Accelerating Obesity Drug Lead Optimization
Inspired by significant advancements in obesity drug development, this challenge focuses on building an AI-driven system to optimize lead compounds for metabolic disorders. Participants will create a multi-agent AI framework that simulates different stages of drug discovery, from target interaction prediction to ADMET property assessment, with the goal of identifying novel drug candidates or improving existing ones. The emphasis is on developing compounds with enhanced efficacy, reduced off-target effects, and favorable pharmacokinetic profiles. This project requires integrating computational chemistry techniques with advanced AI, including generative models for molecular design and multi-agent systems for iterative refinement. Candidates will develop a pipeline that screens virtual libraries, predicts protein-ligand binding affinities, assesses ADMET (Absorption, Distribution, Metabolism, Excretion, Toxicity) properties, and recommends structural modifications. The outcome will be a refined set of lead compounds with improved predicted properties, showcasing the power of AI in accelerating drug discovery.
What you are building
The core problem, expected build, and operating context for this challenge.
Optimize obesity drug leads using generative AI, multi-agent systems, and a vector database for molecular properties.
How work is evaluated
Participants will submit their code, a detailed report on the multi-agent system architecture, the iterative optimization process, and the top 10 optimized lead compounds with their predicted properties. Evaluation will prioritize the robustness of the multi-agent framework, the novelty and quality of the generated compounds, and the predictive accuracy of the integrated models.
Challenge datasets
Public datasets, plus the private datasets you have access to.
What you should walk away with
Implement generative models to propose molecular modifications.
Utilize a vector database for efficient molecular data management and similarity search.
Integrate predictive models for binding affinity and ADMET properties.
Evaluate the proposed compounds based on desired therapeutic criteria.
Design a multi-agent AI system for lead compound optimization in drug discovery.
Reference links and supporting material
Sample data for 3 tasks
Challenge setup
Participants will submit their code, a detailed report on the multi-agent system architecture, the iterative optimization process, and the top 10 optimized lead compounds with their predicted properties. Evaluation wi...
Challenge input
JSON with 'target_protein_id': string, 'initial_lead_SMILES': string.
Agent execution
No agent tool is listed for this challenge.
Evaluated output
JSON with 'agents_initialized': boolean, 'initial_dialogue_summary': string.
- Verifies that agents communicate effectively and roles are clearly defined and executed.
- Checks for iterative improvement in target properties across optimization steps.
- Ensures binding affinity and ADMET models are correctly integrated and yield plausible predictions.
- Best Binding Affinity target: -9.5
- Python execution harness
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
No tool records are attached.
Evaluation contract
- Verifies that agents communicate effectively and roles are clearly defined and executed.
- Checks for iterative improvement in target properties across optimization steps.
- Ensures binding affinity and ADMET models are correctly integrated and yield plausible predictions.
- Confirms molecular data and embeddings are accurately stored and retrievable from Pinecone.
Metrics
- Best_Binding_Affinity · Target -9.5
- ADMET_Compliance_Rate · Target 0.8
- Synthetic_Accessibility_Score · Target 2.5
- Novelty_Score · Target 0.6
Workflow tasks
- Task 1
Input: JSON with 'target_protein_id': string, 'initial_lead_SMILES': string.
Output: JSON with 'agents_initialized': boolean, 'initial_dialogue_summary': string.
Execution harness: Python
- Task 2
Input: JSON with 'current_lead_SMILES': string, 'iteration_number': integer.
Output: JSON array of objects, each with 'SMILES', 'binding_affinity_score', 'ADMET_profile': dictionary.
Execution harness: Python
- Task 3
Input: JSON array of objects, each with 'SMILES', 'embedding': array of floats, 'properties': dictionary, 'query_SMILES': string for similarity search.
Output: JSON array of objects, each with 'SMILES', 'score': float (similarity).
Execution harness: Python
Recipe freeze marker
No recipe freeze time is recorded.