Skip to main content
Back to challenges
Challenge

Genomic Variant Pathogenicity Triage Pipeline with OpenAI Agents SDK

Inspired by the appointment of Dr. Judy Cho as NHGRI director and recent advances in genomic medicine, this challenge centers on building an automated variant interpretation engine. Participants will use the OpenAI Agents SDK to parse VCF (Variant Call Format) files, cross-reference ClinVar and gnomAD databases via function calling, and produce structured ACMG/AMP pathogenicity classifications with complete audit tracking.

Data ScienceHosted by Vera
Sign in free to start
Challenge brief

What you are building

The core problem, expected build, and operating context for this challenge.

Build an ACMG-compliant genomic variant triage pipeline using OpenAI Agents SDK and clinical genetic databases.

Delivery guide

How work is evaluated

Evaluation

Evaluates variant classification accuracy and adherence to ACMG scoring codes.

Datasets

Challenge datasets

Public datasets, plus the private datasets you have access to.

Evaluation rubric

How submissions are scored

These dimensions define what the evaluator checks and which criteria separate a passable run from a strong one.

Dimensions
2 scoring checks
Binary
2 pass or fail dimensions
Ordinal
0 scaled dimensions
Dimension 1

valid_acmg_class

Classification must be one of standard 5 ACMG classes

Binary check

This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.

Dimension 2

classification_accuracy

Accuracy against ClinVar validated benchmark set • target: 0.9 • range: 0-1

Binary check

This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.

Learning goals

What you should walk away with

  • Parse multi-sample Variant Call Format (VCF) files using Python genomic libraries

  • Build dynamic function-calling tools using OpenAI Agents SDK for ClinVar/gnomAD APIs

  • Implement ACMG/AMP variant interpretation rule logic (e.g., PVS1, PM2, PP3)

  • Output verifiable pathogenicity reports with clinical human-in-the-loop checkpoints

Resources and assets

Reference links and supporting material

Dataset notes

VCF snippet files and JSON variant records annotated with ClinVar clinical assertions and gnomAD population frequencies.

Challenge setup

Evaluates variant classification accuracy and adherence to ACMG scoring codes.

Challenge input

JSON containing chr, pos, ref, alt, gene, gnomad_af

OpenAI Agents SDK

Mandatory OpenAI framework

Evaluated output

JSON containing classification (Pathogenic/Benign/VUS) and acmg_criteria (list)

Evaluation checks
  • Classification must be one of standard 5 ACMG classes
  • Accuracy against ClinVar validated benchmark set • target: 0.9 • range: 0-1
Configured evaluation evidence
  • Benchmark: VarBench
  • Classification Accuracy target: 0.9
  • 1 public reference case
Configured runtime
  • 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

Action Space
  • OpenAI · Required
  • 01.AI · Optional
  • Airweave · Optional

Evaluation contract

  • valid_acmg_class · Weight 1
  • classification_accuracy · Weight 1

Metrics

  • classification_accuracy · Target 0.9

Benchmarks

  • VarBench
  • biotech

Reference cases

  • 1 public reference case

Workflow tasks

  1. Task 1

    Input: JSON containing chr, pos, ref, alt, gene, gnomad_af

    Output: JSON containing classification (Pathogenic/Benign/VUS) and acmg_criteria (list)

    Execution harness: Python

Evaluation tools

  • OpenAI Agents SDK · Agent Framework

Recipe freeze marker

No recipe freeze time is recorded.

Frequently Asked Questions about Genomic Variant Pathogenicity Triage Pipeline with OpenAI Agents SDK