Banca de QUALIFICAÇÃO: DÉBORA VIRGÍNIA DA COSTA E LIMA

Uma banca de QUALIFICAÇÃO de MESTRADO foi cadastrada pelo programa.
STUDENT : DÉBORA VIRGÍNIA DA COSTA E LIMA
DATE: 17/12/2021
TIME: 14:00
LOCAL: meet.google.com/efj-rgdo-wyh
TITLE:

The Use of Artificial Neural Networks in Lung Cancer Data Analysis

KEY WORDS:

Machine learning, Artificial intelligence, Oncological data.

PAGES: 52
BIG AREA: Ciências Biológicas
AREA: Biologia Geral
SUMMARY:

Lung cancer is a disease of great importance, representing the main cause of death from cancer, and having, in addition to high levels of incidence, high mortality. As with other types of cancer, it can occur for different causes, from genetic to environmental, so studies carried out from different types of data may be relevant for the control of this neoplasm, especially when considering factors that have an impact on patient survival. Clinical and molecular data obtained from patients result in a large volume of data, so performing pre-processing steps and bioinformatics analysis brings benefits in information discovery and data selection. The steps of this study are: obtaining clinical and molecular data from TCGA (The Cancer Genome Atlas) databases referring to the LUSC (Squamous Cell Carcinoma of the Lung) and LUAD (Lung Adenocarcinoma) cohorts, followed by bioinformatics analysis, organizing and pre-processing the data, and developing neural networks using as input the frequently mutated genes for each cohort. The cohorts showed differences in survival among themselves when submitted to the Kaplan-Meier method and the Log-Rank test, in addition, the statistically relevant clinical variables in the Cox proportional hazards model were race and the presence of metastases. In the genomic analysis, all genes with a mutation frequency greater than 15% were selected, with 35 genes for LUAD and 32 for LUSC being found. The use of these genes as input into the constructed networks enabled the generation of networks with 100% accuracy, through cross-validation, identifying, according to mutations, whether the patient was alive or dead. The developed models consisted of deep MLP (multi-layer perceptron) networks using a dropout technique and Adam's algorithm as a training optimizer. The method of choosing genes with frequent mutations associated with the use of deep learning allows predicting the survival of patients with lung cancer.

BANKING MEMBERS:
Presidente - 347628 - ADRIAO DUARTE DORIA NETO
Externa ao Programa - 1365498 - BEATRIZ STRANSKY FERREIRA
Interno - 2170415 - JORGE ESTEFANO SANTANA DE SOUZA
Interno - 3063244 - TETSU SAKAMOTO