The Australian ProteinFold service: Interactive prediction and visualisation of protein structure

The advanced developments in AI application to the protein-structure prediction challenge have introduced tools that can predict protein structures efficiently and accurately, such as AlphaFold2 and 3, ColabFold, ESMFold and RoseTTAFold. Such tools are compute-intensive and require powerful GPUs for efficient utilization which makes them not readily available for all researchers. To support Australian researchers interested in utilizing protein structure predictors, the Australian BioCommons in collaboration with the Australian structural biology community is working on establishing a national protein structure prediction service based on the Australian Nextflow Seqera Service. The service greatly simplifies these complex workflows to a ‘bring your own data’ approach and is delivered via an easy-to-use web interface to Australian researchers. Improvements on the ProteinFold workflow (originally available through nf-core) have been carried out to allow the execution of different models including AlphaFold2, Colabfold, and ESMFold on the GADI supercomputer at the National Computational Infrastructure (NCI). We have developed and integrated an interactive visualization into the reporting functionality of Seqera platform, allowing users to investigate and interact with predicted structures through 3D interactive models and plots. We aim to integrate several GPGPU infrastructures from different research institutions to distribute the compute required for such analysis. We also aim to include a search functionality through Foldseek to retrieve known protein structures with a high degree of similarity to the predicted structures. The modifications are planned to be contributed to the nf-core workflow and will be available for researchers to reuse.