Bioinformatics identification of characteristic genes of cervical cancer via an artificial neural network.

Background: Artificial neural networks (ANNs) have been extensively used in the field of medicine. The present hypothesis-free study sought to use an ANN to identify the characteristic genes of cervical cancer (CC).

Methods: RNA sequencing profiles were obtained from the GSE7410, GSE9750, GSE63514, and GSE52903 datasets. The differentially expressed genes (DEGs) were identified and compared between the normal and CC tissues. An ANN analysis was conducted to obtain the random-forest tree and to examine differences in gene filtering. A neural network model was established using the characteristic genes of CC, while the verification accuracy of the model was examined by Cox regression. The differences in the immune infiltrating cells between the normal cervical and CC tissues were compared by CIBERSORT (an analytical tool can provide an estimation of the abundances of member cell types in a mixed cell population).

Results: Nine genes' characteristics for CC were identified: cyclin-dependent kinase inhibitor 2A (CDKN2A), chromosome 1 open reading frame 112 (C1orf112), helicase, lymphoid-specific (HELLS), mini-chromosome maintenance protein 5 (MCM5), mini-chromosome maintenance protein 2 (MCM2), kinetochore associated 1 (KNTC1), cysteine-rich secretory protein 3 (CRISP3), phytanoyl-CoA 2-hydroxylase interacting protein (PHYHIP), and cornulin (CRNN).

Conclusions: ANN is a robust neural network model that can be used to potentially predict CC based on the gene score. It can provide novel insights into the pathogenesis and molecular mechanisms of CC.