Generative AI-Driven Optimization for Novel Multi-Target Drug Candidates
Develop an intelligent drug discovery platform capable of generating and optimizing novel small-molecule drug candidates for multi-target activity. This challenge involves leveraging advanced Generative AI and cheminformatics tools to predict binding affinities, assess drug-likeness, and evaluate ADMET profiles. Participants will build a sophisticated agent using Llama 3.1 405B via Groq and Semantic Kernel to orchestrate the design process. The platform will iteratively propose molecular modifications, predict their properties, and refine candidates based on specific therapeutic targets (e.g., GLP-1R, GIPR, GCGR). Experiment tracking will be managed with Weights & Biases, and external knowledge will be integrated using Tavily.
What you are building
The core problem, expected build, and operating context for this challenge.
Build an intelligent agent using Llama 3.1 (via Groq) and Semantic Kernel to iteratively design and optimize small-molecule drug candidates for multi-target activity, tracking experiments with Weights & Biases and integrating search via Tavily.
How work is evaluated
The AI agent's performance will be evaluated based on its ability to propose novel drug candidates that meet specific binding affinity, drug-likeness, and ADMET criteria for multi-target activity. The evaluation will focus on the integrated workflow, the quality of the final molecular proposals, and the effective use of tracking and search tools.
Challenge datasets
Public datasets, plus the private datasets you have access to.
What you should walk away with
Understand the principles of multi-target drug design and polypharmacology.
Apply Generative AI for de novo molecular generation and scaffold hopping.
Develop strategies for multi-objective optimization in drug design, including ADMET considerations.
Implement computational methods for predicting molecular properties and target interactions.
Utilize Semantic Kernel for orchestrating complex AI workflows and external tool integration.
Reference links and supporting material
Sample data for 3 tasks
Challenge setup
The AI agent's performance will be evaluated based on its ability to propose novel drug candidates that meet specific binding affinity, drug-likeness, and ADMET criteria for multi-target activity. The evaluation will...
Challenge input
JSON object with target receptor names and desired activity profiles (e.g., {"targets": ["GLP-1R", "GIPR"], "affinity_target": "low nM"})
Agent execution
No agent tool is listed for this challenge.
Evaluated output
JSON array of initial SMILES strings and associated descriptors (e.g., molecular weight, cLogP).
- Checks if the entire Semantic Kernel workflow executes successfully without critical errors and produces va...
- Verifies that Llama 3.1 (via Groq) API was successfully called for molecular generation and refinement tasks.
- Confirms that Weights & Biases was used to log experimental runs, metrics, and generated molecules througho...
- MultiTargetBindingAffinity target: 20
- 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
- Checks if the entire Semantic Kernel workflow executes successfully without critical errors and produces valid JSON outputs at each stage.
- Verifies that Llama 3.1 (via Groq) API was successfully called for molecular generation and refinement tasks.
- Confirms that Weights & Biases was used to log experimental runs, metrics, and generated molecules throughout the optimization process.
- Verifies that Tavily API was used at least once to retrieve relevant chemical or biological information during the refinement process.
- Ensures that the final optimized drug candidate is represented by a chemically valid SMILES string.
- Checks if the final drug candidate complies with Lipinski's Rule of Five (max 1 violation allowed).
Metrics
- MultiTargetBindingAffinity · Target 20
- ADMETRiskScore · Target 2
- StructuralNovelty · Target 0.7
Workflow tasks
- Task 1
Input: JSON object with target receptor names and desired activity profiles (e.g., {"targets": ["GLP-1R", "GIPR"], "affinity_target": "low nM"})
Output: JSON array of initial SMILES strings and associated descriptors (e.g., molecular weight, cLogP).
Execution harness: Python
- Task 2
Input: JSON array of SMILES strings from initial generation.
Output: JSON array of molecules, each with SMILES, predicted binding affinities, Lipinski violations count, and ADMET scores.
Execution harness: Python
- Task 3
Input: JSON array of scored molecules from previous task.
Output: JSON object representing the single best optimized drug candidate, including its SMILES, comprehensive predicted properties, and a detailed rationale for its selection and optimization pathway.
Execution harness: Python
Recipe freeze marker
No recipe freeze time is recorded.