A diffusion model of viral evolution predicts mutation fitness and evolutionary trajectories

- Repository for the Paper A diffusion model of viral evolution predicts mutation fitness and evolutionary trajectories - 

1 Code source: https://github.com/wuaipinglab/VES

2 Data source:

2.1 Influenza virus (hemagglutinin protein)

• file name: aligned_single_processed_HA.a2m
• collected from GISAID (up to May, 2024)
• quality control and 100% redundancy removal
• multiple sequence alignment using MAFFT, selecting the Influenza A virus (A/WSN/1933(H1N1)) strain as the wild type reference sequence
• a total of 95,560 sequences

2.2 SARS-CoV-2 virus (spike protein)

• file name: nr_no_X_1273_spike.a2m
• collected from GISAID (up to June, 2025)
• quality control and 100% redundancy removal
• multiple sequence alignment using MAFFT, selecting the hCoV-19/Wuhan/WIV04/2019 (WIV04) strain as the wild type reference sequence
• a total of 55,246 sequences 

3 Model checkpoints:

3.1 VES official implementation

• file name: FLU_DM_UNet_Attention_128_ch.pth (Influenza virus)
• file name: COVID_DM_UNet_Attention_4_conv_data_update.pth (SARS-CoV-2 virus)