Single-Cell Organoid Batch Remediation Agent with CrewAI
NIH discoveries on brain organoid developmental clocks emphasize that single-cell RNA transcriptomic datasets are highly sensitive to technical batch effects across organoid differentiation runs. In this challenge, developers will construct a CrewAI agent team (Batch Inspection Agent, Normalization Specialist, Harmonization Evaluator) that automates single-cell batch effect detection, runs Combat/Harmony integration scripts, and verifies dataset alignment.
What you are building
Build a CrewAI multi-agent team to automate single-cell RNA-seq batch effect detection and harmonization.
How work is evaluated
Evaluates batch effect reduction quality and silhouette score improvement.
Datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
check_approval_status
Ensure corrected silhouette < 0.15 leads to APPROVED status
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
silhouette_reduction
Magnitude of batch silhouette score reduction post-harmonization • target: 0.35 • range: 0-1
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
Learning goals
Configure CrewAI multi-agent collaborative workflows for bioinformatics pipelines
Execute Scanpy and Harmony integration tools in Python from agent tasks
Compute quantitative batch silhouette metrics (kBET, silhouette width)
Automate decision workflows for data harmonization approval
Resources
Synthetic multi-batch organoid scRNA-seq expression tables exhibiting engineered technical batch variation.
Challenge setup
Evaluates batch effect reduction quality and silhouette score improvement.
Challenge input
JSON containing batch1_id, batch2_id, raw_batch_silhouette
CrewAI
Mandatory CrewAI framework
Evaluated output
JSON containing corrected_silhouette (float), harmonization_status (APPROVED/REJECTED)
- Ensure corrected silhouette < 0.15 leads to APPROVED status
- Magnitude of batch silhouette score reduction post-harmonization • target: 0.35 • range: 0-1
- Benchmark: scHarmonize-Bench
- Silhouette Reduction target: 0.35
- 1 public reference case
- Python execution harness
- Python sandbox (unavailable on Versalist)
Shows the first task, up to three tools, and up to three evidence items.
View technical recipe
Configured tools
- CrewAI · Required
- crewAI · Optional
- Optuna · Optional
Evaluation contract
- check_approval_status · Weight 1
- silhouette_reduction · Weight 1
Metrics
- silhouette_reduction · Target 0.35
Benchmarks
- scHarmonize-Bench
- biotech
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON containing batch1_id, batch2_id, raw_batch_silhouette
Output: JSON containing corrected_silhouette (float), harmonization_status (APPROVED/REJECTED)
Execution harness: Python
Evaluation tools
- CrewAI · Agent Framework
Recipe freeze marker
No recipe freeze time is recorded.