Jgi Lakehouse

JGI Lakehouse Skill

Quick Start

What is it? JGI's unified data warehouse (651 tables) + filesystem access to genome files.

Two data access methods:

1. Lakehouse (Dremio) → Metadata, annotations, taxonomy (no sequences)
2. JGI Filesystem → Actual genome files (FNA, FAA, GFF) via taxon OID

SQL Dialect: ANSI SQL (not PostgreSQL)

• Use CAST(x AS type) not ::
• Use REGEXP_LIKE() not ~
• Identifiers with dashes need double quotes: "gold-db-2 postgresql"

-- Quick test
SELECT gold_id, project_name FROM "gold-db-2 postgresql".gold.project
WHERE is_public = 'Yes' LIMIT 5;

When to Use

• Query JGI genomics metadata (GOLD, IMG, Mycocosm, Phytozome)
• Find genomes and/or metagenomes by taxonomy, ecosystem, or phenotype.
• Download microbial genomes with IMG taxon OIDs
• Cross-reference GOLD projects with IMG annotations

Best practices

• When reporting results from database queries, always provide a clear summary of the exact criteria that were used for filtering. This should include a list of each field that was used in the query/filter, and the query/filter was applied (string used in exact match, regular expression, exact number searched for or range, etc)

Data Access: Lakehouse vs Filesystem

Critical insight: The Lakehouse is a METADATA warehouse. Genome sequences must be accessed from the JGI filesystem.

Key Data Sources

Full table catalog: See docs/data-catalog.md

Phytozome (plant-db-7 / plant-db-4): Read docs/phytozome.md before writing any queries against these sources.

NUMG (Metagenome Proteins) Agent Workflow

Use NUMG when the task is metagenome protein sequence/domain analysis.

Scope rules:

• numg-iceberg is metagenome-focused.
• Do not use NUMG for isolate genome protein retrieval; use IMG filesystem packages.

Core tables:

• "numg-iceberg"."numg-iceberg".faa
  • oid, gene_oid, faa (protein sequence)
• "numg-iceberg"."numg-iceberg".gene2pfam
  • oid, gene_oid, pfam, evalue, alignment coordinate fields

Recommended query flow:

-- 1) Confirm available NUMG tables
SHOW TABLES IN "numg-iceberg"."numg-iceberg";

-- 2) Inspect schema before writing joins/filters
DESCRIBE "numg-iceberg"."numg-iceberg".faa;
DESCRIBE "numg-iceberg"."numg-iceberg".gene2pfam;

-- 3) Domain filter (use exact lowercase pfam IDs)
SELECT oid, gene_oid, pfam, evalue
FROM "numg-iceberg"."numg-iceberg".gene2pfam
WHERE pfam IN ('pfam00001', 'pfam00004')
LIMIT 100;

-- 4) Join domains to protein sequences
SELECT
  p.oid,
  p.gene_oid,
  p.pfam,
  p.evalue,
  f.faa
FROM "numg-iceberg"."numg-iceberg".gene2pfam p
JOIN "numg-iceberg"."numg-iceberg".faa f
  ON p.oid = f.oid
 AND p.gene_oid = f.gene_oid
WHERE p.pfam = 'pfam00001'
LIMIT 100;

Important NUMG rules:

• Join on both oid and gene_oid (not gene_oid alone).
• Keep Pfam filters exact (pfam00001, not case-transformed).
• Always start with LIMIT and expand only after verifying row shape.

See also: examples/05-query-numg-metagenome-proteins.md

Downloading Genomes with IMG Taxon OIDs

Option 1: JGI Filesystem (Fastest)

# Genome packages are at:
/clusterfs/jgi/img_merfs-ro/img_web/img_web_data/download/{taxon_oid}.tar.gz

# Example: Copy and extract
cp /clusterfs/jgi/img_merfs-ro/img_web/img_web_data/download/8136918376.tar.gz .
tar -xzf 8136918376.tar.gz

Package contents:

• {taxon_oid}.fna - Genome assembly
• {taxon_oid}.genes.faa - Protein sequences
• {taxon_oid}.genes.fna - Gene nucleotide sequences
• {taxon_oid}.gff - GFF annotations
• {taxon_oid}.cog.tab.txt - COG annotations
• {taxon_oid}.pfam.tab.txt - Pfam annotations
• {taxon_oid}.ko.tab.txt - KEGG KO annotations

Linking Assemblies To Reads

Use this workflow when you need to go from an IMG metagenome assembly to the underlying JGI reads.

1. Start from the assembly taxon OID

Assemblies live under:

/clusterfs/jgi/img_merfs-ro/img_web_data_merfs/{taxon_oid}/assembled/

2. Pull the JGI/GOLD linkage fields from metadata

For a metagenome taxon OID, the most useful linkage fields are:

• img_jgi_project_id
• sequencing_gold_id
• sample_gold_id
• study_gold_id
• gold_project_id
• gold_pmo_project_id
• gold_its_spid

In practice, img_jgi_project_id is often the strongest key for JAMO because it behaves like the PMO project identifier used by jamo info ... pmoid.

3. Prefer JAMO pmoid for JGI read lookup

Native JAMO lookup types are listed by:

apptainer run docker://doejgi/jamo-dori:latest jamo info help

For legacy JGI metagenomes, this usually works better than raw_normal spid:

apptainer run docker://doejgi/jamo-dori:latest \
  jamo info all pmoid <img_jgi_project_id>

If you only want FASTQ rows, filter the output:

apptainer run docker://doejgi/jamo-dori:latest \
  jamo info all pmoid <img_jgi_project_id> | rg 'fastq(\\.gz)?'

4. Direct taxon-OID lookup is still useful

This queries JAMO by the IMG taxon OID embedded in metadata:

apptainer run docker://doejgi/jamo-dori:latest \
  jamo info all custom '{"metadata.gold_data.img_oid": 3300000030, "file_name": {"$regex": ".*fastq(\\\\.gz)?$"}}'

This can recover reads even when the older spid route is blank, but in recent re-audits pmoid recovered many more JGI rows.

5. spid is valid, but not sufficient

If you already have a verified sequencing project ID, this is still worth trying:

apptainer run docker://doejgi/jamo-dori:latest \
  jamo info raw_normal spid <gold_its_spid>

But do not stop there. In several JGI cases:

• raw_normal spid returned nothing
• all pmoid <img_jgi_project_id> returned usable FASTQ records

6. Inspect and fetch the actual file

Inspect one metadata record:

apptainer run docker://doejgi/jamo-dori:latest jamo show <metadata_id>

Fetch a file by filename:

apptainer run docker://doejgi/jamo-dori:latest \
  jamo fetch -s dori all filename <file_name>

That prints the staged scratch path, typically under:

/clusterfs/jgi/scratch/dsi/...

Important:

• if the file is already RESTORED, you can use the staged path immediately
• if the file is PURGED, jamo fetch only starts the restore; you must wait until the staged path exists and has non-zero size before using it

Simple wait pattern:

while [[ ! -s /clusterfs/jgi/scratch/dsi/.../file.fastq.gz ]]; do sleep 10; done

Practical rule

For JGI metagenome read recovery, use this priority:

1. jamo info all pmoid <img_jgi_project_id>
2. jamo info all custom '{"metadata.gold_data.img_oid": ...}'
3. jamo info raw_normal spid <gold_its_spid>

Do not assume "no reads" until all three have been checked. Do not assume a fetched file is ready until the staged path is actually restored.

Portal Downloads (Mycocosm / Phytozome)

The portal tracks downloadable files for Mycocosm and Phytozome in "portal-db-1".portal.downloadRequestFiles. Use filePath to copy data from the JGI filesystem (/global/dna/dm_archive/...).

Mycocosm (fungal genomes/proteins):

SELECT filePath, fileType
FROM "portal-db-1".portal.downloadRequestFiles
WHERE LOWER(filePath) LIKE '%mycocosm%'
  AND (filePath LIKE '%.fasta%' OR filePath LIKE '%.fa%' OR filePath LIKE '%.faa%')
LIMIT 20;

Phytozome (plant genomes/proteins):

SELECT filePath, fileType
FROM "portal-db-1".portal.downloadRequestFiles
WHERE LOWER(filePath) LIKE '%phytozome%'
  AND (filePath LIKE '%.fa%' OR filePath LIKE '%.fna%' OR filePath LIKE '%.faa%')
LIMIT 20;

Download from filesystem:

cp /global/dna/dm_archive/<path/from-filePath> .

Notes:

• fileType typically includes Assembly, Annotation, or Sequence.
• virtualPath can provide a user-facing download label but filePath is the real location.

Query Best Practices

⚠️ CRITICAL: When building queries, distinguish between exploration and comprehensive analysis:

Exploration Queries

Use LIMIT for quick validation during development:

-- For testing query structure and results
SELECT gold_id, project_name
FROM "gold-db-2 postgresql".gold.project
WHERE is_public = 'Yes'
LIMIT 10;  -- ✓ OK for testing

Comprehensive Queries

Remove LIMIT and other result-limiting clauses when answering actual questions:

-- For getting actual dataset counts/results
SELECT COUNT(DISTINCT taxon_oid)
FROM "img-db-2 postgresql".img_core_v400.taxon
WHERE genome_type = 'metagenome'
  AND is_public = 'Yes';
-- ✓ No LIMIT - gets true total

Common pitfalls:

• ❌ LIMIT 100 on initial exploration → assumes only 100 results exist
• ❌ LIMIT 50 on a "find all" query → misses 99% of data
• ❌ Using FETCH FIRST N ROWS → same issue as LIMIT

Best practice:

1. Use LIMIT with COUNT(*) or small LIMIT during development
2. Once query logic is correct, remove LIMIT to get true results
3. For very large result sets, use aggregation (COUNT, GROUP BY) to summarize instead

Common Queries

Find Bacterial Isolate Genomes

-- Get count of all finished bacterial isolates
SELECT COUNT(DISTINCT taxon_oid) as total_isolates
FROM "img-db-2 postgresql".img_core_v400.taxon
WHERE domain = 'Bacteria'
  AND genome_type = 'isolate'
  AND is_public = 'Yes'
  AND seq_status = 'Finished';

-- Get sample of isolates (if you need details)
SELECT taxon_oid, taxon_display_name, phylum, genus, species
FROM "img-db-2 postgresql".img_core_v400.taxon
WHERE domain = 'Bacteria'
  AND genome_type = 'isolate'
  AND is_public = 'Yes'
  AND seq_status = 'Finished'
LIMIT 100;

Link GOLD Project to IMG Taxon

SELECT COUNT(DISTINCT t.taxon_oid) as total_linked
FROM "img-db-2 postgresql".img_core_v400.taxon t
WHERE t.sequencing_gold_id IS NOT NULL;

Find Genomes with File Paths (Portal)

SELECT COUNT(DISTINCT taxonOid) as total_tar_gz
FROM "portal-db-1".portal.downloadRequestFiles
WHERE taxonOid IS NOT NULL
  AND filePath LIKE '%.tar.gz';

Critical Pitfalls

Authentication

export DREMIO_PAT=$(cat ~/.secrets/dremio_pat)

Token setup: See docs/authentication.md

API Access

REST API Base: http://lakehouse-1.jgi.lbl.gov:9047/api/v3

# Use scripts/rest_client.py
from rest_client import query
results = query("SELECT * FROM ... LIMIT 10")

Arrow Flight (Python)

For higher-performance programmatic access, use Arrow Flight with Python.

python3 -m venv venv
. venv/bin/activate
pip install \
  https://github.com/dremio-hub/arrow-flight-client-examples/releases/download/dremio-flight-python-v1.1.0/dremio_flight-1.1.0-py3-none-any.whl

Full guide: docs/arrow-flight-python.md

Documentation

IMG Reference

• docs/img_and_gold_terms.md - IMG/GOLD glossary: all terms, project types, quality flags, IDs, sequencing status, taxonomy systems
• docs/IMG_data_types.md - Guide to analysis project types (isolates, MAGs, SAGs, metagenomes) with query patterns and data type counts
• docs/IMG-tables-reference.md - Complete IMG table catalog (244 tables in img_core_v400)
• docs/explore_IMG_genomes.md - Genome metadata queries: NCBI/GTDB taxonomy, genome size, GC content, quality filters

Lakehouse Catalog & SQL

• docs/data-catalog.md - All data sources and key tables (GOLD, IMG, Portal, Mycocosm, Phytozome, NUMG)
• docs/phytozome.md - Phytozome plant genomics: proteomes, genes, families, PFAM/PANTHER/GO, expression, synteny, homologs
• docs/sql-quick-reference.md - Dremio SQL syntax

Access & Downloads

• docs/arrow-flight-python.md - Arrow Flight Python setup and test
• examples/05-query-numg-metagenome-proteins.md - NUMG workflow for protein+Pfam queries
• examples/04-download-img-genomes.md - Download with taxon OIDs
• scripts/download_img_genomes.py - Automated download script