vis/kcc_vis.py

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
from scipy.stats import pearsonr, kendalltau
import seaborn as sns


def get_label_dict():
    """
    返回一个字典
    """
    df = pd.read_excel('../dataset/r2_case.xlsx')
    df = df[~pd.isna(df['MIC GM (ug/ml)'])]
    values = {}
    for _, row in df.iterrows():
        desc = row['SEQUENCE - D-type amino acid substitution']
        value = row['MIC GM (ug/ml)']
        values[desc] = value
    template = list(values.keys())[0].upper()
    t_value = values[template]
    tags = {}
    for mutation, value in values.items():
        if mutation != template:
            tags[mutation] = np.log2(value / t_value)
        
    return tags

# 设置学术风格
sns.set_theme(style="whitegrid")

predicted_csv_files = ["vanilla.csv", "sfda.csv", "uda.csv"]  # 预测值CSV文件列表
titles = ["Vanilla", "SFDA", "UDA"]

# 读取真值
true_data = get_label_dict()

# 创建子图
fig, axes = plt.subplots(1, len(predicted_csv_files), figsize=(3.9 * len(predicted_csv_files), 4), sharex=True, sharey=True)
if len(predicted_csv_files) == 1:
    axes = [axes]  # 确保 axes 是可迭代的

# 遍历预测值文件并绘制子图
for i, csv_file in enumerate(predicted_csv_files):
    # 读取当前预测值
    pred_pd = pd.read_csv(f'./kcc/{csv_file}', index_col='seq')
    true_values, predicted_values, seqs = [], [], []
    for seq, true_value in true_data.items():
        true_values.append(true_value)
        predicted_values.append(pred_pd.loc[seq]['pred'])
        seqs.append(seq)
    
    # 计算PCC
    corr_p, _ = pearsonr(true_values, predicted_values)
    corr_k, _ = kendalltau(true_values, predicted_values)

    
    # 绘制散点图
    ax = axes[i]
    ax.plot([-1, 2.5], [-1, 2.5], color='gray', linestyle='--', linewidth=1)
    ax.scatter(true_values, predicted_values, alpha=0.7)
    ax.set_title(titles[i], fontsize=14)
    ax.set_xlabel("True Log2 MIC Ratio", fontsize=12)
    ax.set_xticks(np.arange(-1, 2.6, 0.5))
    ax.set_ylabel("Predicted Log2 MIC Ratio", fontsize=12)
    ax.set_yticks(np.arange(-1, 2.6, 0.5))
    ax.tick_params(labelleft=True)
    ax.text(0.40, 0.03, f'PCC = {corr_p:.3f}\nKCC = {corr_k:.3f}', transform=ax.transAxes, fontsize=12, color='darkred')
    ax.grid(alpha=0.3)

# 调整布局和显示
plt.subplots_adjust(wspace=0.5)
plt.tight_layout()
plt.savefig('1_kcc_vis.svg')
plt.show()