The molecular profile of the tumor is crucial for classifying brain tumors and, consequently, for selecting the appropriate treatment strategy. Circulating microRNAs are considered promising biomarkers for tumors. MicroRNA NGS sequencing data can be effectively categorized using machine learning.

OBJECTIVE

To create a classifier model using machine learning algorithms for differential diagnosis of glial and benign brain tumors based on high-throughput sequencing data of a tumor-associated pattern of circulating microRNAs.

MATERIALS AND METHODS

MicroRNAs isolated from plasma of patients with glioblastoma (n=22), astrocytoma (n=15), oligodendroglioma (n=5), meningioma (n=13), and control subjects (n=4) were analyzed using high-throughput sequencing on a MiSeq Dx instrument (Illumina, Inc., USA). Differential expression analysis was performed using the DESeq2 software package. For each study group of patients, differentially expressed microRNAs were selected that met the following criteria: |Log2FC|>1 and p<0.05. The sequencing data was normalized and supplemented with synthetic data to eliminate class imbalance, then divided into training and test sets in an 80:20 ratio. The random forest method and the XGBoost algorithm were selected for the machine learning.

RESULTS

Fifty-one differentially expressed tumor-specific microRNAs were identified: 17 in the group of patients with glioblastoma, 10 in the group of patients with astrocytoma, 10 in the group of patients with oligodendroglioma, and 14 in the group of patients with meningioma. With the test dataset, the accuracy of the models trained using the random forest method and the XGBoost algorithm was 85.2% (p<0.001) and 81.5% (p<0.001), respectively. During high-throughput sequencing, bioinformatic analysis, and machine learning methods, 10 microRNAs (hsa-miR-192-5p, hsa-miR-194-5p, hsa-miR-128-3p, hsa-miR-30c-5p, hsa-miR-186-5p, hsa-miR-340-5p, hsa-miR-3168, hsa-miR-19b-3p, hsa-miR-144-5p, hsa-let-7c-5p) were selected for the differential diagnosis of glial tumors and meningiomas.

CONCLUSION

The selected tumor-specific microRNAs can serve as the basis for a minimally invasive diagnostic panel for brain tumors.