----+---------+----------------+
+  6 rows in set (0.00 sec)
+#+end_src
+```
+* For the *individual data*, the =sample= field would be saved as =Name= and the =strain= would be saved as =Name2=. These records would then all be linked to an inbredset group (population?) in the =InbredSet= table via the =StrainXRef= table, whose schema is as follows:
+```
+#+begin_src sql
+  MariaDB [db_webqtl]> DESC StrainXRef;
+  +------------------+----------------------+------+-----+---------+-------+
+  | Field            | Type                 | Null | Key | Default | Extra |
+  +------------------+----------------------+------+-----+---------+-------+
+  | InbredSetId      | smallint(5) unsigned | NO   | PRI | 0       |       |
+  | StrainId         | int(20)              | NO   | PRI | NULL    |       |
+  | OrderId          | int(20)              | YES  |     | NULL    |       |
+  | Used_for_mapping | char(1)              | YES  |     | N       |       |
+  | PedigreeStatus   | varchar(255)         | YES  |     | NULL    |       |
+  +------------------+----------------------+------+-----+---------+-------+
+  5 rows in set (0.00 sec)
+#+end_src
+```
+* Where the =InbredSetId= comes from the =InbredSet= table and the =StrainId= comes from the =Strain= table. The *individual data* would be linked to an inbredset group that is for individuals 
+* For the *average data*, the only value to save would be the =strain= field, which would be saved as =Name= in the =Strain= table and linked to an InbredSet group that is for averages
+*Question 01*: How do we distinguish the inbredset groups?
+*Answer*: The =Family= field is useful for this.
+
+*Question 02*: If you have more derived "datasets", e.g. males-only, females-only, under-10-years, 10-to-25-years, etc. How would the =Strains= table handle all those differences?
+
+## Metadata 
+* The data we looked at had =gene id= and =gene symbol= fields. These fields were used to fetch the *Ensembl ID* and *descriptions* from [[https://www.ncbi.nlm.nih.gov/][NCBI]] and the [[https://useast.ensembl.org/][Ensembl Genome Browser]]
+
+## Files for mapping 
+* Files used for mapping need to be in =bimbam= or =.geno= formats. We would need to do conversions to at least one of these formats where necessary
+
+## Annotation files 
+* Consider the following schema of DB tables 
+#+begin_src sql
+  MariaDB [db_webqtl]> DESC InbredSet;
+  +-----------------+----------------------+------+-----+---------+----------------+
+  | Field           | Type                 | Null | Key | Default | Extra          |
+  +-----------------+----------------------+------+-----+---------+----------------+
+  | Id              | smallint(5) unsigned | NO   | PRI | NULL    | auto_increment |
+  | InbredSetId     | int(5) unsigned      | NO   |     | NULL    |                |
+  | InbredSetName   | varchar(100)         | YES  |     | NULL    |                |
+  | Name            | char(30)             | NO   |     |         |                |
+  | SpeciesId       | smallint(5) unsigned | YES  |     | 1       |                |
+  | FullName        | varchar(100)         | YES  |     | NULL    |                |
+  | public          | tinyint(3) unsigned  | YES  |     | 2       |                |
+  | MappingMethodId | char(50)             | YES  |     | 1       |                |
+  | GeneticType     | varchar(255)         | YES  |     | NULL    |                |
+  | Family          | varchar(100)         | YES  |     | NULL    |                |
+  | FamilyOrder     | int(5)               | YES  |     | NULL    |                |
+  | MenuOrderId     | double               | NO   |     | NULL    |                |
+  | InbredSetCode   | varchar(5)           | YES  |     | NULL    |                |
+  | Description     | longtext             | YES  |     | NULL    |                |
+  +-----------------+----------------------+------+-----+---------+----------------+
+  ⋮
+  MariaDB [db_webqtl]> DESC Strain;
+  +-----------+----------------------+------+-----+---------+----------------+
+  | Field     | Type                 | Null | Key | Default | Extra          |
+  +-----------+----------------------+------+-----+---------+----------------+
+  | Id        | int(20)              | NO   | PRI | NULL    | auto_increment |
+  | Name      | varchar(100)         | YES  | MUL | NULL    |                |
+  | Name2     | varchar(100)         | YES  |     | NULL    |                |
+  | SpeciesId | smallint(5) unsigned | NO   |     | 0       |                |
+  | Symbol    | varchar(20)          | YES  | MUL | NULL    |                |
+  | Alias     | varchar(255)         | YES  |     | NULL    |                |
+  +-----------+----------------------+------+-----+---------+----------------+
+  ⋮
+  MariaDB [db_webqtl]> DESC StrainXRef;
+  +------------------+----------------------+------+-----+---------+-------+
+  | Field            | Type                 | Null | Key | Default | Extra |
+  +------------------+----------------------+------+-----+---------+-------+
+  | InbredSetId      | smallint(5) unsigned | NO   | PRI | 0       |       |
+  | StrainId         | int(20)              | NO   | PRI | NULL    |       |
+  | OrderId          | int(20)              | YES  |     | NULL    |       |
+  | Used_for_mapping | char(1)              | YES  |     | N       |       |
+  | PedigreeStatus   | varchar(255)         | YES  |     | NULL    |       |
+  +------------------+----------------------+------+-----+---------+-------+
+#+end_src
+
+* The =StrainXRef= table creates a link between the Samples/cases/individuals (stored in the =Strain= table) to the group (population?) they belong to in the =InbredSet= table
+* Steps to prepare the TSV file for entering samples/cases into the database are:
+- Clean up =Name= of the samples/cases/individuals in the file:
+  - Names should have no spaces
+  - Names should be the same length of characters: pad those that are shorter e.g. *SampleName12* → *SampleName012* to fit in with other names if, say, the samples range from 1 to 999. In a similar vein, you'd rename *SampleName1* to *SampleName001*
+- Order samples by the names
+- Create a new column, say, =orderId= in the TSV, and assign the order *1, 2, 3, …, n* for the rows, from the first to the "n^{th}" row. The order of the strains is very important and must be maintained
+- retrieve the largest current =Id= value  in the =Strain= table
+- Increment by one (1) and assign that to the first row of your ordered data
+  - Assign subsequent rows, the subsequent values for the ID e.g. Assuming the largest =Id= value in the =Strain= table was *23*, the first row of the new data would have the id *24*. The second row would have *25*, the third, *26* and so on
+- Get the =InbredSetId= for your samples' data. Add a new column in the data and copy this value for all rows
+- Enter data into the =Strain= table
+- Using the previously computed strain ID values, and the =InbredSetId= previously copied, enter data into the =StrainXRef= table
+
+* Some notes on the data:
+- The =Symbol= field in the =Strain= table corresponds to the =Strain= field in the annotation file
+- The =used_for_mapping= field should be set to ~Y~ unless otherwise informed
+- The =PedigreeStatus= field is unknown to us for now: set to ~NULL~
+
+* Annotation file format 
+The important fields are:
+- =ChipId=: The platform that the data was collected from/with
+Consider the following table; 
+#+begin_src sql
+    MariaDB [db_webqtl]> DESC GeneChip;
+    +---------------+----------------------+------+-----+---------+----------------+
+    | Field         | Type                 | Null | Key | Default | Extra          |
+    +---------------+----------------------+------+-----+---------+----------------+
+    | Id            | smallint(5) unsigned | NO   | PRI | NULL    | auto_increment |
+    | GeneChipId    | int(5)               | YES  |     | NULL    |                |
+    | GeneChipName  | varchar(200)         | YES  |     | NULL    |                |
+    | Name          | char(30)             | NO   |     |         |                |
+    | GeoPlatform   | char(15)             | YES  |     | NULL    |                |
+    | Title         | varchar(100)         | YES  |     | NULL    |                |
+    | SpeciesId     | int(5)               | YES  |     | 1       |                |
+    | GO_tree_value | varchar(50)          | YES  |     | NULL    |                |
+    +---------------+----------------------+------+-----+---------+----------------+
+  #+end_src
+ Some of the important fields that were highlighted were:
+ - =GeoPlatform=: Links the details of the platform in our database with NCBI's [[https://www.ncbi.nlm.nih.gov/geo/][Gene Ontology Omnibus (GEO)]] system. This is not always possible, but where we can, it would be nice to have this field populated
+ - =GO_tree_value=:  This is supposed to link the detail we have with some external system "GO". I have not figured this one out on my own and will need to follow up on it.
+ - =Name=: The name corresponds to the =ProbeSetId=, and we want this to be the same value as the identifier on the [[https://www.ensembl.org][Ensembl genome browser]], e.g. For a gene, say =Shh=, for *mouse*, we want the =Name= value to be a variation on [[https://useast.ensembl.org/Mus_musculus/Gene/Summary?db=core;g=ENSMUSG00000002633;r=5:28661813-28672254;t=ENSMUST00000002708][*ENSMUSG00000002633*]]
+ - =Probe_set_Blat_Mb_start=/=Probe_set_Blat_Mb_end=: In Byron's and Beni's data, these correspond to the =geneStart= and =geneEnd= fields respectively. These are the positions, in megabasepairs, that the gene begins and ends at, respectively.
+ - =Mb=: This is the =geneStart=/=Probe_set_Blat_Mb_start= value divided by *1000000*. (*Note to self*: Maybe the Probe_set_Blat_Mb_* fields above might not be in megabase pairs — please confirm)
+ - =Strand_Probe= and =Strand_Gene=: These fields' values are simply ~+~ or ~-~. If these values are missing, you can [[https://ftp.ncbi.nih.gov/gene/README][retrieve them from NCBI]], specifically from the =orientation= field of seemingly any text file with the field
+ - =Chr=: This is the chromosome on which the gene is found 
+
+* The final annotation file will have (at minimum) the following fields (or their
+analogs):
+- =StrainName=
+- =OrderId=
+- =StrainId=: from the database
+- =InbredSetId=: from the database
+- =Symbol=: This could be named =Strain=
+- =GeneChipId=: from the database
+- =EnsemblId=: from the Ensembl genome browser
+- =Probe_set_Blat_Mb_start=: possible analogue is =geneStart=
+- =Probe_set_Blat_Mb_end=: possible analogue is =geneEnd=
+- =Mb=
+- =Strand_Probe=
+- =Strand_Gene=
+- =Chr=
+
+*  =.geno= Files
+- The =.geno= files have sample names, not the strain/symbol. The =Locus= field in the =.geno= file corresponds to the **marker**. =.geno= files are used with =QTLReaper=
+- The sample names in the ~.geno~ files *MUST* be in the same order as the
+strains/symbols for that species. For example; 
+Data format is as follows; 
+```
+#+begin_example
+SampleName,Strain,…
+⋮
+BJCWI0001,BXD40,…
+BJCWI0002,BXD40,…
+BJCWI0003,BXD33,…
+BJCWI0004,BXD50,…
+BJCWI0005,BXD50,…
+⋮
+#+end_example
+```
+and the order of strains is as follows; 
+```
+#+begin_example
+…,BXD33,…,BXD40,…,BXD50,…
+#+end_example
+```
+then, the ~.geno~ file generated by this data should have a form such as shown
+below;
+```
+#+begin_example
+…,BJCWI0003,…,BJCWI0001,BJCWI0002,…,BJCWI0004,BJCWI0005,…
+#+end_example
+```
+The order of samples that belong to the same strain is irrelevant - they share the same data, i.e. the order below is also valid;
+```
+#+begin_example
+…,BJCWI0003,…,BJCWI0002,BJCWI0001,…,BJCWI0004,BJCWI0005,…
+#+end_example
+```
+* =BimBam= Files
+- Used with =GEMMA=
+*  Case Attributes
+- These are metadata about every case/sample/individual in an InbredSet group. The metadata is any data that has nothing to do with phenotypes (e.g. height, weight, etc) that is useful for researchers to have in order to make sense of the data.
+- Examples of case attributes:
+ - Treatment
+ - Sex (Really? Isn't sex an expression of genes?)
+ - batch
+ - Case ID, etc 
+
+* Summary steps to load data to the database 
+- [x] Create *InbredSet* group (think population)
+- [x] Load the strains/samples data
+- [x] Load the sample cross-reference data to link the samples to their
+ InbredSet group
+- Load the case-attributes data
+- [x] Load the annotation (data into ProbeSet table)
+- [x] Create the study for the data (At around this point, the InbredSet group
+  will show up in the UI).
+- [x] Create the Dataset for the data
+- [x] Load the *Log2* data (ProbeSetData and ProbeSetXRef tables)
+- [x] Compute means (an SQL query was used — this could be pre-computed in code
+  and entered along with the data)
+- [x] Run QTLReaper 
+
+
+
+