Discovering collectively informative descriptors from high-throughput experiments.

Area of Science:

• Bioinformatics
• Computational Biology
• Data Science

Background:

• High-throughput technology generates complex biological datasets.
• Combining data from similar studies enhances reliability and generalizability.
• New insights can be derived from integrated datasets to guide future research.

Purpose of the Study:

• To introduce BLANKET, a novel algorithm for symmetric analysis of two related experiments.
• To assess the informativeness of descriptors across different experimental datasets.
• To identify optimal, minimal subsets of descriptors for predicting experimental outcomes.

Main Methods:

• BLANKET analyzes descriptor informativeness from two experiments with intersecting descriptor sets and consistent case/control definitions.
• It generates ranked lists of descriptors and shortlists (p, q) from each experiment.
• The algorithm uses contingency tables and Right Fisher Exact Test (RFET) scores to evaluate shortlist intersections.

Main Results:

• BLANKET identifies pairs of descriptor shortlists with statistically significant intersections (RFET score < threshold).
• The threshold is determined by dataset size (n) and shortlist length limits, indicating a rare quality of intersection.
• The method is computationally moderate and easy to conceptualize.

Conclusions:

• BLANKET aids researchers in finding minimal descriptor subsets for efficient and reliable prediction of experimental outcomes.
• The algorithm can be applied to existing databases to discover optimal predictive descriptor sets.
• This approach enhances the predictive power and accuracy of biological data analysis.