The most informative spacing test effectively discovers biologically relevant outliers or multiple modes in...

Area of Science:

• Bioinformatics
• Statistical genetics
• Transcriptomics

Background:

• Outlier and subgroup identification statistics (OASIS) are used to find transcriptomic features indicating distinct biological processes or patient subgroups.
• Existing OASIS methods have limitations in robustly identifying these features.

Purpose of the Study:

• To develop the 'most informative spacing test' (MIST) for unsupervised detection of transcriptomic features with outliers or multiple expression modes.
• To evaluate MIST's performance against existing OASIS methods.

Main Methods:

• Developed MIST by adapting concepts from existing OASIS methods in bioinformatics and statistics.
• Applied MIST to RNA-seq exon expression, RNA-seq exon junction expression, and microarray exon expression data.
• Compared MIST's performance with seven other OASIS methods.

Main Results:

• MIST robustly identified features that perfectly discriminate subjects by gender or fusion-gene status in pediatric acute megakaryoblastic leukemia.
• MIST effectively identified features related to gender or molecular subtype in adult acute myeloid leukemia.
• MIST demonstrated superior performance compared to seven other OASIS methods across different data types.

Conclusions:

• MIST is a powerful new tool for unsupervised identification of transcriptomic features indicative of biological subgroups.
• MIST offers improved robustness and accuracy in detecting clinically relevant molecular patterns in cancer datasets.
• MIST is available in the OASIS R package for broader research application.