# Artificial intelligence-based miRNA analysis for precision oncology: diagnostic and prognostic insights

## TL;DR

This paper explores how combining AI with miRNA analysis can improve cancer detection, predict outcomes, and guide personalized treatments, while highlighting challenges in implementation.

## Contribution

The paper presents a comprehensive analysis of AI models for miRNA-based cancer diagnostics and prognostics, identifying key miRNA signatures and performance benchmarks.

## Key findings

- AI models analyzing miRNA signatures achieved high diagnostic accuracy (AUC > 0.90) for cancers like gastric, breast, and lung.
- A 3-miRNA combination (hsa-let-7i-3p, miR-362-3p, and miR-3651) predicts cancer stage across eight types.
- Random forest models achieved perfect AUCs (1.00) in some validations, but challenges like data fragmentation and racial bias remain.

## Abstract

MicroRNAs (miRNAs), small molecules that fine-tune gene activity, are consistently disrupted in cancer. Found stably in blood and other fluids, their unique cancer-associated patterns offer a promising route for non-invasive detection and monitoring. Merging artificial intelligence (AI) with miRNA analysis could revolutionize our understanding and treatment of cancer; however, reliably integrating these tools into clinics remains challenging.

A multi-database search was executed until July 2025 using integrated miRNA-related descriptors and AI/ML ontologies such as support vector machine (SVM), random forest (RF), artificial neural network (ANN), logistic regression (LR), principal component analysis (PCA), and hierarchical clustering (HC), to interpret complex miRNA data in cancer. Our focus was on considering research article related to early cancer detection, prediction of patient outcomes, and guiding personalized treatments.

AI models analysing miRNA signatures demonstrate remarkable accuracy [area under the curve (AUC) often exceeding 0.90] in diagnosing various cancers, such as gastric, breast, and lung cancer (LC). For example, SVM proved highly effective for breast cancer (BC) detection. Crucially, AI helps identify small miRNA sets linked to cancer progression, such as a 3-miRNA combination (hsa-let-7i-3p, miR-362-3p, and miR-3651) that predicts disease stage across eight cancers. RF models achieved near-perfect AUCs (1.00) in some validation studies. AI also identifies miRNAs, such as a specific 5-miRNA group in BC, that signal resistance to chemotherapy. However, significant roadblocks persist: fragmented and non-standardized data, AI tools that exhibit disparate performance across demographic groups (evidenced by racial bias in mammography algorithms), and unaddressed validation gaps.

The powerful combination of AI and miRNA biology is reshaping oncology. It enables earlier cancer detection, more accurate forecasts of disease course, and therapies tailored to the individual. Realizing this potential demands AI models that clinicians can understand and trust, diverse datasets to ensure tools work fairly for all patients, and close teamwork across disciplines to integrate these advances into real-world care. This convergence marks a pivotal shift towards proactive, precise, and accessible cancer management globally.