config.py

import math
import os
import torch

"""
Configuration File

This script defines the configuration settings for the GA-DPAMSA framework, including:
- Hyperparameters for Deep Q-Network (DQN) and Genetic Algorithm (GA).
- File paths for datasets, model weights, and results.
- Setup for external multiple sequence alignment (MSA) tools.
- Automatic directory creation to ensure required folders exist.

Author: https://github.com/ZhangLab312/DPAMSA
Co-Author: https://github.com/FLaTNNBio/GA-DPAMSA
"""

# ===========================
# DPAMSA Hyperparameters
# ===========================
GAP_PENALTY = -4  # Penalty for inserting a gap
MISMATCH_PENALTY = -4  # Penalty for a mismatch
MATCH_REWARD = 4  # Reward for a correct match
MAX_EPISODE = 6000  # Maximum number of training episodes
BATCH_SIZE = 128  # Number of experiences sampled per training step
REPLAY_MEMORY_SIZE = 1000  # Capacity of replay memory buffer
ALPHA = 0.0001  # Learning rate for the optimizer
EPSILON = 0.8  # Initial epsilon value for ε-greedy policy
GAMMA = 1  # Discount factor for Q-learning
DELTA = 0.05  # Epsilon decrement step size
DECREMENT_ITERATION = math.ceil(MAX_EPISODE * 0.8 / (EPSILON // DELTA))  # Number of steps to decay epsilon
UPDATE_ITERATION = 128  # Number of iterations before updating the target network
DEVICE_NAME = "cuda:0" if torch.cuda.is_available() else "cpu"  # Auto-detect GPU or CPU
DEVICE = 'cpu'  # Default computation device

# ===========================
# Genetic Algorithm (GA) Parameters
# ===========================
AGENT_WINDOW_ROW = 3  # Number of rows in the agent's observation window
AGENT_WINDOW_COLUMN = 30  # Number of columns in the observation window
GA_ITERATIONS = 3  # Number of iterations for genetic evolution
POPULATION_SIZE = 5  # Population size for genetic algorithm
CLONE_RATE = 0.25  # % of the population to be an exact copy of the input sequences during Population Generation Phase
GAP_RATE = 0.05  # % of Gap to be added to an individual during Population Generation Phase (calculated on seq. length)
SELECTION_RATE = 0.5  # % of the population to be selected following a certain criteria
MUTATION_RATE = 0.25  # % of the population undergo mutation


# Ensure hyperparameter constraints
assert 0 < BATCH_SIZE <= REPLAY_MEMORY_SIZE, "batch size must be in the range of 0 to the size of replay memory."
assert ALPHA > 0, "alpha must be greater than 0."
assert 0 <= GAMMA <= 1, "gamma must be in the range of 0 to 1."
assert 0 <= EPSILON <= 1, "epsilon must be in the range of 0 to 1."
assert 0 <= DELTA <= EPSILON, "delta must be in the range of 0 to epsilon."
assert 0 < DECREMENT_ITERATION, "decrement iteration must be greater than 0."


# ===========================
# File Paths Configuration
# ===========================
PROJECT_ROOT = os.path.dirname(os.path.abspath(__file__))

# Dataset Paths
BASE_DATASETS_PATH = os.path.join(PROJECT_ROOT, "datasets")
FASTA_FILES_PATH = os.path.join(BASE_DATASETS_PATH, "fasta_files")
TRAINING_DATASET_PATH = os.path.join(BASE_DATASETS_PATH, "training_dataset")
INFERENCE_DATASET_PATH = os.path.join(BASE_DATASETS_PATH, "inference_dataset")

# Model Weights Path
DPAMSA_WEIGHTS_PATH = os.path.join(PROJECT_ROOT, "DPAMSA", "weights")

# Tensorboard Training Runs Path
RUNS_PATH = os.path.join(PROJECT_ROOT, "DPAMSA", "runs")

# Results Paths
BASE_RESULTS_PATH = os.path.join(PROJECT_ROOT, "results")
REPORTS_PATH = os.path.join(BASE_RESULTS_PATH, "reports")
DPAMSA_REPORTS_PATH = os.path.join(REPORTS_PATH, "DPAMSA")
GA_DPAMSA_REPORTS_PATH = os.path.join(REPORTS_PATH, "GA-DPAMSA")
DATASETS_REPORTS_PATH = os.path.join(REPORTS_PATH, "datasets")
BENCHMARKS_PATH = os.path.join(BASE_RESULTS_PATH, "benchmarks")
TOOLS_OUTPUT_PATH = os.path.join(BASE_RESULTS_PATH, "tools_output")
CSV_PATH = os.path.join(BENCHMARKS_PATH, "csv")
DATASETS_CSV_PATH = os.path.join(CSV_PATH, "datasets")
INFERENCE_CSV_PATH = os.path.join(CSV_PATH, "inference")
DPAMSA_INF_CSV_PATH = os.path.join(INFERENCE_CSV_PATH, "DPAMSA")
GA_DPAMSA_INF_CSV_PATH = os.path.join(INFERENCE_CSV_PATH, "GA-DPAMSA")
CHARTS_PATH = os.path.join(BENCHMARKS_PATH, "charts")

# Ensure directories exist, creating them if they don't
REQUIRED_DIRECTORIES = [
    DPAMSA_WEIGHTS_PATH,
    RUNS_PATH,
    BASE_RESULTS_PATH,
    DPAMSA_REPORTS_PATH,
    GA_DPAMSA_REPORTS_PATH,
    DATASETS_REPORTS_PATH,
    BENCHMARKS_PATH,
    TOOLS_OUTPUT_PATH,
    CSV_PATH,
    DATASETS_CSV_PATH,
    INFERENCE_CSV_PATH,
    DPAMSA_INF_CSV_PATH,
    GA_DPAMSA_INF_CSV_PATH,
    CHARTS_PATH
]
for path in REQUIRED_DIRECTORIES:
    if not os.path.exists(path):
        os.makedirs(path)


# ===========================
# External MSA Tools Configuration
# ===========================
TOOLS = {
    'ClustalOmega': {
        'command': lambda file_path, output_dir: ['clustalo', '-i', file_path, '-o', output_dir],
        'output_dir': os.path.join(TOOLS_OUTPUT_PATH, 'ClustalOmega'),
        'report_dir': os.path.join(REPORTS_PATH, 'ClustalOmega')
    },
    'MSAProbs': {
        'command': lambda file_path, output_dir: ['msaprobs', file_path, '-o', output_dir],
        'output_dir': os.path.join(TOOLS_OUTPUT_PATH, 'MSAProbs'),
        'report_dir': os.path.join(REPORTS_PATH, 'MSAProbs')
    },
    'ClustalW': {
        'command': lambda file_path, output_dir: ['clustalw', f'-INFILE={file_path}',
                                                  '-OUTPUT=FASTA', f'-OUTFILE={output_dir}'],
        'output_dir': os.path.join(TOOLS_OUTPUT_PATH, 'ClustalW'),
        'report_dir': os.path.join(REPORTS_PATH, 'ClustalW')
    },
    'MAFFT': {
        'command': lambda file_path, output_dir: f"mafft --auto {file_path} > {output_dir}",
        'output_dir': os.path.join(TOOLS_OUTPUT_PATH, 'MAFFT'),
        'report_dir': os.path.join(REPORTS_PATH, 'MAFFT')
    },
    'MUSCLE5': {
        'command': lambda file_path, output_dir: ['muscle5', '-align', file_path, '-output', output_dir],
        'output_dir': os.path.join(TOOLS_OUTPUT_PATH, 'MUSCLE5'),
        'report_dir': os.path.join(REPORTS_PATH, 'MUSCLE5')
    },
    'UPP': {
        'command': lambda file_path, output_dir: ['run_upp.py', '-s', file_path, '-m', 'dna', '-d', output_dir],
        'output_dir': os.path.join(TOOLS_OUTPUT_PATH, 'UPP'),
        'report_dir': os.path.join(REPORTS_PATH, 'UPP')
    },
    'PASTA': {
        'command': lambda file_path, output_dir: ['run_pasta.py', '-i', file_path, '-o', output_dir],
        'output_dir': os.path.join(TOOLS_OUTPUT_PATH, 'PASTA'),
        'report_dir': os.path.join(REPORTS_PATH, 'PASTA')
    }
}