How do GFlowNets assist in drug discovery?

GFlowNets are used to generate diverse molecular structures (SMILES) that satisfy specific reward functions, such as binding affinity and synthesizability, preventing the model from collapsing into a single molecular mode.

Which machine learning frameworks are supported by this skill?

This skill provides implementation guidance and best practices for leading Python-based ML frameworks, specifically PyTorch, JAX, and the specialized torchgfn library.

What is Trajectory Balance in the context of GFlowNets?

Trajectory Balance (TB) is a training objective that ensures the product of forward transition probabilities along a path (multiplied by a partition function) matches the reward and backward probabilities, allowing for efficient credit assignment.

What makes GFlowNets different from Reinforcement Learning?

While traditional Reinforcement Learning focuses on maximizing expected reward to find a single optimal solution, GFlowNets learn to sample candidates with a probability proportional to their reward, ensuring a much more diverse set of high-quality results.

GFlowNet Generative Modeling

Name: GFlowNet Generative Modeling
Author: plurigrid

byplurigrid

•

Data Science & ML

Implements Generative Flow Networks to sample diverse, high-reward candidates for molecule design, causal discovery, and combinatorial optimization.

The GFlowNet skill equips Claude with the specialized knowledge to implement Generative Flow Networks, a paradigm shift in generative modeling that samples candidates proportionally to their reward rather than simply maximizing it. This skill provides deep technical guidance on Trajectory Balance (TB) and Detailed Balance (DB) objectives, enabling the creation of models that discover diverse optima in complex state spaces. It is essential for developers and researchers working on drug discovery, Bayesian structure learning, and scenarios where amortized, off-policy sampling of discrete structures is required using frameworks like PyTorch and JAX.

Key Features

01Diverse candidate generation for drug discovery and molecular SMILES strings

02Implementation of Trajectory Balance (TB) and Detailed Balance (DB) training objectives

03Cross-framework patterns for JAX, PyTorch, and the torchgfn library

04Advanced sampling techniques using Interaction Entropy and GF(3) triads

05Support for Bayesian causal discovery through probabilistic DAG sampling

067 GitHub stars

Use Cases

01Designing novel drug-like molecules with diverse structural properties and high binding affinity

02Discovering underlying causal structures in complex datasets using probabilistic graph sampling

03Solving combinatorial optimization problems where finding multiple varied solutions is critical

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add plurigrid/asi gflownet

For use in Claude.ai and ChatGPT

Download Skill