Abstract:3550 Background: Cell-free DNA (cfDNA) methylation, fragmentation patterns, chromosome instability, and chromatin accessiblity have been previously shown to be valid plasma biomarkers for non-invasive cancer detection. However, conventional whole-genome bisulfite sequencing (WGBS) is unable to simultaneously profile all these biomarkers due to bisulfite-induced DNA damages. Here we developed a machine learning approach to comprehensively integrate multiple types of cancer genomic markers from enzyme-conversion-based low-pass whole-methylome sequencing (WMS) of plasma cfDNA to non-invasively detect colorectal adenocarcinoma. Methods: Plasma cfDNA sampels from 215 colorectal adenocarcinoma patients and 568 healthy individuals were collected and were split into the discovery and independent testing cohort. The discovery cohort includes 150 cancer patients and 398 healthy individuals and the independent testing cohort includes 65 cancer patients and 170 healthy individuals. Whole methylome sequencing (WMS) libraries were generated from enzymatically converted cfDNA and were subsequently paired-end sequenced at ̃2X coverage. The genome-wide methylation density, fragmentation fingerprints, chromosome instability, and chromatin accessibility were extracted from the WMS data and individually modelled via machine learning methods such as SVM, LR, GBDT, random forest. The final predictive model is an ensemble model integrating all uni-modal models. All models were trained and fitted on the discovery cohort. Results: Data of different modalities provide complementary information in separating the cancer patients from the healthy individuals. Unsupervised clustering of the individuals showed clear separation between cancer patients and healthy individuals. The final predictive model achieved AUC =0.982 in the discovery cohort and AUC =0.9664 in the independent testing cohort. Under a specificity of 94.71% (CI: 90% - 98.82%), sensitivity was 78.46% (CI: 60% - 92.31%) in the independent testing cohort. Separating the cancer patients into different stages, we found that the detection power is usuaul lower for early-stage cancer patients. Conclusions: These results demonstrate the first proof of principle on the feasibility of integrating multiple genomic cancer markers to non-invasively detect colorectal adenocarcinoma from WMS plasma cell-free DNA. A large prospective cohort study is planned to further validate its clinical performance.

Multispectral 3D DNA Machine Combined with Multimodal Machine Learning for Noninvasive Precise Diagnosis of Bladder Cancer

Study Progress of Noninvasive Imaging and Radiomics for Decoding the Phenotypes and Recurrence Risk of Bladder Cancer

Ratiometric 3D DNA Machine Combined with Machine Learning Algorithm for Ultrasensitive and High-Precision Screening of Early Urinary Diseases

An Integrated Double-Filtration Microfluidic Device for Detection of Extracellular Vesicles from Urine for Bladder Cancer Diagnosis.

Machine learning assisted microfluidics dual fluorescence flow cytometry for detecting bladder tumor cells based on morphological characteristic parameters

Urine-based SERS and multivariate statistical analysis for identification of non-muscle-invasive bladder cancer and muscle-invasive bladder cancer

Identification of two-dimensional copper signatures in human blood for bladder cancer with machine learning

Multiplexed analysis of small extracellular vesicle-derived mRNAs by droplet digital PCR and machine learning improves breast cancer diagnosis

Machine Learning-Based Detection of Bladder Cancer by Urine cfDNA Fragmentation Hotspots that Capture Cancer-Associated Molecular Features

A radiomics-based interpretable machine learning model to predict the HER2 status in bladder cancer: a multicenter study

Noninvasive Multiplexed Analysis of Bladder Cancer-Derived Urine Exosomes Via Janus Magnetic Microspheres.

Simultaneous detection of membrane protein and mRNA at single extracellular vesicle level by droplet microfluidics for cancer diagnosis

Multimodal analysis of cell-free DNA whole-methylome sequencing for cancer detection and localization

Profiling of Urine Carbonyl Metabolic Fingerprints in Bladder Cancer Based on Ambient Ionization Mass Spectrometry

Non-invasive screening of bladder cancer using digital microfluidics and FLIM technology combined with deep learning

DNA-framework-based multidimensional molecular classifiers for cancer diagnosis

Urinary extracellular vesicles-encapsulated miRNA signatures: A new paradigm for urinary bladder cancer diagnosis and classification

Early detection of colorectal adenocarcinoma by decoding epigenetic and DNA fragmentation fingerprints of plasma cell-free DNA.

Genome-Scale Multimodal Analysis of Cell-Free DNA Whole-Methylome Sequencing for Noninvasive Esophageal Cancer Detection

Applying fragmentomics profiles of urinary cell-free DNA for bladder cancer detection.

Ultra-deep sequencing with unique molecular identifier(UMI) for detection of ctDNA by fragment profiling using machine learning.