ZeitZeiger: supervised learning for high-dimensional data from an oscillatory system.

Area of Science:

• * Systems Biology
• * Computational Biology
• * Chronobiology

Background:

• * Biological systems frequently exhibit temporal or spatial oscillations.
• * Analyzing oscillatory data poses challenges for conventional supervised learning methods.
• * Accurate prediction of periodic biological variables is crucial for understanding these systems.

Purpose of the Study:

• * To introduce ZeitZeiger, a novel computational method for predicting periodic variables from high-dimensional biological data.
• * To develop a robust predictor of circadian time (internal time of day) across multiple organs.
• * To demonstrate ZeitZeiger's efficiency and accuracy compared to existing state-of-the-art methods.

Main Methods:

• * ZeitZeiger learns sparse representations of periodic variations in training data.
• * Maximum-likelihood estimation is employed for prediction on test data.
• * The method was applied to genome-wide gene expression data from mammalian circadian oscillators.

Main Results:

• * ZeitZeiger accurately predicted circadian time in 12 mouse organs within approximately 1 hour using only 13 genes.
• * A multi-organ predictor of circadian time was successfully developed and validated on 20 additional datasets (nearly 800 samples).
• * ZeitZeiger outperformed the state-of-the-art approach in speed, accuracy, and gene utilization.

Conclusions:

• * ZeitZeiger provides an effective tool for predicting periodic variables in biological systems.
• * The method offers insights into the structure and dynamics of biological oscillators.
• * ZeitZeiger is expected to advance the conversion of high-dimensional oscillatory data into biological knowledge.