The VSNi Team18 October 2023
As the developers of ASReml, we take immense pride in presenting a statistical software package that has become a symbol of robustness and accuracy. Here at VSNi, our mission is to empower researchers across the globe with a tool capable of handling complex and diverse datasets while maintaining the highest standards of statistical and computational integrity. In this blog, we will reveal what is behind ASReml's exceptional robustness and explain why it has become an indispensable tool for statisticians, breeders, and researchers alike.
Powerful estimation techniques
We firmly believe that robust statistical theory is the foundation of any reliable analysis. ASReml uses the REML (Restricted Maximum Likelihood) estimation method. This approach allows ASReml to tackle unbalanced data, missing values, and intricate variance structures with ease. By considering data imbalances, REML provides unbiased and efficient estimates of model parameters, making ASReml remarkably stable and reliable even in challenging data scenarios.
Flexibility in model specification
Paramount to all researchers across diverse fields is the requirement for flexibility in their data models. To meet this need, ASReml is engineered to handle a variety of statistical models, including hierarchical, multi-variate, spatial and genomic models, to name a few. ASReml enables breeders to fit complex and flexible models, allowing for the incorporation of unique sets of variance components for each environment/location, and easily integrating pedigree and molecular data. This adaptability allows researchers to explore different research questions and to combine an array of datasets, making ASReml the go-to choice for quantitative geneticists and scientists.
Handling of complex genetic and environmental models
ASReml excels in modelling genetic and environmental effects, allowing researchers to account for genetic relationships, genomic data, and multiple sources of variation. This capability makes ASReml an invaluable tool for calculating heritability, determining the genetic architecture of one or more traits, and understanding breeding responses to selection. ASReml smoothly integrates genomic data into the analysis, facilitating the examination of genetic markers' influence on phenotypic traits. By incorporating genomic information, researchers can conduct genome-wide association studies (GWAS) and genomic selection (GS with GBLUP), allowing the identification of key markers or to build predictive models; all critical tools in modern breeding.
Efficient computation and scalability
ASReml has been developed to handle large datasets efficiently. The underlying algorithms and computational techniques optimise the estimation process, making it less computationally intensive compared to other statistical approaches. ASReml users can analyse vast datasets in a reasonable timeframe, without compromising on accuracy or precision. ASReml-R 4.2 is our fastest version yet! The implementation of optimised and threaded mathematical routines has led to remarkable speed gains. ASReml-R now performs genomic prediction up to 85 times faster, and complex multi-trait and multi-environment analyses are up to 40 times quicker!
Robustness of model assumptions
While models often rely on certain statistical assumptions, we strive to make ASReml flexible enough to accommodate these assumptions. Our software can consider non-Normal data distributions, heteroscedasticity, and correlations among residuals, providing reliable results. This feature is particularly advantageous when dealing with real-world data, which is often subject to additional complexities.
Modern algorithms: The pillars of ASReml's accuracy
ASReml's robustness can be attributed to the sophisticated algorithms at its core. The software employs advanced numerical techniques to handle intricate statistical models effectively. These algorithms are designed to maintain numeric precision but still provide reliable results, even when faced with large datasets, ensuring that ASReml remains stable and dependable under various data scenarios. Furthermore, ASReml uses sparse matrix methods to optimise computations, significantly reducing memory usage and enhancing overall efficiency.
The reliability of any statistical software hinges on rigorous testing. ASReml undergoes extensive testing to validate its procedures, identify potential issues, and ensure optimal and consistent performance. Our team of experts continuously evaluates the software under diverse simulated and real-world datasets, mimicking a wide range of conditions. This dedication to quality assurance ensures that ASReml remains a robust and trustworthy tool, consistently delivering precise and reproducible outcomes.
ASReml: A reliable and versatile statistical tool for researchers
In conclusion, ASReml's robustness is the result of our relentless pursuit of statistical and computational excellence. ASReml uses theoretical-based estimation techniques, allows for model flexibility, handles heterogeneous variance structures, accommodates diverse genetic and environmental models, ensures efficient computation, and stays robust to an array of model assumptions.
Dr. Ruth Butler15 November 2023
Experimental Design: What are Balance, Efficiency and Independence?
Dr. Salvador A. Gezan08 November 2023
Statistical inference is the process of drawing some conclusions about a population based on the sample data at hand.
Salvador A. Gezan01 November 2023
An important aim when fitting linear mixed models (LMM) is the use of random effect estimates. In some analyses, such as genetic evaluations, the main objective of the analysis is to obtain these estimates.