Predicting Gene Mutations in Colon Cancer Using Long-Term Temporal Dependency Learning on a Directed Co-Occurrence...

Area of Science:

• Computational Biology
• Genomics
• Cancer Research

Background:

• Accurate prediction of mutational dependencies is crucial for understanding tumor evolution and cancer progression.
• Colorectal tumorigenesis often exhibits stepwise mutational patterns, necessitating advanced modeling techniques.

Purpose of the Study:

• To develop a graph-based approach for inferring gene mutational order and longest paths in colorectal cancer.
• To enhance the accuracy of predicting gene mutations during tumor evolution using deep learning models.

Main Methods:

• A graph-based approach was used to infer gene mutational order from co-occurrence conditional probabilities.
• Long-Short-Term Memory (LSTM) and dilated Convolutional Neural Network (CNN) models were trained to predict gene mutations.
• Analysis utilized a large gene network and a cohort of human colon adenocarcinoma samples.

Main Results:

• The study inferred a longest path and global mutational orders from a directed co-occurrence asymmetry graph.
• Both LSTM and CNN models demonstrated high prediction accuracies for gene mutations.
• The proposed methods significantly improved precision and recall in mutation prediction compared to previous studies, with the longest weighted path yielding the best performance.

Conclusions:

• The developed graph-based deep learning approach effectively models tumor evolution by predicting gene mutational order.
• This method offers a significant advancement in mutation prediction accuracy for colorectal cancer.
• The findings contribute to a better understanding of cancer progression and have implications for early diagnosis and interventions.