Development and validation of a novel gene signatures based on a random forest algorithm and artificial neural network for predictive diagnosis of cervical squamous cell carcinoma

Background: Effective treatment of cervical carcinoma can be challenging due to the lack of specific symptoms in the initial phase, as well as patients often only seeking medical attention in the middle and late stages of the disease when symptoms become more apparent. This study aims to address these limitations by developing and validating a gene signature for predicting cervical squamous cell carcinoma (CESC) using both the random forest algorithm and artificial neural network. Methods: Potential predictive genes for CESC were identified by analyzing three matrix datasets containing tissues from individuals with normal cervical epithelium and patients with CESC. Then, the random forest algorithm and artificial neural network were used to construct predictive models for CESC diagnosis, which were validated using both an independent validation dataset and in vitro experiments. To confirm the validity of the identified genes, protein and mRNA expression of eight disease signature genes were detected in the two groups using Western blotting and real-time quantitative polymerase chain reaction. Additionally, immunoinfiltration analysis was performed. Results: A total of 241 differentially expressed genes (DEGs) were identified, based on which eight genes with the highest predictive ability were selected and used to construct a molecular prognostic scoring system, which demonstrated exceptional predictive accuracy (Area Under Curve (AUC) = 0.995). Validation using an independent dataset confirmed the model’s remarkable predictive ability (AUC = 1.000). In vitro experiments demonstrated significant differences in the expression of the eight disease signature genes between the two groups. Immunoinfiltration analysis also revealed significant differences in immune cell infiltration, with squamous cell carcinoma of the cervix showing a higher degree of macrophage infiltration than normal cervical epithelium. Conclusions: Random forest algorithm and artificial neural network were used to obtain new gene signatures, based on which a molecular prognostic scoring system was developed to predict CESC and aid clinical decision-making.