improving-platelet-rna-based-diagnostics

Improving-platelet-rna-based-diagnostics performs a comparative analysis of machine learning models for pan-cancer detection and multiclass classification using platelet RNA expression data.

Key Features:

• Large-scale dataset compilation: The final data collection included platelet RNA samples from 1397 cancer patients across 17 cancer types and 354 asymptomatic, presumed healthy donors, with quality filtering applied to assert data collection quality.
• Feature selection of RNA transcripts: The tool assesses feature selection methods for RNA transcripts as part of the machine learning analysis.
• Pan-cancer detection: Simple logistic regression achieves a 68% cancer detection rate at 99% specificity, exceeding the previously used method by 5%.
• Multiclass classification: The tool distinguishes between five cancer types with 79.38% accuracy, exceeding the previously used method by 9.65%.
• Open-source code and data processing pipelines: The code and data processing pipelines are publicly available to serve as a baseline for further research.

Scientific Applications:

• Cancer detection via liquid biopsy: The tool supports minimally invasive cancer detection and monitoring using platelet RNA as a liquid biopsy biomarker, applicable even at early disease stages.
• Multiclass cancer classification: The tool enables classification among multiple cancer types based on platelet RNA expression profiles.
• Benchmarking machine learning methods: The tool provides a comparative framework to evaluate which machine learning techniques are most effective for complex, multidimensional RNA expression data.

Methodology:

The study compiled and quality-filtered existing platelet RNA datasets, then assessed an array of machine learning models and techniques including feature selection of RNA transcripts for pan-cancer detection and multiclass classification, with simple logistic regression identified as the best-performing model.