backend

Backend

FastAPI REST API — the central orchestrator for patient registration, pipeline management, semantic search, and drug discovery.

What It Does

• Registers patients and samples in VastDB; validates FASTQ source files in S3; copies FASTQ to the controlled path (genomics-fastq-files/{patient_id}/{sample_id}/)
• Submits Kubernetes Jobs for Parabricks / mock compute and runs a background job_watcher that polls K8s and syncs job status to VastDB
• Executes vector similarity search via array_cosine_distance (ADBC driver, server-side) with automatic clinical significance re-ranking
• Synthesizes search results into clinical insights using NVIDIA Llama 3.1 Nemotron 70B
• Full drug discovery pipeline: PubChem SMILES lookup → NVIDIA MolMIM molecule generation → RCSB PDB structure search → NVIDIA DiffDock protein-ligand docking → results and researcher annotations cached in VastDB

API Routes

Search & Insights

Route	Description
POST /api/v1/search	Semantic variant search (+ optional LLM synthesis)
POST /api/v1/search/explain	LLM plain-language explanation of a single variant
POST /api/v1/search/insights	Personalized clinical insights for a variant + patient context
GET /api/v1/stats	Platform statistics (patients, variants, cache hits, compute hours)

Drug Discovery

Route	Description
POST /api/v1/search/molecules	Generate novel molecules via NVIDIA MolMIM from a drug name or SMILES string; results cached in VastDB
GET /api/v1/search/structures/{gene}	Search RCSB PDB for protein structures associated with a gene
POST /api/v1/search/dock	Dock a generated molecule against a PDB structure via NVIDIA DiffDock; result cached in VastDB
GET /api/v1/search/molecules/all	List all generated molecules across all variants
GET /api/v1/search/molecules/{variant_id}	Molecules generated for a specific variant
POST /api/v1/search/molecules/{id}/annotate	Append a researcher annotation or status change to a molecule

Patients & Pipelines

Route	Description
POST /api/v1/auth/login	JWT authentication
POST /api/v1/register	Register patient + sample, copy FASTQ, trigger pipeline
GET /api/v1/patients	List all patients
GET /api/v1/patients/{id}	Patient details with samples
GET /api/v1/patients/{id}/variants	Patient's variant records
GET /api/v1/pipelines	All pipeline runs
GET /api/v1/pipelines/{id}	Pipeline detail with K8s logs
GET /health	Health check with processing mode

Drug Discovery Flow

Variant (gene) selected in UI
    → POST /search/molecules   drug_name or SMILES → PubChem lookup → MolMIM (CMA-ES, QED optimization)
                                                                      → novel SMILES + Tanimoto scores
                                                                      → saved to VastDB molecules table
    → GET  /search/structures/{gene}               RCSB PDB search  → ranked protein structures
    → POST /search/dock        molecule_id + pdb_id → DiffDock        → docking poses + confidence scores
                                                                      → best pose SDF + protein PDB cached
    → POST /search/molecules/{id}/annotate          researcher notes / status changes persisted

Both MolMIM results and DiffDock results are cached in VastDB — re-running the same seed SMILES or the same molecule+PDB pair returns the cached result instantly.

Easy to Adjust

All configuration comes from deployments/genomics-k8s-application/values.yaml, mounted as a K8s Secret at /etc/secrets/config.yaml:

Section	Key Settings
vast	access_key, secret_key
s3	endpoint
vdb	endpoint, bucket, schema
nvidia	use_api_catalog, api_key
embedding	host, port, model, dimensions
llm	host, port, model
processing_mode	mock or gpu

What Runs It

• Runtime: Kubernetes deployment in the genomics namespace
• Image: <your-registry>/genomic-engine-backend:<tag>
• Framework: FastAPI (Python 3.11), Uvicorn
• VastDB access: Python SDK for CRUD; ADBC driver (libadbc_driver_vastdb.so) for vector search
• Dependencies: fastapi, vastdb, adbc-driver-manager, boto3, kubernetes, openai