AI-powered biomarkers may struggle to pass clinical trials
A comprehensive review published in the journal Signal Transduction and Targeted Therapy discusses the role of artificial intelligence in biomarker discovery and how high accuracy alone is not enough for clinical success.
A new review article authored by R. Dinc and N. Ardic, published in the journal Signal Transduction and Targeted Therapy, examines the place of artificial intelligence in biomarker discovery, validation, and clinical translation processes.
The Role of Artificial Intelligence in Biomarkers
Across a wide range of data from blood samples to scans and wearable devices, artificial intelligence technologies support biomarker discovery, validation, and clinical integration in disease processes.
Researchers R. Dinc and N. Ardic examine the potential of AI in these processes, detailing the opportunities it offers for modern medicine and the fundamental challenges encountered.
Data Diversity and Encountered Obstacles
AI models have the capacity to integrate many different data types, including genomic, transcriptomic, proteomic, metabolic, imaging, cellular, and digital data.
However, elements such as bias, confounding factors, data heterogeneity, overfitting, and poor external validation can significantly limit the translation of these findings into clinical fields.
Examined Biomarker Fields
The published review study covers many areas, including molecular approaches, cellular biomarkers such as neutrophil extracellular traps, and circulating markers like cell-free DNA.
Additionally, radiomics-based imaging biomarkers and digital data obtained from wearable devices are among the focus points of the study.
Machine Learning Models Used
Classical machine learning and deep learning models help researchers uncover the complex nature of disease signatures.
Advanced technologies such as representation learning, multimodal machine learning, graph neural networks, and foundation models are actively used in this process.
Requirements for Clinical Success
A large portion of AI-derived biomarkers still remain at a correlational level and cannot establish direct causality.
The authors emphasize that mechanistic AI approaches and external validation in independent cohorts are essential to bridge the gap between initial discovery and patient care.