Random Projection for Fast and Efficient Multivariate Correlation Analysis of High-Dimensional Data: A New Approach

Area of Science:

• Computational Biology
• Neuroscience
• Genetics

Background:

• High-dimensional data integration is crucial across scientific disciplines.
• Partial Least Squares Correlation (PLSC) is a common method for multimodal data integration.
• Traditional PLSC is computationally intensive for large datasets, such as in genetic neuroimaging.

Purpose of the Study:

• To develop a computationally efficient method for high-dimensional multimodal data integration.
• To address the limitations of traditional Partial Least Squares Correlation (PLSC) in handling large-scale datasets.
• To propose a novel approach combining Random Projection (RP) with PLSC for improved performance.

Main Methods:

• A new method, PLSC-RP, is proposed, integrating Random Projection (RP) for dimensionality reduction into Partial Least Squares Correlation (PLSC).
• The method was tested using simulated and experimental datasets, including whole-genome SNP data (genotypes) and whole-brain neuroimaging data (phenotypes).
• Performance was evaluated based on computational speed and statistical result equivalence compared to traditional PLSC.

Main Results:

• PLSC-RP demonstrates drastically reduced computation time compared to traditional PLSC.
• The method achieves statistically equivalent results to traditional PLSC.
• Random Projection (RP) based dimensionality reduction is shown to be independent of data type.

Conclusions:

• PLSC-RP offers a significantly faster and efficient solution for high-dimensional multimodal data integration.
• The method maintains statistical validity while improving computational performance.
• PLSC-RP is broadly applicable to various integrative analyses combining diverse data sources.