Thompson Sampling

This folder contains Thompson sampling workflows for peptide screening with cluster-aware selection.

• replay.py: replay workflow that reads precomputed AF2 metrics from data/dict.csv, combines them with cluster assignments, and writes data.csv plus TS outputs.

Inputs

Cluster assignments

Cluster assignments live in ./clusters as CSV files with columns:

• name
• sequence
• cluster

Current clustering result files include:

• cd-0.4.csv, cd-0.5.csv, cd-0.7.csv, cd-0.9.csv
• easy-cluster-0.4.csv, easy-cluster-0.5.csv, easy-cluster-0.7.csv, easy-cluster-0.9.csv
• easy-linclust-0.4.csv, easy-linclust-0.5.csv, easy-linclust-0.7.csv, easy-linclust-0.9.csv

Precomputed AF2 results

data/dict.csv stores AF2 metrics with columns:

• name
• sequence
• plddt_0, dist_0, rog_0
• plddt_1, dist_1, rog_1
• plddt_2, dist_2, rog_2
• plddt_3, dist_3, rog_3
• plddt_4, dist_4, rog_4

For the offline replay workflow, these metrics are converted into a binary binder label and then written to data.csv in the standard format:

• name
• cluster
• label

Example replay run

The ./example folder contains the outputs from:

python replay.py --config data/config.json --library clusters/cd-0.5.csv --dict_path data/dict.csv --data_out example/data.csv --metrics_out example/af2_metrics.csv --out_prefix example/replay

This produces:

• example/data.csv
• example/af2_metrics.csv
• example/replay_seed.csv
• example/replay_selections.csv
• example/replay_curve.csv
• example/replay_summary.json
• example/replay_plot.png

Dependencies

• numpy
• pandas
• matplotlib