Transcriptomics enables comprehensive analysis of RNA molecules, with RNA-sequencing providing key insights into gene expression. However, bioinformatics analyses remain limited by the chosen strategy, an aspect that can be improved through machine learning integration. This study compared three RNA-seq pipelines (step-by-step, nf-core, and OneStop) implemented in HPC and GUI environments, using data from BioProject GSE130437 of breast cancer MCF7 cell lines resistant and susceptible to palbociclib. All pipelines identified a comparable number of genes ($\sim 14000$ per condition), but each detected unique subsets not shared across methods. Differential gene expression analysis revealed distinct transcriptomic profiles, with limited overlap among pipelines. LUM, KCNK2, and LINC00052 transcripts were consistently identified across pipelines and in the reference study, while others appeared only in a pipeline-specific context, highlighting biases introduced by preprocessing steps, statistical models, and default parameters. Performance assessments demonstrated trade-offs, the step-by-step approach offered flexibility but required high expertise and longer runtimes; nfcore achieved the fastest and most reproducible results through containerized workflows; and OneStop provided the most user-friendly interface, although scalability and parameter customization were limited. Overall, pipeline selection substantially impacts DEG outcomes and downstream interpretations. Considering scalability, reproducibility, and machine learning integration, nf-core proved the most practical choice for large datasets, while OneStop lowered entry barriers for non-specialists. These findings emphasize the need for careful pipeline selection and the development of ML-ready outputs to strengthen reproducibility and expand transcriptomics applications toward predictive modeling and personalized medicine.