Exploring gene expression heterogeneity in burn wounds through machine learning models: in silico...

This study used machine learning to assess gene expression heterogeneity among burn wound patients.

Gene expression data were downloaded from the Gene Expression Omnibus database. The raw data were normalized and then reduced using principal component analysis. Then, the machine learning algorithms such as K-Means Clustering, Agglomerative Clustering, Spectral Clustering, and Gaussian Mixture Models were used to perform the clustering, and the performance of each model was compared by way of the Silhouette Score. Linear discriminant analysis (LDA) was run on the resulting clusters using gene expression features with parameters successfully tuned via cross-validation.

The clustering algorithms identified three distinct gene expression clusters in burn patients. Differential gene expression analysis revealed significant variations between clusters, with 303 genes differentially expressed between Clusters 0 and 1, 12 between Clusters 0 and 2, and 429 between Clusters 1 and 2. Age differences were also significant, with Clusters 0 and 1 representing older individuals compared to Cluster 2. No significant differences were found in the time since injury between clusters. The LDA model, a cornerstone of this study, demonstrated impressive accuracy in classifying gene expression data, with a test accuracy of 93% and robust performance metrics across clusters.

This study not only illuminates the gene expression heterogeneity in burn wounds but also suggests the potential influence of age on this variation. By identifying molecular subtypes of burn injury based on gene expression patterns, these findings may help in developing early diagnostic tools and personalized therapeutic approaches.