How anirudhvijayabalaji/reinforcement-learning works
MolRL generates molecules in protein binding pockets through interactive or training workflows: retrieve and voxelize a target, propose atom placements, validate and score the assembled molecule, then inspect or compare results. The graph includes the documented web and desktop interfaces, agent comparison, interaction profiling, and explanation features. Edges are limited to source-index references, shown excerpts, and README statements; several documented subsystems were not sampled, so their internal connections are not inferred.
Main entry points: app.py MolRL Studio · environment.py Placement environment · main.py Training orchestration · visualizer.py Tkinter dashboard
Components 20
User interfaces
- MolRL Studio
app.py - Browser visualization
static/js/app.js - Tkinter dashboard
visualizer.py - Molecule visualization
visualization.py
Molecule generation
- Placement environment
environment.py
Learning and agents
- Training orchestration
main.py - 3D actor-critic
agent.py - 3D attention blocks
cbam_3d.py - PPO optimization
ppo.py - Agent zoo
agents_zoo.py - Agent benchmark
benchmark.py
Chemistry and scoring
- Molecule construction and validation
physics.py - Target reward oracles
reward.py - Formula explanations
formula_explainer.py
Protein and analysis
- Target configuration
target_config.json - Structure retrieval and voxelization
pdb_parser.py - Pocket interaction profiling
interactions.py
Other components
- Web user
- Desktop user
- RCSB and AlphaFold
Connections 26
- Web user → MolRL Studio: uses
README.md - MolRL Studio → Browser visualization: serves visualization
README.md - Desktop user → Tkinter dashboard: uses
README.md - Training orchestration → Placement environment: steps through
main.py - Training orchestration → PPO optimization: optimizes with
main.py - Training orchestration → 3D actor-critic: trains
main.py - Training orchestration → Molecule visualization: generates outputs
main.py - Placement environment → Structure retrieval and voxelization: loads pocket
environment.py - Placement environment → Molecule construction and validation: constructs molecule
environment.py - Placement environment → Target reward oracles: scores molecule
environment.py - Target configuration → Placement environment: selects target
environment.py - Structure retrieval and voxelization → RCSB and AlphaFold: retrieves structures
pdb_parser.py - 3D actor-critic → 3D attention blocks: applies
agent.py - PPO optimization → 3D actor-critic: updates
main.py - MolRL Studio → Placement environment: steps through
app.py - MolRL Studio → Agent zoo: selects agent
app.py - MolRL Studio → Agent benchmark: starts benchmark
app.py - MolRL Studio → Formula explanations: explains assembly
README.md - MolRL Studio → Pocket interaction profiling: profiles contacts
README.md - MolRL Studio → Target reward oracles: shows scores
app.py - Tkinter dashboard → Placement environment: steps through
visualizer.py - Tkinter dashboard → Structure retrieval and voxelization: loads target
visualizer.py - Tkinter dashboard → PPO optimization: runs training
visualizer.py - Pocket interaction profiling → Molecule construction and validation: uses chemistry
interactions.py - Agent benchmark → Agent zoo: compares agents
README.md - Browser visualization → Molecule visualization: renders molecules
README.md