Missense Variant Pathogenicity Triage Agent with OpenAI Agents SDK
Interpreting missense variants requires synthesizing clinical literature and structural pathogenicity predictions. Build a multi-step variant triage agent using OpenAI Agents SDK and DeepSeek R1 to classify genomic variants and generate validated ACMG criteria classifications for patient diagnostic pipelines.
What you are building
The core problem, expected build, and operating context for this challenge.
Develop an autonomous variant triage agent using OpenAI Agents SDK backed by DeepSeek R1 reasoning to evaluate genomic missense functional impacts.
How work is evaluated
Evaluates agent correctness in mapping genomic variants to ACMG criteria and final pathogenicity classifications.
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.
ACMG Code Format Validation
Checks that returned ACMG criteria conform to standard terminology.
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
Classification Accuracy
Accuracy of pathogenic/benign assignments against ClinVar ground truth. • 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 multi-turn genomic agent workflows using OpenAI Agents SDK.
Incorporate DeepSeek R1 chain-of-thought capabilities for ACMG evidence criteria synthesis.
Integrate missense functional prediction scores (e.g., REVEL, AlphaMissense) into decision tool calls.
Produce standardized VCF annotation and clinical pathogenicity outputs.
Reference links and supporting material
Curated set of 500 missense variants with annotated REVEL scores, gnomAD allele frequencies, and ClinVar ground truth classifications.
Challenge setup
Evaluates agent correctness in mapping genomic variants to ACMG criteria and final pathogenicity classifications.
Challenge input
JSON with variant chr, position, ref, alt, gene, and functional scores
OpenAI Agents SDK
Provides function-calling agent orchestration primitives.
DeepSeek R1
Delivers strong chain-of-thought clinical evidence synthesis.
Evaluated output
JSON with classification, applied_acmg_codes, and reasoning_summary
- Checks that returned ACMG criteria conform to standard terminology.
- Accuracy of pathogenic/benign assignments against ClinVar ground truth. • target: 0.92 • range: 0-1
- Benchmark: ClinVar-Eval
- 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
- OpenAI · Required
- BLACKBOX AI · Optional
- DeepSeek R1 · Optional
Evaluation contract
- ACMG Code Format Validation · Weight 1
- Classification Accuracy · Weight 1
Metrics
- Classification Accuracy · Target 0.92
Benchmarks
- ClinVar-Eval
- biotech
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON with variant chr, position, ref, alt, gene, and functional scores
Output: JSON with classification, applied_acmg_codes, and reasoning_summary
Execution harness: Python
Evaluation tools
- OpenAI Agents SDK · Agent Framework
- DeepSeek R1 · Reasoning LLM
Recipe freeze marker
No recipe freeze time is recorded.