A High-Performance Computing Implementation of Iterative Random Forest for the Creation of Predictive Expression...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Biological datasets are rapidly growing, necessitating advanced computational methods.
• Existing tools may not scale efficiently for large-scale genomic analyses.

Purpose of the Study:

• To present a high-performance computing (HPC)-capable implementation of Iterative Random Forest (iRF).
• To introduce a new method, iRF Leave One Out Prediction (iRF-LOOP), for creating large Predictive Expression Networks.
• To enable explainable AI eQTL analysis on massive SNP sets.

Main Methods:

• Developed an HPC-capable implementation of Iterative Random Forest (iRF).
• Introduced iRF Leave One Out Prediction (iRF-LOOP) for Predictive Expression Network construction.
• Benchmarked iRF performance against the R version on supercomputers Summit and Titan.

Main Results:

• The new iRF implementation significantly accelerates analysis of large SNP sets (over 1 million SNPs).
• iRF-LOOP successfully created Predictive Expression Networks for over 40,000 genes.
• Demonstrated iRF-LOOP's capability to identify biologically significant results.

Conclusions:

• The new HPC-capable iRF implementation allows for unprecedented scale in biological data analysis.
• iRF-LOOP provides a powerful tool for constructing complex Predictive Expression Networks.
• This advancement facilitates deeper insights into complex biological systems through large-scale genomic analysis.