IntegratedMRF: random forest-based framework for integrating prediction from different data types.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Random forest methods demonstrated superior performance in drug sensitivity prediction challenges.
• Integrating diverse genomic and epigenetic data can enhance drug response prediction accuracy.

Purpose of the Study:

• To develop an open-source R package, IntegratedMRF, for integrating drug response predictions.
• To provide a computational framework for estimating drug response prediction errors using ensemble approaches.
• To implement both univariate and multivariate random forest models for enhanced prediction.

Main Methods:

• Utilized univariate and multivariate random forest algorithms.
• Integrated various genomic and epigenetic characterizations as input features.
• Employed ensemble approaches for error estimation, including mean and confidence intervals.
• The multivariate random forest model accounts for correlations between output responses.

Main Results:

• The IntegratedMRF package offers a robust framework for drug response prediction.
• Multivariate random forest implementation shows improved performance when drug responses are correlated.
• The package facilitates the estimation of prediction errors with confidence intervals.

Conclusions:

• IntegratedMRF provides an advanced tool for leveraging genomic data in drug response prediction.
• The framework's ability to handle correlated responses offers an advantage over existing methods.
• The open-source R package is readily available for researchers in computational biology and bioinformatics.