A Novel Machine Learning Based Identification Tool (ELECT) for Early Colorectal Cancer Detection Through Advanced Microbiome Composition Analysis

Overview

Colorectal cancer(CRC) ranks third in occurrence and second in mortality among all cancers. Current CRC identification methods are often ineffective due to the invasiveness of such procedures and long waiting times for test results. Most CRC cases are identified in the late stage which has a drastically low 14% 5-year survival rate(5ySR). However, if found at early stages, the 5ySR of CRC cases is around 90%. Thus, early cancer detection is crucial to preventing CRC deaths. The ELECT project’s goal is to accurately detect CRC early on and identify high-correlation cancer biomarker flags. It utilizes an elastic-net regression machine learning model to predict cancer risk based on microbiota samples. Incremental hyperparameter tuning and feature selection were run simultaneously to select the best performing model and increase correlation score. The model was trained, validated, and tested on over 1.5 million unique gut bacterial oncology samples for robustness. After cross-validation, the final model was able to predict CRC with an accuracy greater than the current industry best by 9%. Model results in statistical cluster plots and heatmaps further demonstrate precise and accurate predictions. In addition, this model significantly reduced the dataset size needed by 99% – shrinking the initial pool of 5207 bacteria types needed to 43 of the most critical bacteria biomarkers. This research allows oncologists to quickly, non-invasively, and accurately identify at-risk CRC patients early yielding a greater cancer survivability. Project future work focuses on identifying CRC recurrence and piloting in clinical trials for application to real patient-cases.

Competition history

JSHS 2022 Category not listed

Resources

National JSHS abstract book

Related projects

ISEF · 2021