RAMAN-BASED AI PLATFORM TO ACCELERATE THE DIAGNOSIS AND MANAGEMENT OF GLIOMAS VIA RAPID METHYLATION PROFILING
Recommended Citation
Lita A, Sjöberg J, Păcioianu D, Siminea N, Celiku O, Păun A, Gilbert M, Noushmehr H, Petre I, Larion M. RAMAN-BASED AI PLATFORM TO ACCELERATE THE DIAGNOSIS AND MANAGEMENT OF GLIOMAS VIA RAPID METHYLATION PROFILING. Neuro Oncol 2023; 25(Suppl 5):v182-v182.
Document Type
Conference Proceeding
Publication Date
11-10-2023
Publication Title
Neuro Oncol
Abstract
BACKGROUND: DNA methylation profiling and subtyping of gliomas has become an integral part of the diagnosis and treatment of glioma patients but remains a challenge due to the sample preparation and time requirements of the existing methylation profiling technology. We hypothesized that machine learning and Raman spectroscopy, which creates fast molecular fingerprints of samples in situ based on chemical signatures rather than the use of exogenous dyes, can provide a fast alternative to the existing approaches for methylation profiling and classification of gliomas. METHODS: Raman spectroscopy at the spatial resolution of less than 300nm was used for molecular fingerprinting of the regions of interest using 1mm2 Formalin-Fixed Paraffin-Embedded (FFPE) tissue spots from 45 patient samples with known LGm1 to LGm6 methylation subtypes. Spectral information (over 1738 frequencies) was used to construct tumor/non-tumor, IDH1WT/IDH1mut, and methylation-subtype classifiers. Oversampling was used to obtain subtype-balanced data distributions. Supervised wrapper methods and random forests were used to identify the top 20 most discriminatory Raman frequencies. Stimulated Raman spectroscopy was used to validate the findings. RESULTS: We developed APOLLO - a novel platform based on spontaneous Raman spectroscopy and machine learning - for predicting the DNA methylation subtypes of FFPE glioma tissue specimens. APOLLO discriminates tumors from non-tumor areas with 98% accuracy and discriminates IDH1mut versus IDH1WT tumors with 82% accuracy. APOLLO also achieved high discriminatory power between G-CIMP-high and G-CIMP-low molecular phenotypes (ROC of 0.75), subtypes of lower-grade IDH1mut gliomas with significantly different clinical outcomes. We determined that the Raman shifts important for discriminating IDH1mut versus IDH1wt tumors are associated with novel lipid-metabolism signatures of IDH1mut glioma. CONCLUSIONS: The development of APOLLO allows fast, reliable, and accurate prediction of methylation subtypes of glioma which can speed diagnosis and, once validated on fresh tissue, can be implemented in the operating room.
Volume
25
Issue
Suppl 5
First Page
v182
Last Page
v182
