RPIS_FragmentationLibraries: Molecular Fragments Useful for Design of Molecular Glues for Protein-RNA complexes

RPIS_FragmentationLibraries: Molecular Fragments Useful for Design of Molecular Glues for Protein-RNA Complexes This Database lives at: - [Zenodo] (Large Data Files)- [github] (Executables and README)- [CSBJ] (Original Publication) Overview This repository consists of two primary components: chemical compound libraries in SMILES format, a universally recognized representation in computational molecular sciences. The fragment libraries presented here are part of a peer-reviewed scientific study. These fragments were derived through a comprehensive in silico workflow designed to identify promising stabilizer candidates for protein–RNA interactions (RPIs). The workflow integrates several computational methods, including: - Binding pocket detection and evaluation - Molecular docking- Molecular dynamics simulations - Binding free energy estimations This pipeline yielded a set of stabilizer candidates, which were then used to generate fragment libraries through two distinct approaches: 1. Extended Connectivity Fingerprints (ECFP) – to extract the most representative chemical features. 2. Breaking of Retrosynthetically Interesting Chemical Substructures (BRICS) – to decompose molecules into synthetically meaningful fragments. Both approaches were implemented using Python’s rdkit.Chem module and can be used for a large spectrum of different applications. --- Highlights - 1,000 most abundant ECFP-derived fragments for compound database filtering- Executable scripts demonstrating the database filtering workflow - T38DrugDB filtered database, organized into 9 sub-databases, containing only molecules with at least 2 to 10 fragment matches (provided as .tar.xz archives for portability) - All 213 BRICS fragments decomposed from the 96 high-ranking stabilizer ligands identified in our study - Executables for de novo compound generation from the 213 BRICS fragments, with adjustable maxDepth parameters - Comprehensive 4,000,000-compound database, generated from all possible combinations of the 213 BRICS fragments with maxDepth = 3 Intended Usage The fragment libraries in this repository are divided into two subsets: ECFP-derived fragments and BRICS-derived fragments. Although both provide databases of SMILES strings, their intended applications differ. - ECFP fragments can be used to filter existing compound databases for molecular patterns enriched in RPI stabilizers identified through our in silico workflow. These fragments are particularly useful for identifying chemical motifs associated with stabilizing protein–RNA interactions. - BRICS fragments, on the other hand, also encode combinatorial information. In this method, chemical bonds of RPI stabilizers were broken in a retrosynthetically meaningful way using the rdkit.Chem.BRICS module. The resulting fragments can be recombined following the same retrosynthetic rules, enabling the construction of new compounds enriched with RPI-stabilizing features. These fragments can also be used for database filtering, but since they lack fragment importance scores, filtering must be done naively using all fragments. While these two applications, database filtering (ECFP) and de novo compound generation (BRICS), represent the primary intended uses, they are not the only possibilities. Generative machine learning techniques, for instance, could utilize these fragment libraries to design new potential binders, particularly leveraging the ECFP dataset. In general, the scope of applications for these fragment libraries is broad and limited only by the creativity of the user. The included executable scripts demonstrate the core workflows for database filtering (ECFP) and de novo compound assembly (BRICS). ECFP – Database Filtering Within the github ECFP/ directory, you will find the executable script RefilterDatabaseWithECFP.py and the example dataset sampleDB.csv. The fragment data required for execution is provided in the file MMGBSA_ChemicalAnalysisFragments_cutoff10_ranked_datatable_3orMore.csv. This file must be present to run RefilterDatabaseWithECFP.py without errors. After installing all required dependencies (listed at the beginning of RefilterDatabaseWithECFP.py), the script can be executed to produce nine .csv files named in the format: ECFP_fragmentFilteredLibrary_#N#FragsOrMore.csv. These files serve as example outputs. The results of the filtering process are stored in the Zenode database . That dataset contains the results of applying the filtering procedure to the T38DrugDB, which includes approximately 34 million drug-like compounds published elsewhere. The filtered sub-databases are provided as compressed `.tar.xz` archives for portability. These databases can be used for targeted virtual screening of RPI-stabilizing drug candidates. This example workflow can be executed as follows: # clone this repositorygit clone [e-mail]:Foly93/RPIS_FragmentationLibraries.git ./ # switch to ECFP directorycd RPIS_FragmentationLibraries/ECFP # INSTALL REQUIRED PYTHON PACKAGES FROM YOUR FAVOURITE PACKAGE MANAGER e.g. micromambamicromamba activate your_env_name_goes_here # execute the Python programpython RefilterDatabaseWithECFP.py # Check if the expected output was generated (assuming ubuntu, mac or power shell)ls -rtal ECFP_fragmentFilteredLibrary_*FragsOrMore.csv BRICS – De Novo Compound Assembly The BRICS/ directory contains several executables that serve different purposes, as well as the BRICS fragment library RPIS_BRICS_Fragment_DB.smi. This .smifile contains 213 BRICS fragments, which can be combined to generate novel molecules using the rdkit.Chem.BRICS module. This functionality is demonstrated across three executables:- BRICS_fragment_database_interactive.ipynb- build_one_example_Mol_from_BRICS_fragments.py- generate_N_Mols_from_BRICS_fragments.py The Jupyter notebook BRICS_fragment_database_interactive.ipynb requires a Python environment with jupyter notebook installed. It provides a visual and interactive overview of the fragment set, demonstrating how to:- Display the BRICS fragments - Assemble random molecules from the fragment library - Export the resulting molecules as SMILES strings The assembly process uses the RDKit function BRICS.BRICSBuild. Its parameters are highly sensitive—particularly the maxDepth option, which controls the maximum number of fragments combined into a single molecule. With 213 fragments available, the total combinatorial space is on the order of 4 × 10¹¹ possible assemblies, making exhaustive enumeration computationally infeasible. The script build_one_example_Mol_from_BRICS_fragments.py replicates the notebook’s core functionality in a standalone Python executable. It generates a single example molecule, saving the output to a file identified by its timestamp.The final executable, generate_N_Mols_from_BRICS_fragments.py, creates a specified number of BRICS-assembled molecules and saves them into a timestamped .smi file. While it can be run without command-line arguments, optional parameters are available and can be displayed using the -h flag. The generated .smi files can be directly used for virtual screening or binding affinity prediction in drug discovery workflows. This showcase can be executed as follows:# clone this repository# not necessary if already donegit clone [e-mail]:Foly93/RPIS_FragmentationLibraries.git ./ # switch to BRICS directorycd RPIS_FragmentationLibraries/BRICS # INSTALL REQUIRED PACKAGES FROM YOUR FAVOURITE PACKAGE MANAGER e.g. micromambamicromamba activate your_env_name_goes_here # open the jupyter notebook in your browser and have a good look at the functionalityjupyter notebook BRICS_fragment_database_interactive.ipynb # execute the one-example-python-programpython build_one_example_Mol_from_BRICS_fragments.py # execute the batch creation python programpython generate_N_Mols_from_BRICS_fragments.py # display the options of the batch creation python programpython generate_N_Mols_from_BRICS_fragments.py -h # run the python program with custom flagspython generate_N_Mols_from_BRICS_fragments.py \ --maxDepth 3 \ --numMold 10 \ --scrambleReagents True \ --outputDirectory ../trashFinally, the Zenodo Database also contains BRICS_DB_BuiltMaxDepth_3.txt a data base that contains all possible combinations for maxDepth set to 3. This file contains 3,878,955 compound SMILES strings which is less that the theoretically possible 213 x 213 x 213 = 9,663,597 which results from incompatibilities between some BRICS fragments and duplicate entries. File Description RPIS_FragmentationLibraries/├── README.md # THIS file├── ECFP_fragmentFilteredLibrary_3FragsOrMore.csv.tar.xz # T38DrugDB compounds that contain 3 0r more ECFP fragments├── ECFP_fragmentFilteredLibrary_4FragsOrMore.csv.tar.xz # T38DrugDB compounds that contain 4 0r more ECFP fragments├── ECFP_fragmentFilteredLibrary_5FragsOrMore.csv.tar.xz # T38DrugDB compounds that contain 5 0r more ECFP fragments├── ECFP_fragmentFilteredLibrary_6FragsOrMore.csv.tar.xz # T38DrugDB compounds that contain 6 0r more ECFP fragments├── ECFP_fragmentFilteredLibrary_7FragsOrMore.csv.tar.xz # T38DrugDB compounds that contain 7 0r more ECFP fragments├── ECFP_fragmentFilteredLibrary_8FragsOrMore.csv.tar.xz # T38DrugDB compounds that contain 8 0r more ECFP fragments├── ECFP_fragmentFilteredLibrary_9FragsOrMore.csv.tar.xz # T38DrugDB compounds that contain 9 0r more ECFP fragments├── ECFP_fragmentFilteredLibrary_10FragsOrMore.csv.tar.xz # T38DrugDB compounds that contain 10 0r more ECFP fragments└── BRICS_DB_BuiltMaxDepth_3.txt # Data base containing SMILES of all available BRICS assemblies with maxDepth set to 3 Citation If you use these Fragment libraries, please cite:```Luis Vollmers, Shu-Yu Chen, Martin Zacharias. In Silico Analysis of Potential Stabilizer Binding Sites at Protein–RNA Interfaces. Comput Struct Biotechnol J. 2026;35:0016.DOI:10.34133/csbj.0016``` License This work is licensed under a Creative Commons Attribution 4.0 International License. See creativecommons.org/licenses/by/4.0/ for further information. Contact For questions, issues, or contributions:- [e-mail]- [e-mail]- Publication Link: https://doi.org/10.34133/csbj.0016RPIS_FragmentationLibraries：用于蛋白质-RNA复合物分子胶设计的分子片段库 本数据库的存储位置如下： - [Zenodo]（大型数据文件）- [GitHub]（可执行文件与README文档）- [CSBJ]（原始研究论文） ## 概述 本仓库包含两大核心组成部分：采用SMILES格式的化合物库——SMILES是计算分子科学领域通用的分子表征格式。 本文呈现的片段库源自一项经过同行评审的科学研究。这些片段通过一套完整的in silico（计算机模拟）工作流程得到，该流程旨在筛选可用于蛋白质-RNA相互作用（RPIs）的潜在稳定剂候选物。 该工作流程整合了多种计算方法，具体包括： - 结合口袋检测与评估 - 分子对接 - 分子动力学模拟 - 结合自由能估算 本流程得到了一批稳定剂候选物，随后通过两种不同的方法生成片段库： 1. 扩展连接性指纹（Extended Connectivity Fingerprints，简称ECFP）——用于提取最具代表性的化学特征。 2. 逆合成感兴趣化学子结构断裂法（Breaking of Retrosynthetically Interesting Chemical Substructures，简称BRICS）——将分子拆解为具备合成意义的片段。 两种方法均通过Python的rdkit.Chem模块实现，可应用于广泛的研究场景。 --- ## 核心亮点 - 1000个丰度最高的ECFP衍生片段，用于化合物库筛选 - 配套可执行脚本，演示数据库筛选工作流程 - 经T38DrugDB筛选得到的次级数据库，共分为9个次级数据库，仅包含与至少2~10个片段匹配的化合物，以.tar.xz压缩归档格式提供以保证跨平台兼容性 - 从本研究筛选出的96个优质稳定剂配体中拆解得到的全部213个BRICS片段 - 支持自定义maxDepth参数的、基于213个BRICS片段的全新化合物生成可执行程序 - 基于213个BRICS片段、maxDepth参数设为3的所有可能组合生成的包含400万个化合物的综合数据库 ## 适用场景 本仓库中的片段库分为两个子集：ECFP衍生片段与BRICS衍生片段。尽管二者均提供SMILES格式的字符串数据库，但适用场景有所区别。 - ECFP衍生片段可用于筛选现有化合物库，找出经本研究计算机模拟工作流程确定的、富含RPI稳定剂的分子模式。此类片段尤其适用于识别与稳定蛋白质-RNA相互作用相关的化学基序。 - BRICS衍生片段则携带组合信息。本方法通过rdkit.Chem.BRICS模块，以具备逆合成意义的方式断裂RPI稳定剂的化学键，得到的片段可按照相同的逆合成规则重新组合，从而构建出富含RPI稳定特征的全新化合物。此类片段也可用于数据库筛选，但由于其未附带片段重要性评分，筛选时需采用直接使用全部片段的朴素方式。 尽管数据库筛选（ECFP）与全新化合物生成（BRICS）是这两类片段库的核心适用场景，但并非仅局限于此。例如，生成式机器学习技术可借助这些片段库设计新型潜在结合剂，尤其可利用ECFP数据集。总体而言，本片段库的应用场景十分广泛，仅受限于使用者的创造力。附带的可执行脚本演示了数据库筛选（ECFP）与全新化合物组装（BRICS）的核心工作流程。 ### ECFP——数据库筛选 在GitHub仓库的ECFP/目录下，可找到可执行脚本`RefilterDatabaseWithECFP.py`与示例数据集`sampleDB.csv`。运行该脚本所需的片段数据存储于文件`MMGBSA_ChemicalAnalysisFragments_cutoff10_ranked_datatable_3orMore.csv`中，该文件为脚本正常运行的必备文件。安装所有依赖项（详见`RefilterDatabaseWithECFP.py`文件开头说明）后，执行该脚本将生成9个格式为`ECFP_fragmentFilteredLibrary_#N#FragsOrMore.csv`的CSV文件，作为示例输出结果。数据库筛选流程的最终结果存储于Zenodo数据库中，该数据集包含将筛选流程应用于T38DrugDB的结果——T38DrugDB包含约3400万个已在其他文献中发表的类药化合物。经筛选得到的次级数据库以.tar.xz压缩归档格式提供，以保证跨平台兼容性，可用于RPI稳定剂候选药物的靶向虚拟筛选。本示例工作流程的执行步骤如下： bash # 克隆本仓库 git clone [e-mail]:Foly93/RPIS_FragmentationLibraries.git ./ # 进入ECFP目录 cd RPIS_FragmentationLibraries/ECFP # 通过包管理器安装依赖项，例如micromamba micromamba activate your_env_name_goes_here # 执行Python脚本 python RefilterDatabaseWithECFP.py # 检查预期输出结果（适用于Ubuntu、macOS或PowerShell环境） ls -rtal ECFP_fragmentFilteredLibrary_*FragsOrMore.csv ### BRICS——全新化合物组装 BRICS/目录下包含多个具备不同功能的可执行文件，以及BRICS片段库`RPIS_BRICS_Fragment_DB.smi`。该.smi文件包含213个BRICS片段，可通过rdkit.Chem.BRICS模块组合生成全新分子。相关功能通过三个可执行文件进行演示： - `BRICS_fragment_database_interactive.ipynb` - `build_one_example_Mol_from_BRICS_fragments.py` - `generate_N_Mols_from_BRICS_fragments.py` Jupyter Notebook文件`BRICS_fragment_database_interactive.ipynb`需要搭载已安装Jupyter Notebook的Python环境，可提供可视化的交互式片段集概览，并演示以下操作： - 展示BRICS片段 - 从片段库中随机组装分子 - 将生成的分子导出为SMILES字符串 组装过程使用RDKit的`BRICS.BRICSBuild`函数，其参数设置极为关键——尤其是控制单分子中组合片段最大数量的`maxDepth`参数。本片段库共有213个片段，理论上的组合空间可达4×10¹¹种可能的组装结果，因此穷举所有组合在计算上不可行。 脚本`build_one_example_Mol_from_BRICS_fragments.py`将Notebook的核心功能封装为独立可执行程序，可生成单个示例分子并将结果保存至以时间戳命名的文件中。 最后一个可执行文件`generate_N_Mols_from_BRICS_fragments.py`可生成指定数量的BRICS组装分子，并将结果保存至以时间戳命名的.smi文件中。该脚本支持无命令行参数直接运行，也可通过`-h`标志查看可选参数。生成的.smi文件可直接用于药物发现流程中的虚拟筛选或结合亲和力预测。本演示流程的执行步骤如下： bash # 克隆本仓库（若已克隆则可省略） git clone [e-mail]:Foly93/RPIS_FragmentationLibraries.git ./ # 进入BRICS目录 cd RPIS_FragmentationLibraries/BRICS # 通过包管理器安装依赖项，例如micromamba micromamba activate your_env_name_goes_here # 在浏览器中打开Jupyter Notebook查看功能演示 jupyter notebook BRICS_fragment_database_interactive.ipynb # 执行单分子生成脚本 python build_one_example_Mol_from_BRICS_fragments.py # 执行批量分子生成脚本 python generate_N_Mols_from_BRICS_fragments.py # 查看批量生成脚本的可选参数 python generate_N_Mols_from_BRICS_fragments.py -h # 使用自定义参数运行批量生成脚本 python generate_N_Mols_from_BRICS_fragments.py --maxDepth 3 --numMold 10 --scrambleReagents True --outputDirectory ../trash 此外，Zenodo数据库中还包含`BRICS_DB_BuiltMaxDepth_3.txt`，该数据库包含所有maxDepth设为3的BRICS组装产物。该文件包含3,878,955个化合物的SMILES字符串，少于理论上的213×213×213=9,663,597种组合，这是由于部分BRICS片段之间存在兼容性问题以及存在重复条目导致的。 ## 文件说明 RPIS_FragmentationLibraries/ ├── README.md # 本说明文件 ├── ECFP_fragmentFilteredLibrary_3FragsOrMore.csv.tar.xz # 包含至少3个ECFP片段的T38DrugDB化合物 ├── ECFP_fragmentFilteredLibrary_4FragsOrMore.csv.tar.xz # 包含至少4个ECFP片段的T38DrugDB化合物 ├── ECFP_fragmentFilteredLibrary_5FragsOrMore.csv.tar.xz # 包含至少5个ECFP片段的T38DrugDB化合物 ├── ECFP_fragmentFilteredLibrary_6FragsOrMore.csv.tar.xz # 包含至少6个ECFP片段的T38DrugDB化合物 ├── ECFP_fragmentFilteredLibrary_7FragsOrMore.csv.tar.xz # 包含至少7个ECFP片段的T38DrugDB化合物 ├── ECFP_fragmentFilteredLibrary_8FragsOrMore.csv.tar.xz # 包含至少8个ECFP片段的T38DrugDB化合物 ├── ECFP_fragmentFilteredLibrary_9FragsOrMore.csv.tar.xz # 包含至少9个ECFP片段的T38DrugDB化合物 ├── ECFP_fragmentFilteredLibrary_10FragsOrMore.csv.tar.xz # 包含至少10个ECFP片段的T38DrugDB化合物 └── BRICS_DB_BuiltMaxDepth_3.txt # 包含所有maxDepth设为3的BRICS组装产物SMILES的数据库 ## 引用说明 若使用本片段库，请引用以下文献： Luis Vollmers, Shu-Yu Chen, Martin Zacharias. In Silico Analysis of Potential Stabilizer Binding Sites at Protein–RNA Interfaces. Comput Struct Biotechnol J. 2026;35:0016. DOI:10.34133/csbj.0016 ## 许可协议 本作品采用知识共享署名4.0国际许可协议（Creative Commons Attribution 4.0 International License）进行许可，详细信息请参阅https://creativecommons.org/licenses/by/4.0/。 ## 联系方式 如有疑问、问题或贡献需求，请联系： - [e-mail] - [e-mail] - 论文链接：https://doi.org/10.34133/csbj.0016