A predictive model of aging-related secretion phenotype for osteoarthritis constructed using integrated bioinformatics and machine learning / 中国医科大学学报

ABSTRACT

Objective To explore the predictive markers of senescence-associated secretory phenotype(SASP)in osteoarthritis(OA).Methods OA datasets were screened by the Gene Expression Omnibus(GEO)database,while SASP-related genes were collected by PubMed.Three machine learning algorithms,including least absolute shrinkage and selection operator(LASSO),support vector machines recursive feature elimination(SVM-RFE),and random forest(RF),were used to screen the candidate predictive markers of SASP genes in OA,and the OA prediction model was constructed using the overlapping genes identified by the machine learning algo-rithms.CIBERSORT was used to explore the degree of peripheral blood immune cell infiltration in OA versus normal samples.The miRNA-transcription factor-mRNA regulatory network of the model genes was predicted using Cytoscape.The most valuable genes of the predic-tion model were experimentally verified by real-time quantitative polymerase chain reaction(RT-qPCR)in OA rats and normal control rats(n= 6 per group).Results One OA dataset was screened by the GEO database,and 125 OA-related SASP genes were isolated.A total of seven intersection genes were obtained by the three machine learning algorithms.The area under the curve of the prediction model was 0.891.The CIBERSORT immune infiltration results showed a significant difference in plasma cell infiltration level between OA and normal samples(P= 0.001 3).The RT-qPCR results showed that the expression level of TNFRSF1Awas significantly higher in the OA versus normal group(P<0.0001).Conclusion TNFRSF1Ais highly expressed in OA and may be a potential predictive marker for it.