Dados do Resumo

Título

Evaluation by Machine Learning, by the Randon Forest algorithm of DNA methylation data as a Predictor of High and Low Grade Lesions in Cervical Cancer

Introdução

Cervical cancer in the process of carcinogenesis has an intrinsic interaction with persistent Human Papilloma Virus (HPV) infection, mainly HPV 16 and HPV 18. This is one of the most common types of cancer in women of various ages, and can progress to Cervical Intraepithelial Neoplasias (CINs). Epigenetic processes participate in the emergence of CINs. Studies have elucidated that several tumor suppressor genes are suppressed in their expression by the process of DNA methylation, in women with HPV 16 and HPV 18 in high-grade lesions. On the other hand, Machine Learning techniques are currently used in molecular data to choose the best profiles of a biological state. AIMS: Evaluate by Machine Learning, using DNA methylation data from a panel of tumor suppressor genes, the possibility of predicting women with high and low grade lesions.

Objetivo

Evaluate by Machine Learning, using DNA methylation data from a panel of tumor suppressor genes, the possibility of predicting women with high and low grade lesions.

Métodos

Cross-sectional observational study, 407 cervical cytology samples from women at the Hospital de Amor de Barrretos, between 2014 and 2015, were included, and the DNA methylation profile of 12 tumor suppressor genes was previously analyzed.
Patients characcteristic, cervical cytology samples, pathological analysis, cervical biopsy and DNA methylation, obtaining 53 attributes for analysis with the Randon Forest algorithm.

Resultados

Computational simulation analyzes with the supervised Randon Forest algorithm revealed the relationship between HPV infection status and 8 differentially methylated genes. It was found that the hsamiR 1242, JAM3 and EPB41L3 genes presented a greater probability of separation between CIN 1 and CIN 2+.

Conclusões

It was possible to verify that the methylation levels of the EPB41L3, hsamiR 1242 and JAM3 genes were more relevant in separating high and low grade lesions according to analyzes using the Randon Forest algorithm.

Palavras Chave

Cervical cancer; DNA methylation and Machine Learning