Melbournevirus protein structure prediction - AlphaFold3

AlphaFold3 structure prediction from Melbournevirus proteins in monomeric and different homo-multimeric states, as well as heterodimers detected in the cross-linking experiment. AlphaFold version 3.0.1 was installed and run locally on Ubuntu 24.04.1 LTS by utilizing the Docker platform. Predictions were run on a NVIDIA RTX A6000 GPU (48 GB RAM). Number of recycles was set to default. We used the initial AlphaFold3 input format, the default parameters as well as databases supplied at https://github.com/google-deepmind/alphafold3. Model parameters were received directly from Google. Number of multimer predictions per model was set to 1. The heteromultimeric model of MEL_149 and MEL_368 has been predicted with the AlphaFold3 online search server. pTM and ipTM scores were extracted from .json files using R and PAE data was extracted using python 3. For every monomer or multimer prediction the model with the best agreement with cross-linking data was selected. In case of equally satisfied cross-links, the model with highest pTM, or in case of multimeric predictions ipTM score were selected for analysis.