Evaluation of lncRNAs as Potential Biomarkers for Diagnosis of Metastatic Triple-Negative Breast Cancer through Bioinformatics and Machine Learning

Abstract

Background: Triple-negative breast cancer (TNBC) is highly invasive and metastatic to the lymph nodes. Therefore, it is an urgent priority to distinguish novel biomarkers and molecular mechanisms of lymph node metastasis as the first step to the disease investigation. Long non-coding RNAs (lncRNAs) have widely been explored in cancer tumorigenesis, progression, and invasion.

Objectives: This study aimed to identify and evaluate lncRNAs in the signaling pathway of MMP11 gene in both metastatic and non-metastatic TNBC samples. The potential of lncRNAs in prognosis and diagnosis of the disease was also assessed using bioinformatics analysis, machine learning, and quantitative real-time PCR.

Materials and methods: Using machine learning algorithms, we analyzed the available BC data from the Cancer Genome Atlas Network (TCGA) and identified three potential lncRNAs, gastric adenocarcinoma-associated, positive CD44 regulator, long intergenic noncoding RNA (GAPLINC), TPT1-AS1, and EIF1B antisense RNA 1 (EIF1B-AS1) that could successfully distinguish between metastatic and non-metastatic TNBC.

Results: The results showed the upregulation of GAPLINC lncRNA in metastatic BC tissues, compared to non-metastatic (P<0.01) and normal samples, though TPT1-AS1 and EIF1B-AS1 were downregulated in metastatic TNBC samples (P<0.01).

Conclusion: Given the aberrant expression of candidate lncRNAs and the underlying mechanisms, the above-mentioned RNAs could act as novel diagnostic and prognostic biomarkers in metastatic BC.

Keywords: Biomarkers; Breast neoplasms; Long noncoding RNA; Neoplasm metastasis; Triple negative breast neoplasms.

Figure 1

The heatmap of lncRNAs genes. The genes (n = 117) were grouped into two main clusters using ggplot2 R package.

Figure 2

Cox regression analysis. Multivariate Cox regression model in patients’ survival rate considering two crucial factors, the pathological stage and age, indicating risk scores A) in high- and low-risk patients and B) of the survival plot.

Figure 3

TCGA RNA-seq dataset results. A) Results of TCGA RNA-seq and expression matrix to identify the number of genes related to breast cancer; B) Analysis of lncRNAs expression in cancer and normal samples, showing that 117 lncRNAs were significantly correlated with MMP11 expression level.

Figure 4

Deep learning analysis. A) Model accuracy and loss; B) confusion matrix using creat R package.

Figure 5

Decision tree constructed by R packege rpart. A) The genes that play the most important role in the model and in separation of metastatic from non-metastatic samples; B) The results of SVM showed that TPT1-AS1, LINC01099, GAPLINC, LINC01140 and EIF1B-AS1 had the most essential role in the model.

Figure 6

Comparison of expression level changes of candidate lncRNAs with the MMP11 gene. The expression level of A) MMP11, B) GAPLINC, C) TPT1-AS1, and D) EIF1B-AS1 in cancer compared to normal samples using GhraphPad (V 8).