A comparison of computational methods for expression forecasting.

Area of Science:

• Computational Biology
• Genomics
• Machine Learning

Background:

• Machine learning (ML) methods are increasingly used to predict gene expression changes following genetic perturbations.
• The accuracy and reliability of these predictive models are not well-established.
• A need exists for systematic evaluation of ML methods in this domain.

Purpose of the Study:

• To create a comprehensive benchmarking platform for evaluating gene expression forecasting methods.
• To assess the performance of various ML methods, parameters, and data sources.
• To identify conditions under which expression forecasting can be reliable.

Main Methods:

• Developed a benchmarking platform integrating 11 large-scale perturbation datasets.
• Incorporated a software engine supporting diverse ML-based expression forecasting methods.
• Systematically evaluated method performance, parameter choices, and auxiliary data utilization.

Main Results:

• Found that ML-based expression forecasting methods frequently fail to outperform simple baseline predictions.
• Identified specific parameters and data sources that influence forecasting accuracy.
• Demonstrated variability in performance across different datasets and methods.

Conclusions:

• Current gene expression forecasting methods show limited accuracy and often do not surpass basic approaches.
• The developed platform provides a valuable resource for method improvement and identifying successful forecasting contexts.
• Further research is needed to enhance the predictive power of ML models for gene expression.