How ulton321/protein-language-model-steering works
This is a research repository for protein-language-model steering and fine-tuning, with notebooks for ProGen2, ProLLaMA, and ESM. The documented workflow is to prepare protein data, configure and run a notebook, then evaluate and visualize results. The tree also contains an execution script and analysis outputs, but the sampled script refers to a different notebook path than the tree; connections between it and the listed notebooks are therefore not asserted. Notebook internals were not sampled, so model-specific calls and data flow remain unverified.
Main entry points: run_pipeline.py Notebook runner
Components 14
Research workflow
- Notebook runner
Scripts/run_pipeline.py - Protein scores
data/progen2_200_sequences_scores.csv
Model notebooks
- ProGen2 base
notebook/Progen2/BaseProleg2.ipynb - ProGen2 fine-tuning
notebook/Progen2/Fine_Tuned_Progen.ipynb - ProGen2 task vectors
notebook/Progen2/ProGen2_TaskVec.ipynb - ProLLaMA base
notebook/Prollama/Base_Prollama (1).ipynb - ProLLaMA fine-tuning
notebook/Prollama/Fine_Tuned_Prollama (1).ipynb - ESM base
notebook/esm/base.ipynb - ESM fine-tuning
notebook/esm/Fine_Tuned_ESM3.ipynb
Analysis outputs
- Property results
Output/model_property_results.csv - Best mean summary
Output/summary_best_by_mean.csv - FDR tests
Output/ttest_results_FDR.csv - Results plot
Output/plot/results.pdf
Other components
- Researcher
Connections 1
- Researcher → Notebook runner: runs pipeline
Scripts/run_pipeline.py