Screening disease feature genes and analyzing correlations with immune cell infiltration in knee osteoarthritis chondrocytes based on multiple machine learning algorithms.

This study aimed to comprehensively analyze differentially expressed genes (DEGs) in chondrocytes from patients with knee osteoarthritis (OA) by integrating multiple machine learning algorithms and bioinformatics techniques, to unravel the underlying molecular mechanisms associated with OA chondrocytes, and to provide novel insights for the innovation of clinical therapeutic strategies. We downloaded the GSE117999, GSE114007, GSE169077, GSE246425, and GSE178557 datasets from the public Gene Expression Omnibus (GEO) database as the training set, while GSE57218 served as an independent validation set. To ensure data consistency and comparability, the training set was normalized, and the ComBat algorithm was applied to eliminate batch effects, yielding a merged gene expression dataset. Subsequent differential expression analysis was performed to identify genes with significant changes under disease conditions, followed by enrichment analysis. To more accurately identify genes closely linked to disease characteristics, we independently analyzed the merged dataset using three machine learning algorithms: Lasso regression, random forest, and support vector machine (SVM). The intersection of results from these three methods was used to construct a robust list of disease-related feature genes. These prominent feature genes were validated in the training set and further externally confirmed using the GSE57218 dataset. Additionally, the CIBERSORT algorithm was employed to quantify immune cell infiltration in the normalized gene expression data, selecting infiltration results with high reliability (P < 0.05). Focusing on the target genes, we clarified the strength and significance of their associations with immune cell infiltration levels, comprehensively revealing differences in immune cell infiltration profiles between groups and the potential associations with target genes. DDIT3 and PFKFB3 were significantly downregulated in OA patients. DDIT3 was specifically associated with lipid metabolism, apoptosis, and inflammatory genes (e.g., TNFRSF12A), whereas PFKFB3 was linked to phospholipid synthesis and cell cycle genes (e.g., CHKA). Both genes were associated with core OA-related pathways, including PI3K-Akt and AGE-RAGE. Immune infiltration analysis revealed that DDIT3 was positively correlated with pro-inflammatory mast cells and M1 macrophages, while PFKFB3 was negatively correlated with activated dendritic cells. Collectively, these two genes were associated with immune cell infiltration patterns. The competing endogenous RNA (ceRNA) network analysis indicated that DDIT3 was associated with axes such as LINC00689-miR-769-5p, and PFKFB3 was associated with complex networks like GAS6-AS1-miR-146a-5p. DDIT3 and PFKFB3 are key candidate genes associated with the pathological progression of OA. Their downregulation is correlated with inflammatory and metabolic disturbances in chondrocytes, supporting their potential use as diagnostic biomarkers and therapeutic targets for OA.