MINE: a new way to design genetics experiments for discovery

Area of Science:

• Genetics
• Bioinformatics
• Experimental Design

Background:

• Classical experimental design assumes more samples than parameters (n > p), risking overfitting in omics studies where parameters exceed samples (p > n).
• Existing methods struggle with high-dimensional data common in omics research.

Purpose of the Study:

• Introduce the Maximally Informative Next Experiment (MINE) as a novel adaptive experimental design strategy.
• Address the challenges of parameter-rich (p > n) experimental data, particularly in omics.

Main Methods:

• MINE employs an ensemble method to predict outcomes of future experiments based on existing models.
• It adaptively selects the most informative next experiment to guide discovery.
• The approach integrates smaller, manageable experiments to replace large, classical designs.

Main Results:

• MINE facilitates exploration of data for novel relationships even when n < p.
• Demonstrated effectiveness in multiyear studies involving genetic networks (Neurospora crassa) and genome-wide association studies (Sorghum bicolor).
• MINE provides a model-guided, adaptive strategy for large-scale omics research.

Conclusions:

• MINE offers a robust solution for experimental design in high-dimensional biological data (p > n).
• This adaptive, model-guided approach enhances discovery efficiency in omics and agricultural research.
• MINE enables more tractable and informative experimental trajectories compared to classical designs.