Develop a new data ingest / ETL pipeline for indexing eQTL data into the new mongo database

[karatugo]

We need to develop a robust and scalable data ingest/ETL (Extract, Transform, Load) pipeline to facilitate the reading of eQTL (expression Quantitative Trait Loci) data from FTP sources, indexing it into a MongoDB database, and serving it via an API. This pipeline will ensure efficient data extraction, transformation, and retrieval to support downstream analysis and querying through a web service.

• Scalable data ingest/ETL pipeline
• Read from FTP sources
• Ingest with the correct schema
• Save to MongoDB
• Index MongoDB - is it automatic? discuss with DBA team
• Deploy to Sandbox
• Fix the Sandbox bugs
• Update Confluence docs
• Discuss prod deployment with DBA team
• Deploy to Prod
• Plan for API implementation
• Implement API

[karatugo]

Files to Index

• QTD0000*.all.tsv.gz: Contains comprehensive eQTL data. This should be the primary source for indexing.
• QTD0000*.cc.tsv.gz: Contains specific eQTL data (likely condition-specific or subset). Also useful for indexing.
• QTD0000*.permuted.tsv.gz: Contains permuted eQTL data for significance testing. Useful for specific analyses but not primary indexing.

Suggested MongoDB Schema

Here's a refined schema to capture the necessary details from these files:

1. Study Information:

   • study_id: QTD000021
   • study_name: "Sample eQTL Study"

2. Sample Information:

   • sample_id: Auto-generated or derived from context if available?

3. eQTL Information:

   • molecular_trait_id: Corresponding trait ID.
   • molecular_trait_object_id: Object ID for the molecular trait.
   • chromosome: Chromosome number.
   • position: Position on the chromosome.
   • ref: Reference allele.
   • alt: Alternative allele.
   • variant: Variant identifier.
   • ma_samples: Minor allele sample count.
   • maf: Minor allele frequency.
   • pvalue: P-value of the association.
   • beta: Effect size.
   • se: Standard error.
   • type: Variant type (e.g., SNP).
   • aan: Additional annotation number.
   • r2: R-squared value.
   • gene_id: Gene identifier.
   • median_tpm: Median TPM (Transcripts Per Million).
   • rsid: Reference SNP ID.

4. Permuted eQTL Information:

   • p_perm: Permuted p-value.
   • p_beta: Permuted beta value.

Example MongoDB Document Structure

{
  "study_id": "QTD000021",
  "study_name": "Sample eQTL Study",
  "samples": [
    {
      "sample_id": "sample001",
      "eqtls": [
        {
          "molecular_trait_id": "ENSG00000187583",
          "molecular_trait_object_id": "ENSG00000187583",
          "chromosome": "1",
          "position": 14464,
          "ref": "A",
          "alt": "T",
          "variant": "chr1_14464_A_T",
          "ma_samples": 41,
          "maf": 0.109948,
          "pvalue": 0.15144,
          "beta": 0.25567,
          "se": 0.17746,
          "type": "SNP",
          "aan": 42,
          "r2": 382,
          "gene_id": "ENSG00000187583",
          "median_tpm": 0.985,
          "rsid": "rs546169444",
          "permuted": {
            "p_perm": 0.000999001,
            "p_beta": 3.3243e-12
          }
        }
      ]
    }
  ]
}

Steps to Implement

1. Extract Data:

   • Parse QTD0000*.all.tsv.gz and QTD0000*.cc.tsv.gz to extract eQTL data.
   • Parse QTD0000*.permuted.tsv.gz to extract permuted data and merge with the main eQTL data.

2. Transform Data:

   • Normalize data fields and structure according to the MongoDB schema.

3. Load Data:

   • Insert the structured documents into MongoDB.
   • Ensure appropriate indexes on fields such as gene_id, chromosome, position, and variant for efficient querying.

4. API Development:

   • Develop endpoints for querying the eQTL data based on different parameters.

Indexing Strategy

• Create indexes on key fields for efficient retrieval:
  • gene_id
  • chromosome
  • position
  • variant
  • rsid

[karatugo]

@karatugo Focus on Mongo indexing, deployment and API development

[karatugo]

Deployment to sandbox is in progress. I was able to run build step successfully. Deploy step has some errors at the moment. I'll prioritise this next week.

[karatugo]

Sandbox deployment worked with singularity commands but while automating I got the error below.

• Fix this error and test it in sandbox

FATAL:   could not open image /nfs/public/rw/gwas/deposition/singularity_cache/eqtl-sumstats-service_72de6563bdc84abc0be38ef294c854e3dd30f56e.sif: failed to retrieve path for /nfs/public/rw/gwas/deposition/singularity_cache/eqtl-sumstats-service_72de6563bdc84abc0be38ef294c854e3dd30f56e.sif: lstat /nfs/public: no such file or directory

[karatugo]

Fixed the above error, now working on mongo save failed issue.

[karatugo]

Deployment to sandbox complete.

[karatugo]

Started a full ingestion yesterday evening. In 16h, with 2 concurrent workers only 2 studies/19 datasets were complete.

• Need to adjust accordingly:

  • number of workers
  • sbatch wallclock time
  • sbatch memory

  I'll wait until the ingestion is complete to see what we get at the end of 2 days with 2 workers and 8G mem.

[karatugo]

Sent an email to Kaur for the schemas of .permuted files.

[karatugo]

• Ignored .permuted files
• Used local file system rather than ftp protocol
  This is due to bugfixing in sandbox. There were many ftp connection problems.

[karatugo]

• fixed docker pull rate error in ingest script (docker-login needed)

[karatugo]

• fixed an issue with file last modified date in local fs, demo ingest is running

[karatugo]

Started a full ingestion yesterday evening. In 16h, with 2 concurrent workers only 2 studies/19 datasets were complete.

* [ ]  Need to adjust accordingly:
  
  * number of workers
  * sbatch wallclock time
  * sbatch memory
  
  I'll wait until the ingestion is complete to see what we get at the end of 2 days with 2 workers and 8G mem.

• Run 8 concurrent workers with 64G for 2 days

[karatugo]

* [ ]  Run 8 concurrent workers with 64G for 2 days

Running, will check on monday.

[karatugo]

I realized that there's a typo in memory, it should be 64G rather than 6G. Restarted.