Predicting Survival in Glioblastoma Using Gene Expression Databases: A Neural Network Analysis.

Area of Science:

• Oncology
• Bioinformatics
• Computational Biology

Background:

• Glioblastoma (GBM) is the most aggressive and lethal primary brain tumor.
• Artificial neural networks (ANNs) show promise for enhancing diagnostic and prognostic accuracy in complex diseases.
• Predicting patient survival is crucial for guiding treatment strategies in neuro-oncology.

Purpose of the Study:

• To develop and evaluate an artificial neural network (ANN) model for predicting 15-month survival in glioblastoma (GBM) patients.
• To compare the predictive performance of the ANN model against traditional logistic regression (LR) using genomic and clinical data.
• To identify key molecular and clinical factors influencing GBM patient survival.

Main Methods:

• Genomic data from 551 GBM patients were obtained from multiple databases (CGGA, TCGA, MYO, CPTAC).
• An ANN model and a logistic regression (LR) model were constructed using age, gender, IDH status, and 31 key genes as input features.
• Model performance was assessed using Area Under the ROC Curve (AUC), accuracy, and Hosmer-Lemeshow (H-L) statistic, with data split into training, testing, and validation sets.

Main Results:

• The ANN model achieved a higher predictive accuracy (83.3%) and AUC (0.81) compared to the LR model (AUC 0.71).
• Key genes identified by the ANN model included FN1, ICAM1, MYD88, IL10, and CCL2.
• The ANN model demonstrated superior performance in predicting 15-month survival for GBM patients.

Conclusions:

• Artificial neural networks provide a robust and accurate method for predicting 15-month survival in glioblastoma patients.
• The developed ANN model can aid in personalized treatment planning and improve clinical decision-making for GBM.
• Gene expression patterns hold significant potential for prognostic stratification in aggressive brain tumors.