Machine learning-based investigation of regulated cell death for predicting prognosis and immunotherapy response in glioma patients

Abstract

Glioblastoma is a highly aggressive and malignant type of brain cancer that originates from glial cells in the brain, with a median survival time of 15 months and a 5-year survival rate of less than 5%. Regulated cell death (RCD) is the autonomous and orderly cell death under genetic control, controlled by precise signaling pathways and molecularly defined effector mechanisms, modulated by pharmacological or genetic interventions, and plays a key role in maintaining homeostasis of the internal environment. The comprehensive and systemic landscape of the RCD in glioma is not fully investigated and explored. After collecting 18 RCD-related signatures from the opening literature, we comprehensively explored the RCD landscape, integrating the multi-omics data, including large-scale bulk data, single-cell level data, glioma cell lines, and proteome level data. We also provided a machine learning framework for screening the potentially therapeutic candidates. Here, based on bulk and single-cell sequencing samples, we explored RCD-related phenotypes, investigated the profile of the RCD, and developed an RCD gene pair scoring system, named RCD.GP signature, showing a reliable and robust performance in predicting the prognosis of glioblastoma. Using the machine learning framework consisting of Lasso, RSF, XgBoost, Enet, CoxBoost and Boruta, we identified seven RCD genes as potential therapeutic targets in glioma and verified that the SLC43A3 highly expressed in glioma grades and glioma cell lines through qRT-PCR. Our study provided comprehensive insights into the RCD roles in glioma, developed a robust RCD gene pair signature for predicting the prognosis of glioma patients, constructed a machine learning framework for screening the core candidates and identified the SLC43A3 as an oncogenic role and a prediction biomarker in glioblastoma.

Mime: A flexible machine-learning framework to construct and visualize models for clinical characteristics prediction and feature selection

“…Machine learning (ML), a branch of computer science that learns from complex datasets to develop a high-accuracy predictive model, has become a popular tool in medical research [5] . Several diagnostic and prognostic models based on machine learning have been developed from transcriptional data for various cancer types [6] , [7] , [8] , [9] , [10] . Since the performance of machine learning models from large amounts of transcriptional data can vary, it is often recommended to compare the results derived from several methods and select the optimal one for further application [6] , [11] , [12] .…”