muSignAl: An algorithm to search for multiple omic signatures with similar predictive performance

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics and Proteomics

Background:

• Multidimensional omic datasets frequently contain correlated features, enabling the identification of multiple biological signatures with comparable predictive accuracy for a given phenotype.
• Exploring these signatures is challenging due to limited sample sizes and the high computational expense associated with exhaustive combinatorial searches.

Purpose of the Study:

• To develop an efficient algorithm for selecting multiple biological signatures with similar predictive performance.
• To overcome the limitations of low sample size and high computational cost in exploring feature combinations.

Main Methods:

• Developed the muSignAl (multiple signature algorithm), designed to identify multiple signatures with similar predictive performance.
• The algorithm systematically bypasses the need to explore all possible feature combinations, reducing computational burden.
• Demonstrated the algorithm's workflow using a proteomics dataset.

Main Results:

• The muSignAl algorithm successfully selects multiple signatures with comparable predictive performance.
• The method efficiently navigates the combinatorial search space, mitigating high computational costs.
• The proteomics dataset example illustrates the practical application and workflow of the algorithm.

Conclusions:

• muSignAl offers a valuable tool for bioinformatics research, facilitating the understanding of relationships between biological features and phenotypes.
• The algorithm supports the discovery and development of biomarker panels, potentially optimizing development costs through the identification of equally effective signature sets.
• The source code for muSignAl is publicly available for broader research application.