Application of Machine Learning in the Identification of Diagnostic Cancer Biomarkers Using Transcriptomic Data: A Systematic Review

Background and Objective: Diagnosing cancer at early stages remains one of the most effective ways to lower mortality rates. Transcriptomic datasets such as RNA-Seq and gene expression microarrays offer a deep look into the gene expression changes tied to various malignancies. Yet, because these datasets are complex and high-dimensional, traditional analysis methods often fall short, prompting a shift toward advanced analytical frameworks. Over the past few years, machine learning has become a central approach for handling large-scale transcriptomic data. This systematic review set out to explore how machine learning techniques are applied to discover and validate new diagnostic biomarkers for cancer using transcriptomic profiles. Materials and Methods: We carried out this systematic review following the PRISMA guidelines. A thorough search was conducted across international databases including PubMed, Scopus, and ScienceDirect, alongside Iranian databases like SID and Civilica, covering all publications up to ۲۰۲۶ without any time filters. Keywords used in the search strategy included Machine Learning, Transcriptomics, Cancer Diagnosis, and Biomarkers. Following the PRISMA screening process based on strict inclusion and exclusion criteria, relevant studies were selected and data extraction was performed systematically. Results: Findings from the reviewed literature show that specific machine learning algorithms most notably Random Forest (RF), Support Vector Machine (SVM), and LASSO perform exceptionally well in feature selection, reducing dimensionality, and pinpointing diagnostic gene biomarkers from transcriptomic data. When researchers paired these algorithms with standard bioinformatics workflows, the diagnostic accuracy for cancers such as breast, colorectal, and gastric cancer noticeably improved, while also opening doors to potential therapeutic targets. Conclusion: Machine learning holds robust potential for parsing complex transcriptomic data accurately and efficiently to find reliable cancer biomarkers. Even so, before these models can find a place in routine clinical workflows, they will need rigorous validation in larger clinical cohorts alongside better standardization in data preprocessing.

کلیدواژه ها:

Machine Learning ، Transcriptomics ، Cancer Diagnosis ، Biomarkers ، Gene Expression Analysis