Improved protein complex prediction with AlphaFold-multimer by denoising the MSA profile

Supporting data for AFProfile

casp15.tar.zst - predicted structures and MSAs for the CASP15 set
native_afm_2_6_bench.tar.zst - native cif files for all complexes with ranking confidence <0.75 in the AFM 2-6 chains benchmark ( https://doi.org/10.1093/bioinformatics/btad424)
pred_top_models_afm_2_6_bench.tar.zst - predicted top ranked models and scores for all 100 samples.
afm_opt_metrics.csv - the best models, confidences and MMscores for the AFProfile run on the AFM 2-6 chains benchmark (n=427 structures)
msa_shapes.csv - the shape of the MSA as input to AFM for each structure
directed.tar.zst - contains all models for the AFProfile run on the AFM 2-6 chains benchmark (n=42700 samples)

The directories are compressed with zstd: https://github.com/facebook/zstd
Uncompress:
tar --use-compress-program full/path/to/zstd -xvf file.tar.zst