Poisson stepperのステップ検出アルゴリズム

スクリプト

• ポアソン過程でステップするトラジェクトリのステップ地点を検出するアルゴリズム
• "Assembly dynamics of microtubules at molecular resolution" Nature (2006) 709-712 を参考に作成
• 一番大きいステップをはじめに検出して、だんだん小さいステップを足していくスタイル
• このアルゴリズムのいいところはフリーパラメータが実質ゼロでCalibrationの必要がないところ
• 前に作ったpoisson_stepper.pyを使用してテスト

chi2.py

#!/usr/bin/python                                                                                                                                                                                                                                                                          

import matplotlib.pyplot as plt
import numpy as np
import poisson_stepper
from scipy.optimize import curve_fit
import time
import timer


def detect_step(trajectory):
    # Initialize                                                                                                                                                                                                                                                                           
    min_chi2 = float("inf")
    min_t = 0
    delta_times_length = 0.0
    n = len(trajectory)

    for i in range(1, n - 1):
        # Array preparation                                                                                                                                                                                                                                                                
        left = trajectory[0:i]
        right = trajectory[i:n]

        # Calculate Chi2                                                                                                                                                                                                                                                                   
        current_chi2 = len(left) * np.var(left) + len(right) * np.var(right)

        # Update minimum values                                                                                                                                                                                                                                                            
        if min_chi2 > current_chi2:
            min_chi2 = current_chi2
            min_t = i
            delta_times_length = abs(np.average(left) - np.average(right)) * n

    return min_t, delta_times_length


def plot(trajectory, border, average_trajectory):

    plt.plot(trajectory)
    for i in range(1, len(border)):
        x = np.arange(border[i-1], border[i], 0.1)
        y_value = average_trajectory[i-1]
        y = [y_value for _ in range(0, len(x))]
        plt.plot(x, y, color='k', linewidth=2.0)

        if i == len(border) - 1:
            break

        y = np.arange(average_trajectory[i-1], average_trajectory[i], 0.1)
        x = [border[i] for _ in range(0, len(y))]
        plt.plot(x, y, color='k', linewidth=2.0)

    plt.show()


def detect_step_all_ranges(trajectory, border):
    max_delta_times_length = 0.0
    max_step_t = 0

    for iborder in range(0, len(border)-1):
    start = border[iborder] + 1
        end   = border[iborder+1] - 1
        trajectory_to_process = trajectory[start:end]
        step_t, delta_times_length = detect_step(trajectory_to_process)
        step_t += start

        if max_delta_times_length < delta_times_length:
            max_delta_times_length = delta_times_length
            max_step_t = step_t

    return max_step_t


def chi2(trajectory, t):

    # Calculate border                                                                                                                                                                                                                                                                     
    border = [0, len(trajectory)]

    for i in range(0, 10):
        border.append(detect_step_all_ranges(trajectory, border))
        border = sorted(border)

    # Calculate average trajectory                                                                                                                                                                                                                                                         
    average_trajectory = [np.average(trajectory[border[i]:border[i+1]])
                          for i in range(0, len(border) - 1)]

    # Plot                                                                                                                                                                                                                                                                                 
    plot(trajectory, border, average_trajectory)

    # Calculate step size                                                                                                                                                                                                                                                                  
    step_size = [average_trajectory[i+1] - average_trajectory[i]
                 for i in range(0, len(average_trajectory)-1)]

    return step_size


if __name__ == "__main__":
    trajectory, t = poisson_stepper.poisson_stepper(sigma=10.0)
    step_size = chi2(trajectory, t)

計算結果

• 青線が模擬実験データ、黒線がフィッティング
• まぁまぁ、いい精度で検出できている