Bioinformatics Methods to Select Prognostic Biomarker Genes from Large Scale Datasets: A Review

Area of Science:

• Bioinformatics
• Genomics
• Biostatistics

Background:

• Survival datasets with molecular and clinical data are abundant, leading to numerous proposed prognostic biomarkers.
• Clinical validation and routine use of these biomarkers remain limited, largely attributed to a high false discovery rate.
• The high number of genes tested relative to patient cohorts contributes to this challenge.

Purpose of the Study:

• To review recent methodologies for improving prognostic biomarker discovery from survival data.
• To highlight advancements in variable selection, p-value definition, and biological knowledge integration.
• To illustrate concepts with a renal cancer dataset and provide supporting scripts.

Main Methods:

• Review of historical and recent methodologies in prognostic biomarker discovery.
• Discussion of variable selection techniques, including improved lasso penalization.
• Exploration of accurate p-value definition, false discovery rate control, and network/pathway incorporation.

Main Results:

• Recent developments focus on three key areas: enhanced variable selection, refined p-value and false discovery rate control, and integration of biological networks.
• These advancements aim to improve the reliability and clinical applicability of prognostic biomarkers.
• Illustrative examples using a renal cancer dataset are provided.

Conclusions:

• New methodologies show promise for more accurate prognostic biomarker identification.
• Independent benchmarking with diverse datasets is crucial for validating these developments.
• Further methodological research is necessary to advance the field.