Machine learning based biomarker discovery for chronic kidney disease–mineral and bone disorder (CKD-MBD)

Yuting Li,Yukuan Lou,Man Liu,Siyi Chen,Peng Tan,Xiang Li,Huaixin Sun,Weixin Kong,Suhua Zhang,Xiang Shao

DOI: https://doi.org/10.1186/s12911-024-02421-6

IF: 3.298

2024-02-07

BMC Medical Informatics and Decision Making

Abstract:Chronic kidney disease-mineral and bone disorder (CKD-MBD) is characterized by bone abnormalities, vascular calcification, and some other complications. Although there are diagnostic criteria for CKD-MBD, in situations when conducting target feature examining are unavailable, there is a need to investigate and discover alternative biochemical criteria that are easy to obtain. Moreover, studying the correlations between the newly discovered biomarkers and the existing ones may provide insights into the underlying molecular mechanisms of CKD-MBD.

medical informatics

What problem does this paper attempt to address?

This paper attempts to address the challenges in the early detection of chronic kidney disease - mineral and bone metabolism disorder (CKD - MBD) and in the research on its molecular mechanisms. Specifically, the authors aim to:

1. **Seek alternative biomarkers**: In cases where target - feature examinations are not possible, explore easily accessible alternative biochemical criteria as early - detection indicators for CKD - MBD.
2. **Improve diagnostic accuracy**: Through machine - learning methods, identify the optimal combination of biomarkers that can effectively distinguish different types of CKD - MBD (abnormal calcium, abnormal phosphorus, and abnormal PTH).
3. **Study molecular mechanisms**: By analyzing the correlations between newly discovered biomarkers and existing ones, provide clues for in - depth research on the potential molecular mechanisms of CKD - MBD.

### Main problem description

Chronic kidney disease - mineral and bone metabolism disorder (CKD - MBD) is a complex systemic disease, manifested as abnormalities in mineral and bone metabolism, such as abnormal metabolism of calcium, phosphorus, parathyroid hormone (PTH), or vitamin D, as well as abnormalities in bone turnover, mineralization, volume, linear growth, or strength, and may also be accompanied by calcification of blood vessels and other soft tissues. Although there are already diagnostic criteria, in some cases, measuring blood calcium, phosphorus, and PTH levels may be impractical, infrequent, or costly. In addition, existing diagnostic methods may have false - positive and false - negative problems, so new biochemical criteria need to be explored to assist in diagnosis.

### Solution

To address the above problems, the authors adopted the following methods:

- **Data collection**: Collected 65 blood biochemical indicators from 116 patients and divided the patients into four categories: abnormal calcium, abnormal phosphorus, abnormal PTH, and healthy control group.
- **Machine - learning methods**: Used six machine - learning algorithms (KNN, logistic regression, linear regression, support vector regression, deep neural network, and random forest) for classification and feature selection.
- **Feature selection**: Employed the sequential forward feature selection algorithm to gradually select the most discriminative features and construct the optimal biomarker panel.
- **Validation**: Used an independent cohort of 114 patients to verify the predictive ability of the selected features.

Through these methods, the authors successfully identified a set of highly discriminative biomarkers that can effectively classify different types of CKD - MBD and provided a basis for further research on their molecular mechanisms.

### Conclusion

This study is the first to use machine - learning methods to explore biochemical criteria related to CKD - MBD, discovered alternative biomarkers that can be used as early - detection indicators, and provided new candidate markers for in - depth research on the molecular mechanisms of CKD - MBD. This not only helps to improve the diagnostic accuracy of CKD - MBD but also provides an important reference for future clinical applications and basic research.