Researchers from China have developed a new machine learning-based diagnostic model for idiopathic pulmonary fibrosis (IPF), according to an article in Scientific Reports.

“We found that the [area under the curve (AUC)] efficiency of our model, called neura IPF, was excellent,” the researchers explained. “However, the lack of gene data for IPF in the [Gene Expression Omnibus] database meant that the genetic characteristics of IPF were not included in the construction of the diagnostic model, potentially compromising its diagnostic effectiveness for IPF.”

Through the analysis of gene expression datasets, the researchers identified 6 differentially expressed genes between IPF patients and healthy subjects: CDH3, DIO2, ADAMTS14, HS6ST2, IL13RA2, and IGFL2. The expression of these genes was not affected by patients’ age or sex. These genes were then used to construct the artificial neural network model for IPF diagnosis.

Explore more great content on Idiopathic Pulmonary Fibrosis (IPF) →
New Venture Aims to Improve Diagnosis, Treatment of Pulmonary Fibrosis
The Role of Cellular Senescence in Idiopathic Pulmonary Fibrosis
Understanding the Pathological Formation of Extracellular Matrix in IPF
Continue Reading

The model was validated in 3 independent gene expression datasets, showing a sensitivity of 90%, 100%, and 75%, and a specificity of 85%, 100%, and 90% in datasets 1, 2, and 3, respectively. AUCs were 0.856 (dataset 1), 1 (dataset 2), and 0.880 (dataset 3).

Read about IPF diagnosis

The 6 differentially expressed genes were positively correlated with patients’ lung function. Moreover, survival analysis highlighted CDH3, which encodes for cadherin-3, as a potential prognostic marker in IPF.

In addition, CDH3, ADAMTS14, and DIO2 were associated with overall survival in IPF, as patients with high expression of CDH3 and ADAMTS14 had a poor prognosis, and patients with high expression of DIO2 had a good prognosis.

These preliminary findings need further validation in future studies. Despite the limitations, neura IPF shows advantages over previously developed models, including its strong predictive performance and the use of fewer feature variables. Moreover, neura IPF was validated in a large-scale dataset.

Reference

Li Z, Wang S, Zhao H, et al. Artificial neural network identified the significant genes to distinguish idiopathic pulmonary fibrosis. Sci Rep. 2023;13(1):1225. doi:10.1038/s41598-023-28536-w