Machine–Learning-assisted Exploratory Prioritization of Transcriptomic Candidate Genes in Public Mesenchymal Stem Cell-derived Extracellular Vesicle Response Datasets

Talshyn Amirkhanqyzy Nurulla, Anar Balkashevna Tulyayeva

2026

Full text Open access: Hybrid Metadata (CSL-JSON)
DOI 10.4103/wkrj.wkrj_36_26
Data confidence 70 medium
How this score is calculated

It is built from checkable properties of the record itself: whether the article has a DOI, how many independent sources confirmed its data, whether the required fields are filled in, and how recent the latest confirmation is. Disagreement between sources and a missing identifier lower it. A value of 80 or above means the record was published without further checking; 60 to 79 means it is published at reduced confidence, with a manual check advisable. A record verified by an editor by hand is never pushed below 90 by a recalculation. This scores our data about the article, not the article itself.

Abstract

A BSTRACT Background: Mesenchymal stem cell-derived extracellular vesicles (MSC-EVs) have been associated with regenerative and immunomodulatory effects; however, transcriptomic responses to MSC-EV exposure remain incompletely characterized. Public datasets provide opportunities for exploratory reanalysis, but conclusions are limited when datasets differ in transcriptomic layer, recipient cell type, disease context, and EV source. Objective: The objective is to perform a hypothesis-generating computational reanalysis of public MSC-EV-related transcriptomic datasets and prioritize exploratory candidate genes and micro ribonucleic acids (miRNAs) associated with EV exposure. Materials and Methods: Two public GEO datasets were analyzed independently: GSE255429, a messenger RNA (mRNA) dataset of corneal epithelial stem cells exposed to adipose-derived MSC-EVs, and GSE237992, a miRNA dataset of rheumatoid arthritis synovial fibroblasts exposed to human umbilical cord MSC-derived exosomes. Differential expression was assessed using fold-change analysis, Welch’s t -test, and Benjamini–Hochberg false discovery rate (FDR) correction. Because no features survived FDR correction, downstream analyses were restricted to exploratory candidates selected using raw P value and effect-size thresholds. In the mRNA dataset, LASSO and random forest were used for feature prioritization. Classifier performance was assessed using leave-one-out cross-validation, but interpreted only as an instability check because of the very small sample size. Results: In the mRNA dataset, 955 genes met exploratory criteria based on raw P < 0.05 and |log2FC| ≥1, but none remained significant after FDR correction. In the miRNA dataset, 9 miRNAs met exploratory criteria, also without FDR-significant findings. LASSO and random forest produced nonoverlapping top-ranked gene lists, indicating substantial feature-ranking instability. A transparent exploratory prioritization process identified a 10-gene candidate shortlist: FBXO43, LRRN4, LCK, LINC00261, VWF, SPP1, FERMT2, TEX22, BLNK, and G0S2. Functional enrichment suggested nominal associations with extracellular matrix organization, cell adhesion, and tissue-remodeling processes, but pathway-level findings were interpreted cautiously because corrected significance was limited. Conclusions: This study provides an exploratory reanalysis of two biologically heterogeneous MSC-EV-related public datasets. The findings should not be interpreted as validated biomarkers or clinically actionable targets. Instead, the candidate genes and miRNAs represent hypothesis-generating signals requiring confirmation in larger, independent, and biologically homogeneous MSC-EV datasets, followed by experimental validation.