UNNT: A novel Utility for comparing Neural Net and Tree-based models

Area of Science:

• Computational biology
• Machine learning in oncology

Background:

• Deep learning (DL) is increasingly used in cancer research, particularly for drug response prediction.
• Advances in data generation, algorithms, and computing power accelerate DL adoption.
• Comparing different machine learning models for complex tasks like drug response is challenging.

Purpose of the Study:

• To evaluate tree-based models against deep learning models for cancer drug response prediction.
• To develop an accessible tool for comparing neural network and tree-based models.
• To demonstrate the utility of XGBoost over Convolutional Neural Networks (CNNs) for single drug response problems.

Main Methods:

• Exploration of tree-based models, specifically XGBoost (eXtreme Gradient Boosting).
• Comparison of XGBoost performance against deep learning models like CNNs.
• Development of an open-source library, UNNT (A novel Utility for comparing Neural Net and Tree-based models), for model comparison.
• Application to cancer drug response datasets.

Main Results:

• XGBoost demonstrated superior accuracy for single drug response prediction compared to CNNs.
• The UNNT library facilitates easier training and comparison of CNNs and XGBoost.
• The developed approach is applicable beyond cancer research, including domains like chemistry.

Conclusions:

• Tree-based models, such as XGBoost, offer advantages over deep learning models for specific cancer drug response tasks.
• The UNNT library provides a valuable, open-source resource for researchers comparing different machine learning architectures.
• This work highlights the potential for improved predictive modeling in drug discovery and related scientific fields.