SLEMM

SLEMM applies the stochastic Lanczos algorithm within a mixed models framework to perform efficient genomic prediction and restricted maximum likelihood (REML) estimation on large-scale genotypic and phenotypic datasets.

Key Features:

• Stochastic Lanczos implementation: Uses the stochastic Lanczos algorithm within a mixed models framework to accelerate restricted maximum likelihood (REML) estimation.
• Efficient computational performance: Optimized for high-performance computing to handle datasets involving millions of individuals and SNPs.
• SNP weighting mechanism: Incorporates SNP weighting to assign differential weights to single nucleotide polymorphisms (SNPs) based on relevance to the trait.
• Versatile validation: Tested across seven public datasets covering 19 polygenic traits in three plant species and three livestock species.
• Comparative benchmarking: Demonstrated predictive accuracy and computational efficiency relative to GCTA's empirical BLUP, BayesR, KAML, and LDAK's BOLT.
• Scalability: Simulation analyses showed capability to process up to 3 million individuals and 1 million SNPs.

Scientific Applications:

• Plant breeding: Genomic prediction and marker identification for polygenic traits to inform selection decisions.
• Livestock genetics: Estimation of breeding values and genomic prediction across livestock species.
• Personalized medicine: High-throughput genomic prediction for complex traits and potential risk stratification in human genetics.
• Polygenic trait analysis: Predictive modeling and evaluation of polygenic architectures using large-scale genotype and phenotype data.

Methodology:

Implementation of the stochastic Lanczos algorithm within a mixed models framework for REML estimation, incorporation of SNP weighting, and evaluation via simulation analyses and comparative benchmarking.