Más-o-menos: a simple sign averaging method for discrimination in genomic data analysis

Area of Science:

• Bioinformatics
• Biostatistics
• Machine Learning

Background:

• Translating genomic signatures into clinical practice requires accurate patient subgroup discrimination.
• While sophisticated algorithms exist, simpler methods are often preferred but lack formal investigation.
• This study addresses the need for a well-defined and evaluated simple algorithm for prognostic scoring.

Purpose of the Study:

• To formally define and systematically investigate the performance of a simple prognostic scoring method called más-o-menos.
• To compare the discrimination performance of más-o-menos against established complex methods like lasso and ridge regression.
• To demonstrate the utility of más-o-menos in real-world cancer gene expression datasets.

Main Methods:

• Defined the más-o-menos algorithm: summing standardized predictors weighted by marginal association signs.
• Conducted theoretical analysis and simulations to study the algorithm's behavior.
• Applied más-o-menos to 27 independent gene expression studies across bladder, breast, and ovarian cancers (3833 patients).

Main Results:

• The más-o-menos method achieves good discrimination performance.
• Its performance is comparable to, and sometimes superior to, computationally intensive methods like lasso and ridge regression.
• Demonstrated effectiveness in extensive analyses of diverse cancer types.

Conclusions:

• The simple más-o-menos algorithm is a viable and effective tool for prognostic scoring in clinical settings.
• It offers a computationally efficient alternative to complex machine learning models for genomic data analysis.
• The method is readily available for use in survival analysis through the survHD package.