AI-Based Early Detection and Subtyping of Non-Small Cell Lung Cancer from Blood Samples Using Orphan Non-Coding RNAs

Generative AI allows us to identify the pattern of oncRNAs in blood, enhancing the sensitivity and specificity of an early detection assay

X: oncRNAs

Q: annotated RNAs

Z, S; Latent variables of the underlying distribution

X̂; re-constructed oncRNAs

NSCLC study cohort design with an emphasis on early-stage detection

• 887 subjects including 320 NSCLC cases and 567 individuals withoutcancer diagnosis
• NSCLC and non-cancer subjects were comparable with regards to age, sex, and BMI
• Individuals were recruited as part of four independent studies

Exai’s generative AI model allows learning a cross-validated^† pattern of oncRNAs among multiple studies

Generative AI model vs Linear model
for cancer detection from serum oncRNAs

^† Model was trained with 10-fold cross validation. Error bars and ROC boundaries show 95% confidence interval.

oncRNAs and AI enable novel subtyping of tumors from blood

Accuracy on par with tissue immunohistochemistry (IHC); potential to expand actionable therapeutic indications

X: oncRNAs

Q: annotated RNAs

Z, S; Latent variables of the underlying distribution

X̂; re-constructed oncRNAs

Summary