Optimizing sample size for supervised machine learning with bulk transcriptomic sequencing: a learning curve...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Accurate sample classification via transcriptomics is vital for personalized medicine.
• Existing sample size calculation methods may not suit machine learning (ML) for sample classification.
• A methodological gap exists in determining optimal sample sizes for ML-based transcriptomics analyses.

Purpose of the Study:

• To present a novel computational approach for establishing the accuracy-versus-sample size relationship in transcriptomics data.
• To address the need for appropriate sample size determination in ML-driven transcriptomics studies.
• To facilitate the development of clinically useful classifiers for personalized treatment.

Main Methods:

• A novel computational approach using data augmentation and learning curve fitting.
• Comprehensive performance evaluation on microRNA and RNA sequencing data.
• Consideration of diverse data characteristics and algorithm configurations.

Main Results:

• The developed approach effectively establishes the accuracy-sample size relationship for transcriptomics data.
• Performance was validated across various data types and ML algorithms.
• The method provides a robust framework for sample size estimation.

Conclusions:

• The novel computational approach enhances the adoption of ML in transcriptomics.
• This method accelerates the translation of transcriptomics findings into clinical applications.
• Accessible code (Python, R) is provided to promote reproducibility and implementation.