使用Keras构建基于LSTM的DNA序列分类模型

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
from sklearn.model_selection import train_test_split
from keras.models import Sequential
from keras.layers import Dense, Flatten, Conv1D, MaxPooling1D, LSTM
from keras.utils import to_categorical

def read_fasta(file):
    seqs = []
    labels = []
    with open(file, 'r') as f:
        for line in f:
            if line.startswith('>'):
                labels.append(line.strip()[1:])
            else:
                seqs.append(line.strip())
    label_dict = {l:i for i,l in enumerate(set(labels))}
    label_indices = [label_dict[l] for l in labels]
    seqs = np.array(seqs)[np.argsort(label_indices)]
    labels = np.array(label_indices)
    return seqs, labels

def seq_to_num(seqs):
    seq_num = []
    for seq in seqs:
        seq_num.append([int(i) for i in seq.replace('A', '1').replace('C', '2').replace('G', '3').replace('T', '4')])
    return np.array(seq_num)

def encode_labels(labels):
    label_dict = {l:i for i,l in enumerate(set(labels))}
    encoded_labels = [label_dict[l] for l in labels]
    return to_categorical(encoded_labels)

def create_model():
    model = Sequential()
    model.add(Conv1D(filters=32, kernel_size=3, activation='relu', input_shape=(1000, 1)))
    model.add(MaxPooling1D(pool_size=2))
    model.add(Conv1D(filters=64, kernel_size=3, activation='relu'))
    model.add(MaxPooling1D(pool_size=2))
    model.add(Flatten())
    model.add(Reshape((-1, 64)))  # 新增的Reshape层
    model.add(LSTM(64))
    model.add(Dense(64, activation='relu'))
    model.add(Dense(2, activation='softmax'))
    model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])
    return model

seqs, labels = read_fasta('CP015726.fasta')
seq_num = seq_to_num(seqs)
encoded_labels = encode_labels(labels)
X_train, X_test, y_train, y_test = train_test_split(seq_num, encoded_labels, test_size=0.2, random_state=42)
model = create_model()
history = model.fit(X_train.reshape((-1, 1000, 1)), y_train, epochs=10, batch_size=32, validation_split=0.2)
score = model.evaluate(X_test.reshape((-1, 1000, 1)), y_test, verbose=0)
print('Test loss:', score[0])
print('Test accuracy:', score[1])

代码修改说明：

在 read_fasta 函数中，将 labels 转换为整数索引，然后根据索引对 seqs 进行排序，而不是将整个 labels 数组排序。

修改后的 read_fasta 函数：

def read_fasta(file):
    seqs = []
    labels = []
    with open(file, 'r') as f:
        for line in f:
            if line.startswith('>'):
                labels.append(line.strip()[1:])
            else:
                seqs.append(line.strip())
    label_dict = {l:i for i,l in enumerate(set(labels))}
    label_indices = [label_dict[l] for l in labels]
    seqs = np.array(seqs)[np.argsort(label_indices)]
    labels = np.array(label_indices)
    return seqs, labels

修改后，代码可以正常运行，并训练出基于LSTM的DNA序列分类模型。