Run Experiment

PIPELINE LAUNCHER

Execute PaTz pipeline scripts from the UI. Script 03 (training) has full UI launch + live WebSocket monitoring. Scripts 01/02/04/05 show configuration and terminal commands.

Data Acquisitionscripts/01_acquire_data.py→ data/raw/

Download HIV-1 and M. tuberculosis protein sequences from UniProtKB/Swiss-Prot. Fetches GO annotations. Saves raw JSONL files.

python scripts/01_acquire_data.py

ESM-2 Embedding Extractionscripts/02_extract_embeddings.py→ data/embeddings/embeddings.h5

Load ESM-2 650M and extract 1280-dim per-residue embeddings for all proteins. Requires CUDA (RTX 3050). Saves HDF5 cache.

Batch Size

python scripts/02_extract_embeddings.py --batch-size 4

Train PaTzscripts/03_train_patz.py→ checkpoints/patz_best.pt

Train the full PaTz model (Pairformer + Score Diffusion + Physics Loss). Streams live metrics via WebSocket.

Epochs

Batch Size

Learning Rate

Grad Accumulation

Disable FP16--no-fp16

XAI Pipelinescripts/04_xai_pipeline.py→ results/xai_results.json

Run Integrated Gradients attribution + MC Dropout uncertainty for all XAI proteins. Saves xai_results.json.

Batch Size

python scripts/04_xai_pipeline.py --batch-size 4

Drug Screeningscripts/05_screening.py→ results/screening_results.csv

Multi-objective screening of flavonoids against HIV-TB co-infection protein pairs. Uses PaTz structural + functional scores.

Max Pairs

python scripts/05_screening.py --max-pairs 200