Linking physiology and transcriptional profiles by quantitative predictive models.

Area of Science:

• Systems Biology
• Genomics
• Computational Biology

Background:

• Physiological processes are complex and influenced by gene expression.
• Predictive modeling from transcriptional data is challenging due to noise and variability.
• Integrating diverse biological data types is crucial for a holistic understanding.

Purpose of the Study:

• To develop a robust methodology for constructing quantitative, predictive physiological models from transcriptional profiles.
• To enable accurate prediction of physiological traits using gene expression data.
• To provide an integrative framework for building predictive models using various biological data.

Main Methods:

• Utilized a modified partial least squares (PLS) regression for model construction.
• Incorporated gene pre-selection based on signal-to-noise ratio (SNR).
• Employed consensus ranking across thousands of trials with varying training samples for robust gene set identification.

Main Results:

• Successfully constructed quantitative predictive models for mouse age, insulin, and leptin levels using liver transcriptional data.
• Developed predictive models for yeast mutant growth rate and Drosophila sample age from literature data.
• Demonstrated that highly predictive models are achievable with current DNA microarray precision when physiological variation is controlled.

Conclusions:

• The developed methodology enables the construction of accurate predictive physiological models from transcriptional data.
• Genes identified are crucial for collectively predicting physiological phenotypes.
• The method offers a flexible framework adaptable to various physiological or cellular data types for integrative modeling.