comparison COBRAxy/src/flux_simulation.py @ 539:2fb97466e404 draft

Uploaded

539:2fb97466e404

"""
Flux sampling and analysis utilities for COBRA models.

This script supports two modes:
- Mode 1 (model_and_bounds=True): load a base model and apply bounds from
    separate files before sampling.
- Mode 2 (model_and_bounds=False): load complete models and sample directly.

Sampling algorithms supported: OPTGP and CBS. Outputs include flux samples
and optional analyses (pFBA, FVA, sensitivity), saved as tabular files.
"""

import argparse
import utils.general_utils as utils
from typing import List
import os
import pandas as pd
import numpy as np
import cobra
import utils.CBS_backend as CBS_backend
from joblib import Parallel, delayed, cpu_count
from cobra.sampling import OptGPSampler
import sys
import utils.model_utils as model_utils


################################# process args ###############################
def process_args(args: List[str] = None) -> argparse.Namespace:
    """
    Processes command-line arguments.
    
    Args:
        args (list): List of command-line arguments.
    
    Returns:
        Namespace: An object containing parsed arguments.
    """
    parser = argparse.ArgumentParser(usage='%(prog)s [options]',
                                     description='process some value\'s')
    
    parser.add_argument("-mo", "--model_upload", type=str,
        help="path to input file with custom rules, if provided")
    
    parser.add_argument("-mab", "--model_and_bounds", type=str,
        choices=['True', 'False'],
        required=True,
        help="upload mode: True for model+bounds, False for complete models")
    
    parser.add_argument("-ens", "--sampling_enabled", type=str,
        choices=['true', 'false'],
        required=True,
        help="enable sampling: 'true' for sampling, 'false' for no sampling")
    
    parser.add_argument('-ol', '--out_log',
                        help="Output log")
    
    parser.add_argument('-td', '--tool_dir',
                        type=str,
                        required=True,
                        help='your tool directory')
    
    parser.add_argument('-in', '--input',
                        required=True,
                        type=str,
                        help='input bounds files or complete model files')
    
    parser.add_argument('-ni', '--name',
                        required=True,
                        type=str,
                        help='cell names')
    
    parser.add_argument('-a', '--algorithm',
                        type=str,
                        choices=['OPTGP', 'CBS'],
                        required=True,
                        help='choose sampling algorithm')
    
    parser.add_argument('-th', '--thinning',
                        type=int,
                        default=100,
                        required=True,
                        help='choose thinning')
    
    parser.add_argument('-ns', '--n_samples',
                        type=int,
                        required=True,
                        help='choose how many samples (set to 0 for optimization only)')
    
    parser.add_argument('-sd', '--seed',
                        type=int,
                        required=True,
                        help='seed for random number generation')
    
    parser.add_argument('-nb', '--n_batches',
                        type=int,
                        required=True,
                        help='choose how many batches')
    
    parser.add_argument('-opt', '--perc_opt',
                        type=float,
                        default=0.9,
                        required=False,
                        help='choose the fraction of optimality for FVA (0-1)')
    
    parser.add_argument('-ot', '--output_type',
                        type=str,
                        required=True,
                        help='output type for sampling results')
    
    parser.add_argument('-ota', '--output_type_analysis',
                        type=str,
                        required=False,
                        help='output type analysis (optimization methods)')

    parser.add_argument('-idop', '--output_path',
                        type=str,
                        default='flux_simulation/',
                        help = 'output path for fluxes')
    
    parser.add_argument('-otm', '--out_mean',
                    type = str,
                    required=False,
                    help = 'output of mean of fluxes')
    
    parser.add_argument('-otmd', '--out_median',
                    type = str,
                    required=False,
                    help = 'output of median of fluxes')

    parser.add_argument('-otq', '--out_quantiles',
                    type = str,
                    required=False,
                    help = 'output of quantiles of fluxes')
    
    parser.add_argument('-otfva', '--out_fva',
                    type = str, 
                    required=False,
                    help = 'output of FVA results')
    parser.add_argument('-otp', '--out_pfba',
                    type = str,
                    required=False,
                    help = 'output of pFBA results')
    parser.add_argument('-ots', '--out_sensitivity',
                    type = str,
                    required=False,
                    help = 'output of sensitivity results')
    ARGS = parser.parse_args(args)
    return ARGS
########################### warning ###########################################
def warning(s :str) -> None:
    """
    Log a warning message to an output log file and print it to the console.

    Args:
        s (str): The warning message to be logged and printed.
    
    Returns:
      None
    """
    with open(ARGS.out_log, 'a') as log:
        log.write(s + "\n\n")
    print(s)


def write_to_file(dataset: pd.DataFrame, path: str, keep_index:bool=False, name:str=None)->None:
    """
    Write a DataFrame to a TSV file under path with a given base name.

    Args:
        dataset: The DataFrame to write.
        name: Base file name (without extension). If None, 'path' is treated as the full file path.
        path: Directory path where the file will be saved.
        keep_index: Whether to keep the DataFrame index in the file.

    Returns:
        None
    """
    dataset.index.name = 'Reactions'
    if name:
        dataset.to_csv(os.path.join(path, name + ".csv"), sep = '\t', index = keep_index)
    else:
        dataset.to_csv(path, sep = '\t', index = keep_index)

############################ dataset input ####################################
def read_dataset(data :str, name :str) -> pd.DataFrame:
    """
    Read a dataset from a CSV file and return it as a pandas DataFrame.

    Args:
        data (str): Path to the CSV file containing the dataset.
        name (str): Name of the dataset, used in error messages.

    Returns:
        pandas.DataFrame: DataFrame containing the dataset.

    Raises:
        pd.errors.EmptyDataError: If the CSV file is empty.
        sys.exit: If the CSV file has the wrong format, the execution is aborted.
    """
    try:
        dataset = pd.read_csv(data, sep = '\t', header = 0, index_col=0, engine='python')
    except pd.errors.EmptyDataError:
        sys.exit('Execution aborted: wrong format of ' + name + '\n')
    if len(dataset.columns) < 2:
        sys.exit('Execution aborted: wrong format of ' + name + '\n')
    return dataset



def OPTGP_sampler(model: cobra.Model, model_name: str, n_samples: int = 1000, thinning: int = 100, n_batches: int = 1, seed: int = 0) -> None:
    """
    Samples from the OPTGP (Optimal Global Perturbation) algorithm and saves the results to CSV files.
    
    Args:
        model (cobra.Model): The COBRA model to sample from.
        model_name (str): The name of the model, used in naming output files.
        n_samples (int, optional): Number of samples per batch. Default is 1000.
        thinning (int, optional): Thinning parameter for the sampler. Default is 100.
        n_batches (int, optional): Number of batches to run. Default is 1.
        seed (int, optional): Random seed for reproducibility. Default is 0.
        
    Returns:
        None
    """
    import numpy as np
    
    # Get reaction IDs for consistent column ordering
    reaction_ids = [rxn.id for rxn in model.reactions]
    
    # Sample and save each batch as numpy file
    for i in range(n_batches):
        optgp = OptGPSampler(model, thinning, seed)
        samples = optgp.sample(n_samples)
        
        # Save as numpy array (more memory efficient)
        batch_filename = f"{ARGS.output_path}/{model_name}_{i}_OPTGP.npy"
        np.save(batch_filename, samples.to_numpy())
        
        seed += 1
    
    # Merge all batches into a single DataFrame
    all_samples = []
    
    for i in range(n_batches):
        batch_filename = f"{ARGS.output_path}/{model_name}_{i}_OPTGP.npy"
        batch_data = np.load(batch_filename, allow_pickle=True)
        all_samples.append(batch_data)
    
    # Concatenate all batches
    samplesTotal_array = np.vstack(all_samples)
    
    # Convert back to DataFrame with proper column names
    samplesTotal = pd.DataFrame(samplesTotal_array, columns=reaction_ids)
    
    # Save the final merged result as CSV
    write_to_file(samplesTotal.T, ARGS.output_path, True, name=model_name)
    
    # Clean up temporary numpy files
    for i in range(n_batches):
        batch_filename = f"{ARGS.output_path}/{model_name}_{i}_OPTGP.npy"
        if os.path.exists(batch_filename):
            os.remove(batch_filename)


def CBS_sampler(model: cobra.Model, model_name: str, n_samples: int = 1000, n_batches: int = 1, seed: int = 0) -> None:
    """
    Samples using the CBS (Constraint-based Sampling) algorithm and saves the results to CSV files.
    
    Args:
        model (cobra.Model): The COBRA model to sample from.
        model_name (str): The name of the model, used in naming output files.
        n_samples (int, optional): Number of samples per batch. Default is 1000.
        n_batches (int, optional): Number of batches to run. Default is 1.
        seed (int, optional): Random seed for reproducibility. Default is 0.
        
    Returns:
        None
    """
    import numpy as np
    
    # Get reaction IDs for consistent column ordering
    reaction_ids = [reaction.id for reaction in model.reactions]
    
    # Perform FVA analysis once for all batches
    df_FVA = cobra.flux_analysis.flux_variability_analysis(model, fraction_of_optimum=0).round(6)
    
    # Generate random objective functions for all samples across all batches
    df_coefficients = CBS_backend.randomObjectiveFunction(model, n_samples * n_batches, df_FVA, seed=seed)
    
    # Sample and save each batch as numpy file
    for i in range(n_batches):
        samples = pd.DataFrame(columns=reaction_ids, index=range(n_samples))
        
        try:
            CBS_backend.randomObjectiveFunctionSampling(
                model, 
                n_samples, 
                df_coefficients.iloc[:, i * n_samples:(i + 1) * n_samples], 
                samples
            )
        except Exception as e:
            utils.logWarning(
                f"Warning: GLPK solver has failed for {model_name}. Trying with COBRA interface. Error: {str(e)}",
                ARGS.out_log
            )
            CBS_backend.randomObjectiveFunctionSampling_cobrapy(
                model, 
                n_samples, 
                df_coefficients.iloc[:, i * n_samples:(i + 1) * n_samples],
                samples
            )
        
        # Save as numpy array (more memory efficient)
        batch_filename = f"{ARGS.output_path}/{model_name}_{i}_CBS.npy"
        utils.logWarning(batch_filename, ARGS.out_log)
        np.save(batch_filename, samples.to_numpy())
    
    # Merge all batches into a single DataFrame
    all_samples = []
    
    for i in range(n_batches):
        batch_filename = f"{ARGS.output_path}/{model_name}_{i}_CBS.npy"
        batch_data = np.load(batch_filename, allow_pickle=True)
        all_samples.append(batch_data)
    
    # Concatenate all batches
    samplesTotal_array = np.vstack(all_samples)
    
    # Convert back to DataFrame with proper column namesq
    samplesTotal = pd.DataFrame(samplesTotal_array, columns=reaction_ids)
    
    # Save the final merged result as CSV
    write_to_file(samplesTotal.T, ARGS.output_path, True, name=model_name)
    
    # Clean up temporary numpy files
    for i in range(n_batches):
        batch_filename = f"{ARGS.output_path}/{model_name}_{i}_CBS.npy"
        if os.path.exists(batch_filename):
            os.remove(batch_filename)



def model_sampler_with_bounds(model_input_original: cobra.Model, bounds_path: str, cell_name: str) -> List[pd.DataFrame]:
    """
    MODE 1: Prepares the model with bounds from separate bounds file and performs sampling.

    Args:
        model_input_original (cobra.Model): The original COBRA model.
        bounds_path (str): Path to the CSV file containing the bounds dataset.
        cell_name (str): Name of the cell, used to generate filenames for output.

    Returns:
        List[pd.DataFrame]: A list of DataFrames containing statistics and analysis results.
    """

    model_input = model_input_original.copy()
    bounds_df = read_dataset(bounds_path, "bounds dataset")
    
    # Apply bounds to model
    for rxn_index, row in bounds_df.iterrows():
        try:
            model_input.reactions.get_by_id(rxn_index).lower_bound = row.lower_bound
            model_input.reactions.get_by_id(rxn_index).upper_bound = row.upper_bound
        except KeyError:
            warning(f"Warning: Reaction {rxn_index} not found in model. Skipping.")
    
    return perform_sampling_and_analysis(model_input, cell_name)


def perform_sampling_and_analysis(model_input: cobra.Model, cell_name: str) -> List[pd.DataFrame]:
    """
    Common function to perform sampling and analysis on a prepared model.

    Args:
        model_input (cobra.Model): The prepared COBRA model with bounds applied.
        cell_name (str): Name of the cell, used to generate filenames for output.

    Returns:
        List[pd.DataFrame]: A list of DataFrames containing statistics and analysis results.
    """

    returnList = []

    if ARGS.sampling_enabled == "true":
    
        if ARGS.algorithm == 'OPTGP':
            OPTGP_sampler(model_input, cell_name, ARGS.n_samples, ARGS.thinning, ARGS.n_batches, ARGS.seed)
        elif ARGS.algorithm == 'CBS':
            CBS_sampler(model_input, cell_name, ARGS.n_samples, ARGS.n_batches, ARGS.seed)

        df_mean, df_median, df_quantiles = fluxes_statistics(cell_name, ARGS.output_types)

        if("fluxes" not in ARGS.output_types):
            os.remove(ARGS.output_path + "/" + cell_name + '.csv')

        returnList = [df_mean, df_median, df_quantiles]

    df_pFBA, df_FVA, df_sensitivity = fluxes_analysis(model_input, cell_name, ARGS.output_type_analysis)

    if("pFBA" in ARGS.output_type_analysis):
        returnList.append(df_pFBA)
    if("FVA" in ARGS.output_type_analysis):
        returnList.append(df_FVA)
    if("sensitivity" in ARGS.output_type_analysis):
        returnList.append(df_sensitivity)

    return returnList

def fluxes_statistics(model_name: str,  output_types:List)-> List[pd.DataFrame]:
    """
    Computes statistics (mean, median, quantiles) for the fluxes.

    Args:
        model_name (str): Name of the model, used in filename for input.
        output_types (List[str]): Types of statistics to compute (mean, median, quantiles).

    Returns:
        List[pd.DataFrame]: List of DataFrames containing mean, median, and quantiles statistics.
    """

    df_mean = pd.DataFrame()
    df_median= pd.DataFrame()
    df_quantiles= pd.DataFrame()

    df_samples = pd.read_csv(ARGS.output_path + "/"  +  model_name + '.csv', sep = '\t', index_col = 0).T
    df_samples = df_samples.round(8)

    for output_type in output_types:
        if(output_type == "mean"):
            df_mean = df_samples.mean()
            df_mean = df_mean.to_frame().T
            df_mean = df_mean.reset_index(drop=True)
            df_mean.index = [model_name]
        elif(output_type == "median"):
            df_median = df_samples.median()
            df_median = df_median.to_frame().T
            df_median = df_median.reset_index(drop=True)
            df_median.index = [model_name]
        elif(output_type == "quantiles"):
            newRow = []
            cols = []
            for rxn in df_samples.columns:
                quantiles = df_samples[rxn].quantile([0.25, 0.50, 0.75])
                newRow.append(quantiles[0.25])
                cols.append(rxn + "_q1")
                newRow.append(quantiles[0.5])
                cols.append(rxn + "_q2")
                newRow.append(quantiles[0.75])
                cols.append(rxn + "_q3")
            df_quantiles = pd.DataFrame(columns=cols)
            df_quantiles.loc[0] = newRow
            df_quantiles = df_quantiles.reset_index(drop=True)
            df_quantiles.index = [model_name]
    
    return df_mean, df_median, df_quantiles

def fluxes_analysis(model:cobra.Model,  model_name:str, output_types:List)-> List[pd.DataFrame]:
    """
    Performs flux analysis including pFBA, FVA, and sensitivity analysis. The objective function
    is assumed to be already set in the model.

    Args:
        model (cobra.Model): The COBRA model to analyze.
        model_name (str): Name of the model, used in filenames for output.
        output_types (List[str]): Types of analysis to perform (pFBA, FVA, sensitivity).

    Returns:
        List[pd.DataFrame]: List of DataFrames containing pFBA, FVA, and sensitivity analysis results.
    """

    df_pFBA = pd.DataFrame()
    df_FVA= pd.DataFrame()
    df_sensitivity= pd.DataFrame()

    for output_type in output_types:
        if(output_type == "pFBA"):
            solution = cobra.flux_analysis.pfba(model)
            fluxes = solution.fluxes
            df_pFBA.loc[0,[rxn.id for rxn in model.reactions]] = fluxes.tolist()
            df_pFBA = df_pFBA.reset_index(drop=True)
            df_pFBA.index = [model_name]
            df_pFBA = df_pFBA.astype(float).round(6)
        elif(output_type == "FVA"):
            fva = cobra.flux_analysis.flux_variability_analysis(model, fraction_of_optimum=ARGS.perc_opt, processes=1).round(8)
            columns = []
            for rxn in fva.index.to_list():
                columns.append(rxn + "_min")
                columns.append(rxn + "_max")
            df_FVA= pd.DataFrame(columns = columns)
            for index_rxn, row in fva.iterrows():
                df_FVA.loc[0, index_rxn+ "_min"] = fva.loc[index_rxn, "minimum"]
                df_FVA.loc[0, index_rxn+ "_max"] = fva.loc[index_rxn, "maximum"]
            df_FVA = df_FVA.reset_index(drop=True)
            df_FVA.index = [model_name]
            df_FVA = df_FVA.astype(float).round(6)
        elif(output_type == "sensitivity"):
            solution_original = model.optimize().objective_value
            reactions = model.reactions
            single = cobra.flux_analysis.single_reaction_deletion(model)
            newRow = []
            df_sensitivity = pd.DataFrame(columns = [rxn.id for rxn in reactions], index = [model_name])
            for rxn in reactions:
                newRow.append(single.knockout[rxn.id].growth.values[0]/solution_original)
            df_sensitivity.loc[model_name] = newRow
            df_sensitivity = df_sensitivity.astype(float).round(6)
    return df_pFBA, df_FVA, df_sensitivity

############################# main ###########################################
def main(args: List[str] = None) -> None:
    """
    Initialize and run sampling/analysis based on the frontend input arguments.

    Returns:
        None
    """

    num_processors = max(1, cpu_count() - 1)

    global ARGS
    ARGS = process_args(args)

    if not os.path.exists('flux_simulation'):
        os.makedirs('flux_simulation')

    # --- Normalize inputs (the tool may pass comma-separated --input and either --name or --names) ---
    ARGS.input_files = ARGS.input.split(",") if ARGS.input else []
    ARGS.file_names = ARGS.name.split(",")
    # output types (required) -> list
    ARGS.output_types = ARGS.output_type.split(",") if ARGS.output_type else []
    # optional analysis output types -> list or empty
    ARGS.output_type_analysis = ARGS.output_type_analysis.split(",") if ARGS.output_type_analysis else []

    # Determine if sampling should be performed
    if ARGS.sampling_enabled == "true":
        perform_sampling = True
    else:
        perform_sampling = False

    print("=== INPUT FILES ===")
    print(f"{ARGS.input_files}")
    print(f"{ARGS.file_names}")
    print(f"{ARGS.output_type}")
    print(f"{ARGS.output_types}")
    print(f"{ARGS.output_type_analysis}")
    print(f"Sampling enabled: {perform_sampling} (n_samples: {ARGS.n_samples})")
    
    if ARGS.model_and_bounds == "True":
        # MODE 1: Model + bounds (separate files)
        print("=== MODE 1: Model + Bounds (separate files) ===")
        
        # Load base model
        if not ARGS.model_upload:
            sys.exit("Error: model_upload is required for Mode 1")

        base_model = model_utils.build_cobra_model_from_csv(ARGS.model_upload)

        validation = model_utils.validate_model(base_model)

        print("\n=== MODEL VALIDATION ===")
        for key, value in validation.items():
            print(f"{key}: {value}")

        # Set solver verbosity to 1 to see warning and error messages only.
        base_model.solver.configuration.verbosity = 1

        # Process each bounds file with the base model
        results = Parallel(n_jobs=num_processors)(
            delayed(model_sampler_with_bounds)(base_model, bounds_file, cell_name) 
            for bounds_file, cell_name in zip(ARGS.input_files, ARGS.file_names)
        )

    else:
        # MODE 2: Multiple complete models
        print("=== MODE 2: Multiple complete models ===")
        
        # Process each complete model file
        results = Parallel(n_jobs=num_processors)(
            delayed(perform_sampling_and_analysis)(model_utils.build_cobra_model_from_csv(model_file), cell_name) 
            for model_file, cell_name in zip(ARGS.input_files, ARGS.file_names)
        )

    # Handle sampling outputs (only if sampling was performed)
    if perform_sampling:
        print("=== PROCESSING SAMPLING RESULTS ===")
        
        all_mean = pd.concat([result[0] for result in results], ignore_index=False)
        all_median = pd.concat([result[1] for result in results], ignore_index=False)
        all_quantiles = pd.concat([result[2] for result in results], ignore_index=False)

        if "mean" in ARGS.output_types:
            all_mean = all_mean.fillna(0.0)
            all_mean = all_mean.sort_index()
            write_to_file(all_mean.T, ARGS.out_mean, True)

        if "median" in ARGS.output_types:
            all_median = all_median.fillna(0.0)
            all_median = all_median.sort_index()
            write_to_file(all_median.T, ARGS.out_median, True)
        
        if "quantiles" in ARGS.output_types:
            all_quantiles = all_quantiles.fillna(0.0)
            all_quantiles = all_quantiles.sort_index()
            write_to_file(all_quantiles.T, ARGS.out_quantiles, True)
    else:
        print("=== SAMPLING SKIPPED (n_samples = 0 or sampling disabled) ===")

    # Handle optimization analysis outputs (always available)
    print("=== PROCESSING OPTIMIZATION RESULTS ===")
    
    # Determine the starting index for optimization results
    # If sampling was performed, optimization results start at index 3
    # If no sampling, optimization results start at index 0
    index_result = 3 if perform_sampling else 0
    
    if "pFBA" in ARGS.output_type_analysis:
        all_pFBA = pd.concat([result[index_result] for result in results], ignore_index=False)
        all_pFBA = all_pFBA.sort_index()
        write_to_file(all_pFBA.T, ARGS.out_pfba, True)
        index_result += 1
        
    if "FVA" in ARGS.output_type_analysis:
        all_FVA = pd.concat([result[index_result] for result in results], ignore_index=False)
        all_FVA = all_FVA.sort_index()
        write_to_file(all_FVA.T, ARGS.out_fva, True)
        index_result += 1
        
    if "sensitivity" in ARGS.output_type_analysis:
        all_sensitivity = pd.concat([result[index_result] for result in results], ignore_index=False)
        all_sensitivity = all_sensitivity.sort_index()
        write_to_file(all_sensitivity.T, ARGS.out_sensitivity, True)

    return
        
##############################################################################
if __name__ == "__main__":
    main()