Mime: A flexible machine-learning framework to construct and visualize models for clinical characteristics prediction...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• High-throughput sequencing has advanced cancer research, enabling machine learning models for outcome prediction.
• Existing tools lack a user-friendly, open-source R package for state-of-the-art machine learning algorithms.
• Predictive modeling in cancer requires robust and accessible computational frameworks.

Purpose of the Study:

• To introduce Mime, a flexible R package for constructing machine learning integration models.
• To streamline the development of accurate predictive models using complex transcriptional data.
• To identify critical genes and signatures associated with patient prognosis and clinical response.

Main Methods:

• Developed Mime, an open-source R package providing a computational framework for machine learning models.
• Constructed de novo PIEZO1-associated signatures using Mime for in silico validation.
• Applied various machine learning algorithms within Mime to assess immunotherapy response prediction.

Main Results:

• Mime demonstrated high accuracy in predicting patient outcomes using PIEZO1-associated signatures, outperforming existing models.
• PIEZO1-associated signatures accurately inferred immunotherapy response across different algorithms.
• SDC1 was identified as a potential therapeutic target for glioma based on the PIEZO1-associated signatures.

Conclusions:

• The Mime package offers a user-friendly solution for building machine learning-based predictive models.
• Mime facilitates the identification of prognostic biomarkers and potential therapeutic targets in cancer.
• This framework has broad applicability for advancing cancer research and clinical insights.