Multiple omics-based machine learning reveals specific macrophage sub-clusters in renal ischemia-reperfusion injury and constructs predictive models for transplant outcomes.

BACKGROUND

Ischemia-reperfusion injury (IRI) is closely associated with numerous severe postoperative complications, including acute rejection, delayed graft function (DGF) and graft failure. Macrophages are central to modulating the aseptic inflammatory response during the IRI process. The objective of this study is to conduct an analysis of the developmental and differentiation characteristics of macrophages in IRI, identify distinct molecules subtypes of IRI, and establish robust predictive strategies for DGF and graft survival.

METHOD

We analyzed scRNA-Seq data from GEO database to identify macrophage sub-clusters specific to renal IRI, and use the hdWGCNA algorithm to screen gene modules closely associated with this sub-cluster. Integrating these module genes with the results from bulk RNA-Seq differential analysis to obtain hub genes, and delineating the different IRI molecular subtypes through consensus clustering based on the expression profiles of hub genes. Innovatively, the gene expression matrix was transformed into a unique graphic pixel module and applied advanced computer vision processing algorithms to construct a DGF predictive model. Additionally, we also employed 111 combinations of 10 machine learning algorithms to develop a predictive signature for graft survival. Finally, we validated the expression of the key gene ANXA1 in a mouse IRI model using qRT-PCR, WB, and IHC.

RESULT

This study successfully identified a subset of macrophages closely associated with renal IRI, and cell communication and pseudo-time analysis implied that they may be instrumental in both the maintenance and exacerbation of the IRI process. Utilizing the expression patterns of hub genes, recipients can be clustered into two subtypes (CI and C2) with unique clinical and molecular features. We innovatively applied deep learning algorithms to construct a model for DGF prediction, which can effectively mitigate batch effects among IRI recipients. Compared to other existing models, our model demonstrated superior performance with AUC of 0.816 and 0.845 in the training and validation set. Furthermore, we also used the random survival forest algorithm to develop a high-precision predictive signature for graft failure. The mouse IRI model confirmed a marked upregulation of ANXA1 mRNA and protein expression in renal tissue following IRI.

CONCLUSION

This study successfully revealed the macrophage sub-cluster closely associated with renal IRI. Two distinct IRI subgroups with different characteristics were identified and robust strategies were constructed for predicting DGF and graft survival, which can offer potential therapeutic targets for the treatment of IRI and reference for early prevention of various postoperative complications.