Scalable and Portable Pipelines for Predicting 3D Protein Structures on Standalone and HPC systems

Advances in machine learning techniques are enabling improved protein structure prediction solutions. These solutions include AlphaFold, ESMFold, OmegaFold, and others. Configuring each solution with associated software dependencies and data files is a barrier for many scientists. Singularity containers were developed for AlphaFold, ESMFold, and OmegaFold to enable parallelization of these solutions on high performance computing (HPC) systems. These containers also enable portability to cloud-based platforms. These folding prediction solutions were characterized for performance with a series of human proteins with increasing protein sequence lengths. The current solutions all encounter scaling limitations by protein length due to memory usage. The Singularity containers for AlphaFold, ESMFold, and OmegaFold are provided as open source.