AlphaFold Unmasked data sets

Here are deposited all of the predictions generated for the test cases presented in "AlphaFold Unmasked: integration of experiments and predictions with a smarter template mechanism" (doi: https://doi.org/10.1101/2023.09.20.558579) along with the log files necessary to reproduce the experiments.

Each tar.gz file includes one or more AlphaFold experiments, where multiple predictions have been generated either with AlphaFold-Multimer (standard pipeline, v2.2 and/or v2.3 parameters) or with AF_unmasked. An experiment is made of a set of 3D structure predictions (.pdb files) along with the ancillary data generated by AlphaFold (pickle files) and the corresponding inputs (Multiple Sequence Alignments, sequences). Scripts to reproduce the results are included along with the log files generated during the experiments.

H1111, H1142, T1109 and T1110 are multimeric prediction targets from CASP15 (https://predictioncenter.org/casp15/) chosen because most or all predictors failed to correctly predict these complexes in the 2021 edition of CASP.

Rubisco, NF1 and ClpB are examples of large and/or challenging targets where Cryo-EM data is available to be integrated in the prediction pipeline.

The PDB benchmark is made of a set of protein heterodimeric structures deposited in the PDB before January 2022, i.e. before AlphaFold v2.3 was trained and released. These heterodimers have been redundancy reduced by structural similarity (MMalign score threshold: 0.4) to increase their diversity