Optimal microRNA Sequencing Depth to Predict Cancer Patient Survival with Random Forest and Cox Models

Area of Science:

• Oncology
• Bioinformatics
• Genomics

Background:

• Tumor profiling using RNA sequencing (RNA-seq) is crucial for predicting patient survival.
• Study design requires calibrating sequencing depth and patient cohort size under cost constraints.
• Balancing these parameters is essential for effective survival prediction models.

Purpose of the Study:

• To benchmark the impact of patient number and RNA-seq sequencing depth (miRNA-seq and mRNA-seq) on predictive capabilities.
• To compare the performance of the Cox model with elastic net penalty and random survival forest for survival prediction.
• To evaluate the contribution of miRNA and mRNA data, individually and combined, to prediction accuracy.

Main Methods:

• Comparative analysis of Cox model with elastic net penalty and random survival forest.
• Benchmarking predictive performance using varying numbers of patients and sequencing depths for miRNA-seq and mRNA-seq data.
• Assessment of prediction improvement over clinical data alone.

Main Results:

• Cox model and random survival forest showed comparable predictive capabilities, with cancer-specific variations.
• miRNA and/or mRNA data significantly improved prediction accuracy compared to clinical data alone.
• mRNA-seq generally outperformed miRNA-seq, except in lung adenocarcinoma where miRNA-seq was superior.
• Reduced sequencing depth maintained predictive ability for most cancers, enabling cost savings.
• Fewer patients in training datasets were sufficient for both models, allowing subgroup analysis.

Conclusions:

• RNA-seq data, particularly mRNA, enhances cancer survival prediction, with miRNA-seq being valuable for specific cancers like lung adenocarcinoma.
• Sequencing depth and patient numbers can be reduced without compromising predictive power, optimizing study costs.
• These findings facilitate the development of more cost-effective and accurate patient survival prediction strategies.