Software

APEX: high performance computing solution for commercial genetic evalution and R&D

---

Abstract

Linear model analysis software is the cornerstone of any genetic evaluation. For commercial deployment it must be versatile, highly performant, easy to master and absolutely fail prove.APEX was designed in 2021 with these goals in mind.Programming: APEX is programmed entirely in C++, the standard language for high-performance computing, to allow for top performance but also to make use of highly optimised third-party libraries and heterogenous computing devices. Template meta-programming techniques and object-orientated programming guarantee extremely fast executing code while maintaining flexibility to accommodate almost instantly for new developments in modelling and algorithms. As a direct result, APEX comes as a single-executable program which covers all tasks which are usually spread over several different programs.Interface: APEX can be deployed in executable or library form, allowing for easy integration into script languages like R and Python. It uses a parameter file in YAML format to communicate all necessary aspects of the task and linear models. YAML is not only standardized, making it possible to use public packages in R or Python for reading, manipulation and writing, it also allows to convey information in a highly structured and logical manner, which is key for ease-of-use. For communicating the linear model itself, APEX uses a linear model syntax which very closely resembles that in scientific publications: "Y=Xb+Zu+e" becomes "Y=Xb+Zu" in APEX, which not only makes the model readable, it also is self-revealing and immediately understood by new users. Algorithms, Task and linear models: APEX covers all algorithms currently used in quantitative genetics: BLUP, AIREML, genomic AIREML, EMREML, multi-chain Gibbs sampling on shared resources, reliability approximation, true reliability calculation, and de-regression. Further APEX can be used as a command line tool for calculation of parameters of "A" or "H", extraction of blocks from "A" or "H", manipulating genotype files, calculating effective record contributions, back-solving of marker effects, calculation of meta-founder co-variance matrices and verifying parentage. APEX accommodates all models commonly used in genetic evaluation including threshold models, random regression models, and all types of single step models including ssSNPBLUP models with marker-specific co-variance matrices. For complex plant breeding data APEX supports auto-regressive co-variance structures(AR1 x AR1).Performance: APEX has been used for ssSNPBLUP models of up to 1.3 billion equations and over 6 million genotyped animals. It's AIREML facility conducted variance component estimation in multi-trait models of up to 17 genetic effects and over 300 parameter to estimate requiring factorization and inversion of sparse matrices with over 25 billion non-zero elements. Since 2022 APEX is actively deployed to animal and plant breeding industry and is regularly used for commercial genetic evaluation and R&D in various species.

Keywords: 2026

How to Cite:

Boerner, V., (2026) “APEX: high performance computing solution for commercial genetic evalution and R&D”, World Congress on Genetics Applied to Livestock Production Digital Archive 2026(1): 2269450. doi: https://doi.org/10.31274/wcgalp.23401

Rights: 1