The team began with microarray data from two public peripheral-blood cohorts. The larger training dataset, GSE66360, contained 49 people with acute myocardial infarction and 50 healthy controls, while the independent validation dataset, GSE61144, contained seven patients and ten controls. Comparing patients with controls in the training cohort identified 477 differentially expressed genes, including 351 with increased expression and 126 with decreased expression. Intersecting this list with a previously published set of 822 genes encoding regulators of 20 protein post-translational modifications produced 17 candidate genes. Functional enrichment connected the candidates with processes including polyubiquitination, ubiquitin-ligase complexes, secretory-granule lumens and viral-response pathways, indicating that much of the emerging signal involved protein turnover and immune regulation.

To narrow the 17 candidates, the researchers applied three complementary machine-learning feature-selection methods: support vector machine recursive feature elimination, extreme gradient boosting and random forest. Their overlapping selections included GABARAPL1, PELI1, TNFAIP3 and NLRP3. TNFAIP3 did not show consistently significant differential expression between the two public datasets, so it was removed. The remaining three genes were consistently increased in acute myocardial infarction samples in both cohorts, and each had a receiver operating characteristic area under the curve above 0.80 in the training and validation sets. These values describe discrimination between patients and healthy controls in the analyzed data; they do not establish performance for diagnosing acute myocardial infarction in a general clinical population.

A deep neural network built around the three biomarkers achieved an area under the curve of 0.963 in the full training dataset. To reduce the risk of overestimating performance, the authors also used fivefold stratified cross-validation, generating an out-of-fold area under the curve of 0.894, with a 95% confidence interval of 0.832 to 0.957. A logistic-regression model using the same biomarkers reached 0.826, with a 95% confidence interval of 0.736 to 0.916. The difference between the two models was not statistically significant, with P equal to 0.1269. The comparison therefore shows that the neural network produced a higher point estimate under cross-validation, not that it was demonstrably superior to logistic regression.

Gene-set enrichment linked GABARAPL1 to 88 pathways, PELI1 to 84 and NLRP3 to 76, with shared signals involving messenger RNA splicing, Leishmania infection and chemokine signaling. The Leishmania result is a pathway-level annotation from human gene sets rather than evidence of parasitic infection. At the cellular level, the analysis estimated the abundance of 28 immune populations and found significant differences between patients and controls for 23 subsets, including activated B cells, activated CD4-positive T cells and activated dendritic cells. Macrophages and neutrophils were strongly correlated, with a Spearman coefficient of 0.87 and P below 0.001. GABARAPL1, PELI1 and NLRP3 were each most strongly associated with plasmacytoid dendritic cells, with coefficients of 0.66, 0.69 and 0.65, respectively.

The biological interest of the three markers lies in the different processes with which they are already associated. GABARAPL1 encodes a protein linked to autophagosome formation and maturation, a pathway involved in cellular stress responses and clearance of damaged components. PELI1 encodes an E3 ubiquitin ligase that participates in nuclear factor-kappa-B and JAK/STAT signaling and is connected with inflammatory responses and macrophage polarization. NLRP3 encodes a core component of the inflammasome, whose activation is associated with caspase-1 signaling and maturation of the inflammatory cytokines interleukin-1 beta and interleukin-18. Prior work discussed by the authors also links ubiquitination, deacetylation and phosphorylation of NLRP3 to myocardial injury. The present study measured messenger RNA, however, rather than the modifications of NLRP3 or other target proteins.

Database-based network reconstruction suggested possible layers of regulation around the biomarkers. The researchers predicted 15 transcription factors for GABARAPL1, seven for PELI1 and 14 for NLRP3. Predicted microRNAs numbered 10 for GABARAPL1, six for PELI1 and one for NLRP3, and these links were used to assemble an interaction network containing 1,393 microRNA–long non-coding RNA pairs. Such networks are computational hypotheses assembled from known or predicted binding relationships; they do not demonstrate that any factor changes the expression or activity of a biomarker in acute myocardial infarction. The analysis also linked the three genes to 80, 29 and 42 candidate compounds, respectively, before selecting representative pairs for molecular docking.

Docking simulations gave 1,4-chrysenequinone a score of −7 kilocalories per mole with GABARAPL1 and −8.9 kilocalories per mole with PELI1. Calcitriol produced a score of −10.1 kilocalories per mole with NLRP3. More negative scores were interpreted by the authors as stronger predicted binding affinity. These calculations do not establish that the compounds can safely or effectively treat acute myocardial infarction, nor do they capture metabolism, concentration, tissue delivery or effects on the intact cardiovascular system. The authors appropriately identified cellular and animal experiments as necessary steps for testing whether the proposed compounds alter biomarker function or cardiac outcomes.

Clinical expression validation used reverse-transcription quantitative PCR on blood from five patients with acute myocardial infarction and five healthy controls recruited at the Second Hospital of Hebei Medical University. All three transcripts were significantly higher in the patient group, with P below 0.001, consistent with the direction of the transcriptomic findings. The result is encouraging but limited by the very small sample, the selection of healthy rather than clinically suspected non-AMI controls and the absence of direct protein measurements. The study also used public cohorts with modest sample sizes, which may not represent the full variation in ethnicity, geography, comorbidity or clinical presentation. Proteomic and post-translational-modification assays, larger independent cohorts and appropriately designed diagnostic studies will therefore be needed before the biomarkers can be considered clinically validated or used as therapeutic targets.

Subject of Research: Transcriptomic screening of protein post-translational-modification regulators in acute myocardial infarction

Article Title: Exploring biomarkers and regulatory mechanisms associated with 20 protein post-translational modifications in acute myocardial infarction based on transcriptomics

Article References: Zhang, J., Li, B., Yang, M., Lu, J., Jiang, H., & Chen, F. (2026). Exploring biomarkers and regulatory mechanisms associated with 20 protein post-translational modifications in acute myocardial infarction based on transcriptomics. Frontiers in Cardiovascular Medicine, 13, 1848328. https://doi.org/10.3389/fcvm.2026.1848328

Image Credits: AI Generated

DOI: 10.3389/fcvm.2026.1848328

Keywords: acute myocardial infarction, blood biomarkers, transcriptomics, post-translational modification, GABARAPL1, PELI1, NLRP3, immune infiltration, molecular docking, machine learning, Exploring, biomarkers