Integrative transcriptomic analysis and explainable machine learning reveal the potential mechanisms of acupuncture in Crohn's disease.

BACKGROUND

Acupuncture has shown clinical benefits in Crohn's disease (CD), but the molecular links between acupuncture-responsive endogenous mediators and CD-related targets remain unclear.

METHODS

We integrated literature-derived acupuncture-responsive endogenous mediators with transcriptomic data from the Gene Expression Omnibus (GEO). Differential expression analysis and weighted gene co-expression network analysis (WGCNA) were used to identify CD-associated signature genes, which were intersected with predicted mediator targets. Eight machine-learning algorithms with repeated nested cross-validation and independent external validation were used to prioritize candidate targets, followed by SHapley Additive exPlanations (SHAP) interpretation. Immune-cell composition was estimated using Cell-type Identification by Estimating Relative Subsets of RNA Transcripts (CIBERSORT), and molecular docking was performed between the prioritized target and acupuncture-responsive endogenous mediators.

RESULTS

Ten acupuncture-responsive endogenous mediators and six shared candidate targets were identified. Random forest (RF) achieved the highest mean area under the receiver operating characteristic curve (AUC) in repeated nested cross-validation (0.905 ± 0.008), whereas gradient boosting machine (GBM) showed the highest AUC in the independent validation cohort, with an AUC of 0.789 and a 95% confidence interval (CI) of 0.652-0.925. FER ranked among the top three features in all eight models and showed the highest mean absolute SHAP value (0.267). FER expression was positively correlated with neutrophil abundance (ρ=0.268, P < 0.001) and negatively correlated with M2 macrophage abundance (ρ=-0.265, P < 0.001). Among the ten endogenous mediators examined, corticosterone showed the most favorable predicted docking score with the FER Src homology 2 (SH2) domain (-6.53 kcal/mol).

CONCLUSION

FER emerged as the prioritized target and was associated with neutrophil and M2 macrophage abundance. The predicted structural compatibility between corticosterone and the FER SH2 domain provides a testable hypothesis for future studies of acupuncture-responsive endogenous mediators in CD.