Using Machine Learning Methods in Identifying Genes Associated with COVID-19 in Cardiomyocytes and Cardiac Vascular Endothelial Cells

Figures

• 

  Figure 1. Flow chart of the entire analysis process. Gene profile data for 104,182 cardiomyocytes and 22,438 vascular endothelial cells were analyzed. Each cell contained 29,071 gene expression levels. Genes were ordered according to the degree of correlation with COVID-19 using four feature selection methods. The obtained four ordered gene lists were fed into the incremental feature selection (IFS) method, where decision tree (DT) was employed as the classification algorithm. Finally, the best feature subsets and classification rules were extracted according to the IFS results using DT.

• 

  Figure 2. IFS curves for evaluating the performance of the decision tree under different numbers of top features in four lists. Four feature lists were generated by four feature selection methods, corresponding to four IFS curves (cardiomyocytes). (A) IFS curves based on the feature list yielded by LASSO. (B) IFS curves based on the feature list yielded by LightGBM. (C) IFS curves based on the feature list yielded by MCFS. (D) IFS curves based on the feature list yielded by RF.

• 

  Figure 3. IFS curves for evaluating the performance of the decision tree under different numbers of top features in four lists. Four feature lists were generated by four feature selection methods, corresponding to four IFS curves (vascular endothelial cells). (A) IFS curves based on the feature list yielded by LASSO. (B) IFS curves based on the feature list yielded by LightGBM. (C) IFS curves based on the feature list yielded by MCFS. (D) IFS curves based on the feature list yielded by RF.

• 

  Figure 4. Venn diagram of four subsets of important genes identified by LASSO, LightGBM, MCFS, and RF (cardiomyocytes). Overlapping circles indicate the number of genes identified by multiple feature selection methods.

• 

  Figure 5. Venn diagram of four subsets of important genes identified by LASSO, LightGBM, MCFS, and RF (vascular endothelial cells). Overlapping circles indicate the number of genes identified by multiple feature selection methods.