MACHINE LEARNING APPROACH FOR THE PREDICTION OF BLADDER CANCER STAGES BASED ON NEXT-GENERATION SEQUENCING DATA

Abstract

ABSTRACT Purpose: The purpose of this paper is to apply Machine learning algorithms for the classification of various stages of bladder Cancer (BCa) based on RNA-Seq transcriptome per million(TPM) gene counts data and its corresponding pathological stages from the TCGA database. The objective is to assess classification performance across different stages. Design/Methodology/Approach: This study applied a computational research design on publicly available BCa gene expression data from The Cancer Genome Atlas (TCGA). Multiple supervised machine learning predictive modelling algorithms were trained and evaluated, with a nested crossvalidation design. A forward feature selection technique was used to select the best features for ML classifiers, in conjunction with 3-fold nested cross-validation (nCV), applied to binary classification using machine learning algorithms. The dataset preprocessing was carried out in two phases using the R and Python programming languages. Research Limitation: Reliance on downloaded data raises concerns about the data generator’s bias. Findings: This study suggests that TPM profiles of bulk RNA-seq samples are unreliable for separating adjacent stages of bladder cancer. These findings suggest that bulk transcriptomic data should not be used solely to inform treatment decisions for bladder cancer. Rather, it will be more informative to integrate molecular subtyping with multi-omics data or to make models that can directly predict clinical outcomes. Practical Implication: In practical terms, these findings suggest that bulk RNAseq TPM transcriptomic data should not be solely relied on for staging bladder cancer in clinical or predictive settings. Instead, more informative approaches such as combining molecular subtypes, integrating multi-omics data, or focusing on models that predict clinical outcomes are likely to provide greater value for decision-making and future research. Social Implication: This highlights the effect of over-relying on AI diagnostics that do not capture the full biological characteristics, which is essential for protecting patient safety. Originality/Value: This research examined the application of machine learning algorithms to predict bladder cancer stages using RNA-seq TPM gene-count NGS data from the TCGA database, a method that researchers have not previously considered. Keywords: Bioinformatics. bladder cancer. machine learning, next-generation sequencing