Back to evaluations
Public evaluation

InitialPigmentGeneration

The AI agent's performance will be evaluated on its ability to propose novel, sustainable sunscreen pigment molecules that meet specific UV absorption criteria, stability, and biodegradability. The evaluation will assess the pipeline's automation, the quality of generated structures, and the accuracy of property predictions.

Evaluation type
task based
Challenge
AI-Accelerated Sustainable Material Design for Bio-Inspired Sunscreens
Difficulty
Advanced
Rigor
Unspecified

Evaluation overview

How the linked challenge is judged: tasks, benchmarks, and criteria count.

Tasks
3
Benchmarks
0
Criteria
0

Task templates

Inputs and expected outputs.

Task 1

InitialPigmentGeneration

Generate an initial set of 5-10 diverse molecular structures representing potential bio-inspired sunscreen pigments using Claude Opus 4.1.

Input format

JSON object with target properties (e.g., {"uv_absorption_peak_nm": 320, "biodegradability_score_target": 0.8})

Output format

JSON array of SMILES strings and initial predicted properties for generated molecules.

Task 2

PropertyPredictionAndRefinement

For a given set of generated SMILES strings, use Fireworks to simulate and predict UV absorption maxima, quantum yield, and an estimated biodegradability score. Then, use AgentFlow and Claude Opus 4.1 to refine the top 3 candidates based on these predictions, aiming for optimized properties.

Input format

JSON array of SMILES strings from initial generation.

Output format

JSON array of refined SMILES strings, each with comprehensive predicted properties (UV abs, stability, biodegradability) and a justification for refinement.

Task 3

MultiObjectiveOptimization

Implement a multi-objective optimization routine (e.g., using a genetic algorithm or similar heuristic) to find a final 'best' candidate that balances UV absorption, stability, and biodegradability. The agent should present the top optimized molecule.

Input format

JSON array of refined SMILES strings and their properties.

Output format

JSON object representing the single best optimized molecule, including its SMILES string, all predicted properties, and a rationale for its selection.