Reference-free transcriptome signatures for prostate cancer prognosis.

Area of Science:

• Computational biology
• Cancer genomics
• Transcriptomics

Background:

• RNA sequencing (RNA-seq) data is crucial for developing prognostic signatures in cancer outcome prediction.
• Current prognostic predictors often overlook non-canonical RNAs by relying on fixed gene annotations.
• Reference-free transcriptome classifiers, utilizing k-mers, offer an alternative approach independent of gene annotations.

Purpose of the Study:

• To compare the efficacy of conventional gene-based and novel reference-free signatures for prostate cancer risk and relapse prediction.
• To evaluate the performance of k-mer based classifiers against gene expression matrix based classifiers.

Main Methods:

• A standardized procedure was implemented to process both k-mer count matrices and gene expression matrices.
• Signatures were extracted from input data and subsequently validated on an independent dataset.
• Comparative analysis focused on risk and relapse prediction accuracy for prostate cancer.

Main Results:

• Both gene-based and k-mer based classifiers demonstrated comparable high performance in risk prediction.
• Both approaches showed markedly lower performance in predicting cancer relapse.
• Reference-free signatures identified novel long non-coding RNAs (lncRNAs) and variable regions of cancer driver genes not captured by gene-based methods.

Conclusions:

• Reference-free classifiers represent a promising strategy for discovering novel prognostic RNA biomarkers.
• This approach facilitates the identification of previously unrecognized RNA elements with prognostic value.
• Further research into reference-free methods could enhance cancer outcome prediction accuracy.