Missense Variant Triage Agent with OpenAI Agents SDK and Google Jules
In response to emerging missense functional prediction tools, genomic diagnostic labs require automated pipelines to classify novel variants. Build an autonomous triage agent using OpenAI Agents SDK integrated with Google Jules for automated code patch generation and Biopython validation. Achieve 92% classification accuracy on ClinVar benchmarking sets.
What you are building
The core problem, expected build, and operating context for this challenge.
Build an automated missense variant triage workflow using OpenAI Agents SDK and Biopython for clinical genomic annotation.
How work is evaluated
Evaluates the agent's ability to classify missense variants and produce valid HGVS variant annotations.
Challenge datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
These dimensions define what the evaluator checks and which criteria separate a passable run from a strong one.
hgvs_syntax_check
Ensures HGVS transcript notation follows standard nomenclature.
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
classification_accuracy
Accuracy of pathogenic vs benign variant categorization • target: 0.92 • range: 0-1
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
What you should walk away with
Implement autonomous variant annotation tool workflows using OpenAI Agents SDK
Integrate Google Jules for automated execution of bioinformatics script updates
Parse and validate VCF and ClinVar data streams using Biopython
Evaluate variant pathogenicity scores using multi-agent verification checks
Reference links and supporting material
Curated set of 500 missense single-nucleotide variants with gold-standard ClinVar pathogenicity labels and functional prediction scores.
Challenge setup
Evaluates the agent's ability to classify missense variants and produce valid HGVS variant annotations.
Challenge input
JSON containing a list of genomic variant objects with chromosome, position, ref, and alt alleles.
OpenAI Agents SDK
Required framework for building multi-turn tool-using agents.
Google Jules
Automated code modification and tool fixing during pipeline failure.
Evaluated output
JSON object with classified pathogenic status, score, and verified HGVS nomenclature string.
- Ensures HGVS transcript notation follows standard nomenclature.
- Accuracy of pathogenic vs benign variant categorization • target: 0.92 • range: 0-1
- Benchmark: ClinVar Variant Pathogenicity Benchmark
- Classification Accuracy target: 0.92
- 1 public reference case
- Python execution harness
- Python sandbox (unavailable on Versalist)
Summary shows the first task, up to three tools, and up to three evidence items. Open the technical recipe to view all configured items.
View technical recipe
Configured tools
- Google Jules · Required
- OpenAI · Optional
- Agno · Optional
- Google Jules · Required
- Agno · Optional
Evaluation contract
- hgvs_syntax_check · Weight 1
- classification_accuracy · Weight 1
Metrics
- classification_accuracy · Target 0.92
Benchmarks
- ClinVar Variant Pathogenicity Benchmark
- biotech
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON containing a list of genomic variant objects with chromosome, position, ref, and alt alleles.
Output: JSON object with classified pathogenic status, score, and verified HGVS nomenclature string.
Execution harness: Python
Evaluation tools
- OpenAI Agents SDK · Agent Orchestration
- Google Jules · AI Code Maintenance
Recipe freeze marker
No recipe freeze time is recorded.