Multimodal analysis of methylomics and fragmentomics in plasma cell-free DNA for multi-cancer early detection and localization

BACKGROUND

• Cancer when being detected early can provide higher chance of survival and better quality of life.
• Circulating tumor DNA (ctDNA) has emerged as a promising biomarker for detecting and locating cancers as it harbors both genetic and epigenetic alterations present in the tumor. However, the low fraction of ctDNA in patients' plasma and its heterogeneity among cancer types lead to the high sequencing cost.
• We developed a multimodal assay called SPOT-MAS (Screening for the Presence Of Tumor by Methylation And Size) to analyze multiple ctDNA features for simultaneously detecting and locating cancer from a single blood test.

METHODS

Workflow of SPOT-MAS assay for multi-cancer detection and localization

Patient recruitment:

• The discovery cohort dataset was used to train a binary classification model (cancer detection) and a multiclass classification model (cancer localization).
• The validation cohort dataset was used to externally validate the performance of machine learning models

RESULTS

The cancer detection model (ensemble stacking) exhibited the best performance and outperformed the single-feature models

Discovery

Validation

Discovery

Validation

Discovery

Validation

The cancer localization model (graph convolutional neutral network) enabled prediction of cancer types, with an accuracy of 70%

Discovery

Validation

FUTURE DIRECTIONS FOR RESEARCH

We are conducting prospective clinical trials to evaluate the utility of SPOT-MAS as a complementary method in a national screening program.