PRISM-Gen Demo

Description

PRISM-Gen Demo is a read-only data analysis skill for exploring pre-calculated molecular screening results from the PRISM-Gen broad-spectrum coronavirus Mpro inhibitor discovery pipeline.

All analysis is performed locally on CSV files bundled with the skill. No network access, no external API calls, no credential requirements.

What This Skill Does

• Retrieve data from any of 10 pipeline stage CSV files (step3a through step5b)
• Filter molecules by property thresholds (pIC50, QED, MW, LogP, gap_ev, hERG_Prob, etc.)
• Sort and rank molecules by any numeric column
• Merge data across pipeline stages using SMILES as join keys
• Score molecules with composite scoring, worst-case multi-target analysis, summary statistics, and Pareto front identification
• Plot property distributions, scatter plots, docking score heatmaps, pipeline attrition funnels, and Pareto fronts
• Summarize the full pipeline with attrition statistics and key findings

What This Skill Does NOT Do

• Does NOT run any computational chemistry calculations (no DFT, no docking, no MD)
• Does NOT access any network or external services
• Does NOT require or use any credentials, API keys, or tokens
• Does NOT modify any files outside its own output directory
• Does NOT contain any shell scripts
• Does NOT use hardcoded absolute paths

Commands

retrieve

Retrieve and display data from any pipeline stage.

python3 scripts/retrieve.py --stage step5b --columns name,pIC50,Broad_Spectrum_Score --max_rows 10
python3 scripts/retrieve.py --list_stages
python3 scripts/retrieve.py --stage step4a --list_columns
python3 scripts/retrieve.py --stage step5b --name mol_16

filter

Filter molecules by property thresholds. Multiple conditions are combined with AND.

python3 scripts/filter.py --stage step4a --where "pIC50>7.5" "QED>0.7" "hERG_Prob<0.5"
python3 scripts/filter.py --stage step5b --where "Broad_Spectrum_Score<-7.0" "Is_Final_Top==True"

sort

Sort and rank molecules by any column.

python3 scripts/sort.py --stage step5b --by Broad_Spectrum_Score --ascending --top 10
python3 scripts/sort.py --stage step4c --by R_global --top 20

merge

Merge data across pipeline stages on SMILES keys.

python3 scripts/merge.py --stages step3c,step4a --on smiles --columns pIC50,gap_ev,Lipinski_Pass,hERG_Prob

score

Compute composite scores, worst-case analysis, statistics, and Pareto fronts.

python3 scripts/score.py --mode worst_case --stage step5a --top 10
python3 scripts/score.py --mode composite --stage step4c --weights "pIC50:1.0,QED:0.5,R_ADMET:2.0"
python3 scripts/score.py --mode stats --stage step5b --columns pIC50,QED,Broad_Spectrum_Score
python3 scripts/score.py --mode pareto --stage step5b --obj1 pIC50 --obj2 QED

plot

Generate visualizations (requires matplotlib).

python3 scripts/plot.py --mode histogram --stage step4a --column pIC50 --output hist.png
python3 scripts/plot.py --mode scatter --stage step5b --x pIC50 --y Broad_Spectrum_Score --output scatter.png
python3 scripts/plot.py --mode heatmap --stage step5a --output heatmap.png
python3 scripts/plot.py --mode funnel --output funnel.png
python3 scripts/plot.py --mode pareto --stage step5b --x pIC50 --y QED --output pareto.png

summary

Generate a full pipeline summary report.

python3 scripts/summary.py
python3 scripts/summary.py --detailed

Available Pipeline Stages

Stage Key	Description	Rows	Columns
step3a	RL-optimized molecules (generation + surrogate scoring)	200	10
step3a_top	Top 200 molecules by Reward	200	10
step3b	GFN2-xTB electronic structure results	200	6
step3c	xTB-refined ranking with GEM scoring	200	24
step4a	ADMET filtering (Lipinski, hERG, QED)	200	38
step4b	B3LYP/6-31G* DFT validation (PySCF)	46	34
step4c	Master summary (all stages merged)	200	65
step5a	Broad-spectrum docking (3 targets)	36	13
step5b	Final candidates with full annotations	36	75
step5b_master	Master summary with docking	200	74

Dependencies

• Required: Python 3.7+ (core functionality uses only Python standard library: csv, json, argparse, math, os, sys)
• Optional: matplotlib (for visualization commands only; all non-plot commands work without it)

Security

• Pure Python implementation — no shell scripts, no subprocess calls, no os.system calls
• No hardcoded paths — all file paths are resolved relative to the skill root directory
• No network access — all data is bundled locally in the data/ directory
• No credentials — no environment variables, API keys, or tokens required
• Read-only on input data — CSV files in data/ are never modified
• Output isolation — generated plots are written only to user-specified paths

License

MIT-0 — Free to use, modify, and redistribute. No attribution required.

Author

@SenaZeng

Related

• GitHub: https://github.com/SenaZeng/PRISM-Gen
• Zenodo: https://doi.org/10.5281/zenodo.18764996