Prism Gen Demo

PRISM-Gen Demo Skill

English: PRISM-Gen pre-calculation result display demo: Provides retrieval, filtering, sorting, merging, and visualization of multiple CSV result files from molecular generation/screening, offering portable query-based skills.

中文: PRISM-Gen预计算结果展示Demo：对分子生成/筛选的多个CSV结果文件进行检索、过滤、排序、合并和可视化，提供可移植的查询型技能。

English Section

Overview

PRISM-Gen Demo is a portable skill for analyzing pre-calculated molecular screening results. It provides query-based data retrieval, filtering, sorting, merging, and visualization capabilities without requiring HPC connections or computational workflows.

Design Goals

• Portability: Does not trigger HPC computation workflows, only processes existing CSV files
• Stability: Core functions work offline; Python dependencies optional for advanced visualization
• Query-based: Provides retrieval, filtering, sorting, and merging functions
• Structured: Returns results in a clear structured format
• Visualization: Provides data visualization and profile summarization (requires Python)

Usage Scenarios

✅ Use this skill when:

• "Show PRISM Demo results"
• "Retrieve molecular data" / "Filter CSV results"
• "Sort molecules" / "Top N screening"
• "Merge multiple stage results" / "Profile summarization"
• "Visualization analysis" / "Chart display"
• "Export results" / "Format conversion"

Core Function Architecture

1. Data Source Management

• List data sources: Display all available CSV files
• Source information: Show file structure and statistics
• Data preview: Quick view of sample data

2. Data Query

• Conditional filtering: Single or multi-column condition screening
• Top N selection: Sort by specified column to get best molecules
• Range queries: Support numerical ranges and string matching

3. Data Analysis

• Correlation analysis: Calculate Pearson, Spearman correlation coefficients
• Regression analysis: Linear regression and trend lines
• Distribution analysis: Histograms, box plots, Q-Q plots

4. Data Visualization

• Distribution plots: Univariate distribution visualization
• Scatter plots: Bivariate correlation analysis
• Statistical charts: Publication-quality statistical charts

5. Data Export

• CSV export: Save filtered and sorted results
• Chart export: PNG, PDF, SVG formats
• Report generation: Structured analysis reports

Supported CSV Files

• step3a_optimized_molecules.csv - Surrogate model optimized molecules
• step3b_dft_results.csv - xTB+DFT electronic screening results
• step3c_dft_refined.csv - GEM re-ranking results
• step4a_admet_final.csv - ADMET filtering results
• step4b_top_molecules_pyscf.csv - DFT validation (PySCF) results
• step4c_master_summary.csv - Master summary table
• step5a_broadspectrum_docking.csv - Broad-spectrum docking results
• step5b_final_candidates.csv - Final candidate molecules

Key Molecular Properties

• Identifiers: smiles, molecule_id, name
• Activity: pIC50, reward, broad
• Physicochemical properties: LogP, MW, TPSA, HBD, HBA
• Safety: hERG_Prob, AMES, hepatotoxicity
• Drug-likeness: QED, SA, Lipinski
• Electronic properties: gap, energy, dipole
• Docking results: docking_score, binding_energy

Quick Start Examples

Example 1: List Data Sources

bash scripts/demo_list_sources.sh

Example 2: Filter High-Activity Molecules

# Filter molecules with pIC50 > 7.0
bash scripts/demo_filter.sh step4a_admet_final.csv pIC50 '>' 7.0

Example 3: Get Top 10 Active Molecules

bash scripts/demo_top.sh step4a_admet_final.csv pIC50 10

Example 4: Generate Distribution Plot

bash scripts/demo_plot_distribution.sh step4a_admet_final.csv pIC50

Example 5: Correlation Analysis

bash scripts/demo_plot_scatter.sh step4a_admet_final.csv pIC50 QED --trendline --correlation

Technical Requirements

Basic Functions (No Installation Required, Fully Offline)

• ✅ Bash shell environment
• ✅ Standard Unix tools (awk, sed, grep)
• ✅ File read/write permissions
• 🚫 No network connection required
• 🚫 No Python installation required

Advanced Functions (Requires Python, Works Offline After Installation)

• 🐍 Python 3.10+
• 📦 Core packages: pandas, numpy, matplotlib, seaborn
• 🔬 Scientific computing: scipy, scikit-learn (optional)
• ⚠️ Network required for installation only
• ✅ Offline usage after installation

Compatibility

• ✅ Linux / macOS / WSL2
• ✅ Local file system
• ✅ Basic functions: Fully offline
• ⚠️ Advanced functions: Network required for installation only

Project Structure

prism-gen-demo/
├── README.md                    # This document
├── SKILL.md                     # OpenClaw skill definition
├── requirements.txt             # Python dependencies
├── data/                        # Pre-calculation result CSV files
├── scripts/                     # Core scripts
├── config/                      # Configuration files
├── examples/                    # Usage examples
├── docs/                        # Documentation
├── output/                      # Output directory
└── plots/                       # Chart output

中文部分

概述

PRISM-Gen Demo 是一个用于分析预计算分子筛选结果的可移植技能。它提供基于查询的数据检索、过滤、排序、合并和可视化功能，无需HPC连接或计算工作流。

设计目标

• 可移植性: 不触发HPC计算流程，只处理既有CSV文件
• 稳定性: 核心功能离线工作；Python依赖仅用于高级可视化（可选）
• 查询型: 提供检索、过滤、排序、合并功能
• 结构化: 以清晰的结构化方式返回结果
• 可视化: 提供数据可视化和profile汇总（需要Python）

使用场景

✅ 使用此技能当：

• "查看PRISM Demo结果" / "展示预计算结果"
• "检索分子数据" / "过滤CSV结果"
• "排序分子" / "Top N筛选"
• "合并多个阶段结果" / "Profile汇总"
• "可视化分析" / "图表展示"
• "导出结果" / "格式转换"

核心功能架构

1. 数据源管理

• 列出数据源: 显示所有可用CSV文件
• 数据源信息: 显示文件结构和统计信息
• 数据预览: 快速查看样本数据

2. 数据查询

• 条件过滤: 基于单列或多列条件筛选分子
• Top N筛选: 按指定列排序获取最佳分子
• 范围查询: 支持数值范围和字符串匹配

3. 数据分析

• 相关性分析: 计算Pearson、Spearman相关系数
• 回归分析: 线性回归和趋势线
• 分布分析: 直方图、箱线图、Q-Q图

4. 数据可视化

• 分布图: 单变量分布可视化
• 散点图: 双变量相关性分析
• 统计图表: 论文质量的统计图表

5. 数据导出

• CSV导出: 保存过滤和排序结果
• 图表导出: PNG、PDF、SVG格式
• 报告生成: 结构化分析报告

支持的CSV文件

• step3a_optimized_molecules.csv - 代理模型优化分子
• step3b_dft_results.csv - xTB+DFT电子筛选结果
• step3c_dft_refined.csv - GEM重排序结果
• step4a_admet_final.csv - ADMET过滤结果
• step4b_top_molecules_pyscf.csv - DFT验证(PySCF)结果
• step4c_master_summary.csv - 主汇总表
• step5a_broadspectrum_docking.csv - 广谱对接结果
• step5b_final_candidates.csv - 最终候选分子

关键分子属性

• 标识符: smiles, molecule_id, name
• 活性: pIC50, reward, broad
• 物化性质: LogP, MW, TPSA, HBD, HBA
• 安全性: hERG_Prob, AMES, hepatotoxicity
• 药物相似性: QED, SA, Lipinski
• 电子性质: gap, energy, dipole
• 对接结果: docking_score, binding_energy

快速开始示例

示例1：列出数据源

bash scripts/demo_list_sources.sh

示例2：筛选高活性分子

# 筛选pIC50 > 7.0的分子
bash scripts/demo_filter.sh step4a_admet_final.csv pIC50 '>' 7.0

示例3：获取Top 10活性分子

bash scripts/demo_top.sh step4a_admet_final.csv pIC50 10

示例4：生成分布图

bash scripts/demo_plot_distribution.sh step4a_admet_final.csv pIC50

示例5：相关性分析

bash scripts/demo_plot_scatter.sh step4a_admet_final.csv pIC50 QED --trendline --correlation

技术要求

基础功能（无需安装，完全离线）

• ✅ Bash shell环境
• ✅ 标准Unix工具 (awk, sed, grep)
• ✅ 文件读写权限
• 🚫 无需网络连接
• 🚫 无需Python安装

高级功能（需要Python，安装后离线工作）

• 🐍 Python 3.10+
• 📦 核心包: pandas, numpy, matplotlib, seaborn
• 🔬 科学计算包: scipy, scikit-learn (可选)
• ⚠️ 仅安装需要网络
• ✅ 安装后可离线使用

兼容性

• ✅ Linux / macOS / WSL2
• ✅ 本地文件系统
• ✅ 基础功能：完全离线
• ⚠️ 高级功能：仅安装需要网络

项目结构

prism-gen-demo/
├── README.md                    # 本文档
├── SKILL.md                     # OpenClaw技能定义
├── requirements.txt             # Python依赖
├── data/                        # 预计算结果CSV文件
├── scripts/                     # 核心脚本
├── config/                      # 配置文件
├── examples/                    # 使用示例
├── docs/                        # 文档
├── output/                      # 输出目录
└── plots/                       # 图表输出

License / 许可证

MIT License - See LICENSE file for details. / MIT许可证 - 详见LICENSE文件。

Contact / 联系方式

For questions or suggestions, please refer to the documentation or contact the skill author. / 如有问题或建议，请参考文档或联系技能作者。