A0908
Title: The correlation-based weighted adaptive LASSO estimator for identifying brain tumor molecular subtypes
Authors: Mina Norouzirad - Center for Mathematics and Applications (NOVAMath) (Portugal) [presenting]
Tomas Bandeira - NOVA University Lisbon (Portugal)
Marta Belchior Lopes - Universidade NOVA de Lisboa (Portugal)
Abstract: Gliomas are brain tumors marked by substantial inter- and intratumoral heterogeneity and are generally associated with poor prognosis. Accurate diagnosis and subtype classification are crucial for improving clinical outcomes. The focus is on classifying glioma molecular subtypes, including astrocytoma, glioblastoma, and oligodendroglioma, using RNA-sequencing data from 619 samples and 20,504 genes sourced from The Cancer Genome Atlas (TCGA). A key challenge is identifying novel biomarkers and understanding their role in tumor progression. The aim is to propose a two-phase data-driven approach: (i) estimate feature relevance using a correlation-based estimator in multinomial logistic regression, and (ii) use the inverse absolute values of these estimates as adaptive weights in an adaptive LASSO framework. This adaptive LASSO model is then benchmarked against traditional regularization techniques, including LASSO, ridge regression, and elastic net. The proposed methodology aims to enhance predictive accuracy and support the discovery of biologically meaningful biomarkers for glioma subtype classification.
