COOBoostR: An Extreme Gradient Boosting-Based Tool for Robust Tissue or Cell-of-Origin Prediction of Tumors.

Area of Science:

• Computational biology
• Genomics
• Cancer research

Background:

• Identifying the tissue or cell of origin is crucial for understanding cancer development and guiding treatment strategies.
• Existing computational methods for predicting cancer origin have limitations in speed and accuracy.

Purpose of the Study:

• To introduce COOBoostR, a novel computational method for predicting the tissue or cell of origin for various cancer types.
• To leverage regional somatic mutation density and chromatin mark features for improved prediction accuracy and speed.

Main Methods:

• Developed COOBoostR, an extreme gradient boosting-based machine learning algorithm.
• Integrated ChIP-seq derived chromatin data with regional somatic mutation density data.
• Ranked chromatin marks to identify the best explanatory features for somatic mutation landscapes.

Main Results:

• COOBoostR demonstrated superior prediction speed compared to existing random forest methods.
• Achieved high prediction accuracy: 76.99% for normal cells/tissue, 95.65% for precancerous lesions, and 89.39% for cancer cells.
• Identified a dynamic somatic mutation accumulation process intertwined with chromatin mark changes in early-stage development.

Conclusions:

• COOBoostR provides an effective and efficient tool for predicting cancer tissue of origin.
• Chromatin marks play a significant role in shaping the somatic mutation landscape, even before precancerous lesion initiation.
• Findings suggest early-stage chromatin alterations influence somatic mutation accumulation, impacting cancer development.