Deep generative AI models analyzing circulating orphan non-coding RNAs enable detection of early-stage lung cancer
Mehran Karimzadeh1, Amir Momen-Roknabadi1, Taylor B Cavazos1
1Exai Bio Inc., Palo Alto, CA, US.
Nature Communications
|November 21, 2024
Summary
This study introduces Orion, an AI model for liquid biopsies, detecting non-small cell lung cancer (NSCLC) using orphan non-coding RNAs. Orion achieves high sensitivity and specificity, outperforming existing methods for early cancer detection.
Area of Science:
- Oncology
- Artificial Intelligence
- Biomarker Discovery
Background:
- Liquid biopsies offer non-invasive cancer detection potential.
- Developing robust tests requires high-dimensional data from diverse patient cohorts.
- Orphan non-coding RNAs (oncRNAs) are emerging as promising blood-based biomarkers.
Purpose of the Study:
- To develop a robust and generalizable liquid biopsy test for early cancer detection.
- To leverage variational auto-encoders for learning blood-based biomarker signatures.
- To evaluate a multi-task generative AI model, Orion, for non-small cell lung cancer (NSCLC) detection.
Main Methods:
- Analysis of serum samples from 1050 individuals with NSCLC and matched controls.
- Utilizing a multi-task generative AI model (Orion) based on variational auto-encoders.
- Comparative analysis against commonly used methods for biomarker signature learning.
Main Results:
- Orion demonstrated superior performance and generalizability compared to existing methods.
- The AI model achieved 94% sensitivity and 87% specificity for NSCLC detection across all stages.
- Orion outperformed other methods by over 30% in sensitivity on held-out validation datasets.
Conclusions:
- Generative AI models like Orion can effectively learn robust signatures from complex biological data for liquid biopsies.
- Orion represents a significant advancement in non-invasive cancer detection, particularly for NSCLC.
- The developed AI model shows promise for revolutionizing early tumor detection and cancer care.
