InPETM

An integrated analysis using association rule mining and network pharmacology to identify therapeutic combinations of herbal materials and compounds in traditional medicine.

Traditional medicine (TM) has been used to treat a variety of symptoms and diseases through the combination of herbal materials, and it also contributes to the pharmaceutical industry with several advantages such as fewer side effects and significant cost reductions. However, the rules for combining ingredients are not well organized, and complex multi-compound characteristics make it difficult to understand the pharmacological mechanisms among the herbal materials used in TM. In silico approaches that have been proposed to analyze TM and herbal materials require large amount of high-quality structural information or physicochemical properties or have limitations due to ease of interpretation or scope of analysis.

In this work, we proposed an approach named InPETM, that integrates association rule mining (ARM) and network pharmacology analyses to identify polypharamcological effects of herbal materials and compounds from TM. Specifically, InPETM performs analyses combining ARM and network pharmacology-based method at the herb-level and compound-level, respectively, and identifies potential herbal material combination and compound candidates for the phenotype. InPETM provided results of pharmacological effects of herbal material combination and compound and identification of mechanism of action in human protein interactome network, which were confirmed by further structural network analysis and literature review analysis. These results indicate that InPETM can contribute to drug development in TM through better understanding of polypharmacological features of herbal materials.

Installation

This source code was tested on the basic environment with:

• python==3.9.7
• conda==4.10.3
• PostgreSQL == version 17

It is also recommended to use the following versions of packages:

• numpy==1.24.3
• pandas==2.0.3
• scipy==1.13.1

Data Statistics

Note: In the full_data and data directories, you can see the raw files used for the preprocessing. The entities and amounts of data provided by each raw file are shown in the following table. Also for COCONUT, herbal material data with biological classifications higher than species are included because they are standardized based on taxonomy identifiers from the NCBI taxonomy database. Therefore, when standardizing the collected herbal material data, all biological classifications were considered and assigned identifiers, but only species and infraspecies were used in the experiment.

• full_data

File name	Entity	Amount of data	Note
prescription.csv	prescription	16260
herb_lev_data.csv	herbal material	870	herb-level analysis
comp_lev_data.csv	herbal material	9088	compound-level analysis
compound.csv	compound	48991
gene.csv	gene	25344
phenotype.csv	phenotype	17519
prescription_phenotype.csv	precription-phenotype	66291	interaction
prescription_herb.csv	precription-herb	110497	interaction
herb_phenotype.csv	herb-phenotype	407204	interaction
herb_compound.csv	herb-compound	1628746	interaction
compound_gene.csv	compound-gene	3933224	interaction
gene_phenotype.csv	gene-phenotype	2990372	interaction

• full_ARM_file - preprocessed

File name	Amount of data	Note
ARM_herb_dataframe.csv	16260 (prescriptions) x 628 (herbal materials)
ARM_compound_dataframe.csv	5187 (herbal materials) x 43454 (compounds)
herb_level_DF.csv	16260	dataframe for preprocessing
comp_level_DF.csv	5187	dataframe for preprocessing

• data (Sample)

File name	Entity	Amount of data	Note
prescription.csv	prescription	100
herb_lev_data.csv	herbal material	100	herb-level analysis
comp_lev_data.csv	herbal material	100	compound-level analysis
compound.csv	compound	102
gene.csv	gene	755
phenotype.csv	phenotype	100
prescription_phenotype.csv	precription-phenotype	200	interaction
prescription_herb.csv	precription-herb	163	interaction
herb_phenotype.csv	herb-phenotype	200	interaction
herb_compound.csv	herb-compound	31123	interaction
compound_gene.csv	compound-gene	3291	interaction
gene_phenotype.csv	gene-phenotype	9467	interaction

• ARM_file - preprocessed

File name	Amount of data	Note
ARM_herb_dataframe.csv	17 (prescriptions) x 77 (herbal materials)
ARM_compound_dataframe.csv	7 (herbal materials) x 49 (compounds)
herb_level_DF.csv	17	dataframe for preprocessing
comp_level_DF.csv	7	dataframe for preprocessing

Getting Started

• In order to get InPETM, please clone this repo:

  git clone git@github.com:lapis423/InPETM.git
  cd InPETM

• Then unzip the "InPETM.zip", "envs.zip" files into the current directory (\InPETM), and create environment using the 'InPETM.yaml' file.

  unzip InPETM.zip
  unzip envs.zip
  conda env create --file InPETM.yaml
  conda activate InPETM

• The network proximity calculation requires a biogrid.csv and the construction of a shortest distance database through the PostgreSQL environment. We provide example python code for this below:

  #Creating database
  postgres_config = {
        'dbname': 'postgres',
        'user': 'postgres',
        'password': '1234',
        'host': 'localhost', 
        'port': 5432 
    }
  
  new_db_name = "biogrid"
  
  try:
        connection = psycopg2.connect(**postgres_config)
        connection.set_isolation_level(ISOLATION_LEVEL_AUTOCOMMIT)
        cursor = connection.cursor()
        cursor.execute(f"CREATE DATABASE {new_db_name};")
        print(f"Database '{new_db_name}'is successfully constructed.")
  
    except Exception as e:
        print(f"Error occurred during the database construction: {e}")
  
    finally:
        if cursor:
            cursor.close()
        if connection:
            connection.close()
        print("Connection closed.")
  
  #Creating Table
  aws_postrgesql_url='postgresql://postgres:1234@localhost:5432/biogrid'
    engine_postgresql=create_engine(aws_postgesql_url)
    
    drop_table_query = "DROP TABLE IF EXISTS shortest_paths"
  
    with engine_postgresql.connect() as connection:
        connection.execute(text(drop_table_query))
  
    create_table_query = """
    CREATE TABLE IF NOT EXISTS shortest_paths (
        source TEXT,
        target TEXT,
        distance INTEGER,
        PRIMARY KEY (source, target)
    )
    """
    with engine_postgresql.connect() as connection:
        connection.execute(text(create_table_query))
  
    insert_query = """
    INSERT INTO shortest_paths (source, target, distance)
    VALUES (:source, :target, :distance)
    ON CONFLICT (source, target) DO NOTHING
    """
  
    with engine_postgresql.connect() as connection:
        for source in tqdm(G.nodes()):
            transaction = connection.begin()
            shortest_paths = nx.single_source_shortest_path_length(G, source)
        
            for target, distance in shortest_paths.items():
                connection.execute(
                    text(insert_query),
                    {"source": source, "target": target, "distance": distance}
                )
            transaction.commit() 
  
    print("All shortest paths are saved.")

• Use the following code to run the integrated analysis for the phenotype.

  python main.py phenotype_name True  # In 'phenotype_name', write the phenotype you want to analyze. (e.g., asthma, diabetes)
  python main.py phenotype_name False  # False means to use full data instead of sample data.

Note: Analyzing with full data rather than sample data can take at least several hours, depending on your computer's specifications. If the full_ARM_file directory does not exist, the first run may take even longer.

Examples

Try out the following examples:

• Asthma

  # Sample analysis for asthma
  python main.py asthma True  # For sample data, there may be herbal materials or compounds that do not yield results.

  >>> Inferred Herb combination:  ['Aster tataricus', 'Paeonia lactiflora']
  

  >>> * * * Herb-level Network Analysis Results * * *
  >>>              herb                       related disease target genes  ...   z-score proximal
  0  Paeonia lactiflora  [BCL2, PYHIN1, ESR1, PSAP, TNF, NOS2, BRD2, RA...  ... -14.691928      Yes
  

  >>> * * * Compound-level Network Analysis Results * * *
  >>> compound                       related disease target genes  average distance  ...   z-score  proximal is_in_arm
  0  catechin  [BCL2, PYHIN1, ESR1, PSAP, TNF, NOS2, BRD2, RA...               1.0  ... -14.89503       Yes        No
  

• Stroke

  # Sample analysis for stroke
  python main.py stroke True  # For sample data, there may be herbal materials or compounds that do not yield results.

  >>> Inferred Herb combination:  ['Glycyrrhiza uralensis', 'Paeonia suffruticosa']
  

  >>> * * * Herb-level Network Analysis Results * * *
  >>>                 herb                       related disease target genes  ...   z-score proximal
  0   Paeonia suffruticosa  [SOD1, MTHFR, TRIM29, PLAT, VKORC1, ITGAV, WNK...  ... -6.290607      Yes
  1  Glycyrrhiza uralensis  [SOD1, IL1B, PLAT, ITGB3, ALDH2, ITGA2B, ALB, ...  ... -2.562244      Yes
  

  >>> * * * Compound-level Network Analysis Results * * *
  >>>  compound                       related disease target genes  average distance  ...   z-score  proximal is_in_arm
  1  catechin  [SOD1, MTHFR, TRIM29, PLAT, VKORC1, ITGAV, WNK...             1.500  ... -6.494325       Yes        No
  0  apigenin  [SOD1, IL1B, PLAT, ITGB3, ALDH2, ITGA2B, ALB, ...             1.375  ... -2.475444       Yes       Yes
  

• Try it on your own

  Create a txt file in the same format as the files in the UMLS_code directory. Write the UMLS identifiers (starting with C) of the phenotypes you want to analyze in the txt file, line by line, and save it as the name of the phenotype. Use the name of the file to perform your analysis.

  python main.py phenotype_name False  # For sample data, there may be herbal materials or compounds that do not yield results.