Inconsistencies in Correspondence between Drugs and Diseases

Oct 13, 2025 by wbrum.org 65 views

Introduction

In the field of drug repositioning, datasets play a crucial role in the development and evaluation of models. However, inconsistencies in these datasets can lead to inaccurate results and undermine the reliability of the models. In this article, we will investigate the potential issues with the datasets used in the paper "Regulation-aware Graph Learning for Drug Repositioning over Heterogeneous Biological Networks" and "SiSGC: A Drug Repositioning Prediction Model Based on Heterogeneous Simplifying Graph Convolution".

Inconsistencies in Drug-Protein Associations

One of the first issues that arose was the discrepancy in the number of drug-protein associations between the two papers. The number of associations in the paper "Regulation-aware Graph Learning for Drug Repositioning over Heterogeneous Biological Networks" was found to be the opposite of those in the paper "SiSGC: A Drug Repositioning Prediction Model Based on Heterogeneous Simplifying Graph Convolution". Furthermore, one of the authors is the same in both papers, which raises suspicions about the dataset versions used.

Inconsistencies in Drug-Disease Associations

In the replication of the paper, it was found that the drug-disease associations appear to be inconsistent. For example, the total number of drugs in the F-Dataset is 592, as recorded in DrDiNum.csv and DrPr.csv. However, according to the paper, it should be 593. This discrepancy raises concerns about the accuracy of the dataset.

Inconsistencies in Correspondence between Drugs and Diseases

Additionally, the correspondence between drugs and diseases may be incorrect. A comparison was made between DrDiNum.csv and the data from Gdataset, which originates from the paper titled "Drug repositioning with adaptive graph convolutional networks". The comparison code is as follows:

import pandas as pd
import numpy as np

def find_differences(ori_file, error_file):
    """比较两个CSV文件的差异，使用双指针方法"""
    # 读取文件
    df_ori = pd.read_csv(ori_file, header=None)
    df_error = pd.read_csv(error_file, header=None)

    # 获取两个文件的长度
    len_ori = len(df_ori)
    len_error = len(df_error)
    print(f"原始文件行数: {len_ori}")
    print(f"错误文件行数: {len_error}")

    # 初始化指针
    i = 0  # 原始文件指针
    j = 0  # 错误文件指针
    diff_start = None

    while i < len_ori and j < len_error:
        ori_val = df_ori.iloc[i, 1]
        error_val = df_error.iloc[j, 1]

        if ori_val != error_val:
            if diff_start is None:
                diff_start = (i, j)
                print(f"\n发现差异起始位置:")
                print(f"原始文件第 {i+1} 行: ({df_ori.iloc[i, 0]}, {ori_val})")
                print(f"错误文件第 {j+1} 行: ({df_error.iloc[j, 0]}, {error_val})")

            # 尝试在错误文件中向前查找匹配
            found_match = False
            look_ahead = 1
            while j + look_ahead < len_error and look_ahead < 5:  # 最多向前看5行
                if df_error.iloc[j + look_ahead, 1] == ori_val:
                    print(f"在错误文件向前第{look_ahead}行找到匹配")
                    j += look_ahead
                    found_match = True
                    break
                look_ahead += 1

            # 如果在错误文件中没找到，尝试在原始文件中向前查找
            if not found_match:
                look_ahead = 1
                while i + look_ahead < len_ori and look_ahead < 5:  # 最多向前看5行
                    if df_ori.iloc[i + look_ahead, 1] == error_val:
                        print(f"在原始文件向前第{look_ahead}行找到匹配")
                        i += look_ahead
                        found_match = True
                        break
                    look_ahead += 1

            # 如果都没找到，两个指针都向前移动一步
            if not found_match:
                i += 1
                j += 1
                continue

        # 找到匹配
        if diff_start is not None:
            print(f"差异结束位置:")
            print(f"原始文件第 {i+1} 行")
            print(f"错误文件第 {j+1} 行")
            print(f"这段差异包含: 原始文件 {i - diff_start[0]} 行, 错误文件 {j - diff_start[1]} 行")
            print(f"当前匹配行: ({df_ori.iloc[i, 0]}, {ori_val})")
            diff_start = None

        # 移动两个指针
        i += 1
        j += 1

    # 检查是否还有未处理的差异
    if diff_start is not None:
        print(f"\n最后一个差异段未结束:")
        print(f"从原始文件第 {diff_start[0]+1} 行到末尾")
        print(f"从错误文件第 {diff_start[1]+1} 行到末尾")

    # 检查是否有未处理的行
    if i < len_ori:
        print(f"\n原始文件还有 {len_ori - i} 行未处理")
        print("最后几行:")
        for k in range(i, min(i+5, len_ori)):
            print(f"({df_ori.iloc[k, 0]}, {df_ori.iloc[k, 1]})")

    if j < len_error:
        print(f"\n错误文件还有 {len_error - j} 行未处理")
        print("最后几行:")
        for k in range(j, min(j+5, len_error)):
            print(f"({df_error.iloc[k, 0]}, {df_error.iloc[k, 1]})")

def main():
    ori_file = "data/F-Dataset/DrDiNumOri.csv"
    error_file = "data/F-Dataset/DrDiNum.csv"
    find_differences(ori_file, error_file)

if __name__ == "__main__":
    main()

The comparison code is used to find the differences between the two files. The output of the code is as follows:

原始文件行数: 604
错误文件行数: 592
发现差异起始位置:
原始文件第 1 行: (1, 1)
错误文件第 1 行: (1, 0)
在错误文件向前第 1 行找到匹配
原始文件第 2 行: (2, 1)
错误文件第 2 行: (2, 0)
在错误文件向前第 1 行找到匹配
原始文件第 3 行: (3, 1)
错误文件第 3 行: (3, 0)
在错误文件向前第 1 行找到匹配
...

The output shows that there are 12 less lines of associations in DrDiNum.csv compared to DrDiNumOri.csv, but in addition there are 12 more lines of different associations.

Inconsistencies in Duplicate Rows

Additionally, DrPrNum.csv and DiPrNum.csv have a large number of duplicate rows. The removal of duplicate associations does not match the number of associations described in the paper.

Conclusion

In conclusion, the datasets used in the paper "Regulation-aware Graph Learning for Drug Repositioning over Heterogeneous Biological Networks" and "SiSGC: A Drug Repositioning Prediction Model Based on Heterogeneous Simplifying Graph Convolution" appear to have inconsistencies. The discrepancies in the number of drug-protein associations, drug-disease associations, and correspondence between drugs and diseases raise concerns about the accuracy of the dataset. Furthermore, the presence of duplicate rows in DrPrNum.csv and DiPrNum.csv also raises concerns about the reliability of the dataset. Therefore, it is essential to investigate and correct these inconsistencies to ensure the accuracy and reliability of the dataset.

Future Work

Future work should focus on investigating the causes of these inconsistencies and correcting them. This may involve re-collecting data, re-processing the data, or re-annotating the data. Additionally, it is essential to develop a robust and reliable method for detecting and correcting inconsistencies in the dataset.

References

[1] Regulation-aware Graph Learning for Drug Repositioning over Heterogeneous Biological Networks. Paper
[2] SiSGC: A Drug Repositioning Prediction Model Based on Heterogeneous Simplifying Graph Convolution. Paper
[3] Drug repositioning with adaptive graph convolutional networks. Paper

Introduction

In our previous article, we investigated the potential issues with the datasets used in the paper "Regulation-aware Graph Learning for Drug Repositioning over Heterogeneous Biological Networks" and "SiSGC: A Drug Repositioning Prediction Model Based on Heterogeneous Simplifying Graph Convolution". In this article, we will answer some of the frequently asked questions (FAQs) related to the inconsistencies in the datasets.

Q: What are the inconsistencies in the datasets?

A: The inconsistencies in the datasets include discrepancies in the number of drug-protein associations, drug-disease associations, and correspondence between drugs and diseases. Additionally, there are duplicate rows in DrPrNum.csv and DiPrNum.csv.

Q: Why are the inconsistencies in the datasets a concern?

A: The inconsistencies in the datasets raise concerns about the accuracy and reliability of the dataset. If the dataset is not accurate, it can lead to inaccurate results and undermine the reliability of the models.

Q: How can the inconsistencies in the datasets be corrected?

A: The inconsistencies in the datasets can be corrected by re-collecting data, re-processing the data, or re-annotating the data. Additionally, a robust and reliable method for detecting and correcting inconsistencies in the dataset should be developed.

Q: What are the potential causes of the inconsistencies in the datasets?

A: The potential causes of the inconsistencies in the datasets include errors in data collection, processing, or annotation. Additionally, inconsistencies in the data sources or methods used to collect the data may also contribute to the inconsistencies.

Q: How can the accuracy and reliability of the dataset be ensured?

A: The accuracy and reliability of the dataset can be ensured by developing a robust and reliable method for detecting and correcting inconsistencies in the dataset. Additionally, regular audits and quality control checks should be performed to ensure the accuracy and reliability of the dataset.

Q: What are the implications of the inconsistencies in the datasets?

A: The implications of the inconsistencies in the datasets include inaccurate results, undermined reliability of the models, and potential harm to patients. Therefore, it is essential to investigate and correct the inconsistencies in the datasets.

Q: How can researchers and practitioners ensure the accuracy and reliability of the datasets used in their research?

A: Researchers and practitioners can ensure the accuracy and reliability of the datasets used in their research by:

Verifying the accuracy and reliability of the dataset before using it
Developing a robust and reliable method for detecting and correcting inconsistencies in the dataset
Regularly auditing and quality controlling the dataset
Using multiple datasets and methods to validate the results
Reporting any inconsistencies or errors in the dataset

Q: What are the future directions for research on the inconsistencies in the datasets?

A: The future directions for research on the inconsistencies in the datasets include:

Investigating the causes of the inconsistencies in the datasets
Developing a robust and reliable method for detecting and correcting inconsistencies in the dataset
Ensuring the accuracy and reliability of the dataset
Validating the results using multiple datasets and methods
Reporting any inconsistencies or errors in the dataset

Conclusion

In conclusion, the inconsistencies in the datasets used in the paper "Regulation-aware Graph Learning for Drug Repositioning over Heterogeneous Biological Networks" and "SiSGC: A Drug Repositioning Prediction Model Based on Heterogeneous Simplifying Graph Convolution" raise concerns about the accuracy and reliability of the dataset. Therefore, it is essential to investigate and correct the inconsistencies in the datasets to ensure the accuracy and reliability of the dataset.

References

[1] Regulation-aware Graph Learning for Drug Repositioning over Heterogeneous Biological Networks. Paper
[2] SiSGC: A Drug Repositioning Prediction Model Based on Heterogeneous Simplifying Graph Convolution. Paper
[3] Drug repositioning with adaptive graph convolutional networks. Paper