MR with model splitting using Slice-n-Dice¶
Slice-n-Dice is an automated MR pipeline designed to pre-process AlphaFold2 and RosettaFold models by removing low confidence regions and converting confidence scores into predicted B-factors. It also slices predicted models into distinct structural units then automatically places the slices using Phaser. The slicing step can use AlphaFold2’s predicted aligned error (PAE), or can operate via a variety of Cɑ atom clustering algorithms, extending applicability to structures of any origin.
Slice-n-Dice helps to deal with inaccuracy in domain-domain orientations of predicted models.
Structure prediction software puts pLLDT scores (AlphaFold) or rmsd estimates (RoseTTAFold) in place of B-factors. This gives a wrong perception regarding the regions with low or high reliability, and, therefore, the scores must be recalculated as B-factors. That’s why important to select B-factor treatment option correctly. The default pLDDT threshold for AlphaFold2 models is 70 and the default RMS threshold for RosettaFold models is 1.75.
Slice¶
Slice option in CCP4i2 uses Birch clustering algorithm based on the coordinates of the Cɑ atoms.
The number of splits (clusters) can be specified in the Options tab. MIN and MAX splits are the minimum and the maximum number of splits to make to the search models. For example, a MIN split of 1 and a MAX split of 3 will result in 3 MR jobs:
using an unsplit model,
a model split in two and
a model split three ways.
Note
The large target protein needs to be modeled in several chunks. An increasing number of splits allows slice and dice to test a range of different splits.
Dice¶
Dice method performs molecular replacement on the individual slices produced by Slice by Phaser. Phaser automatically places as many slices as possible. A number of cores for Phaser can be specified in the Options tab. After that Refmac is used to assess if the placed slice has improved the solution. A number of refinement cycles in Refmac can be specified in the Options tab.
ACKNOWLEDGEMENTS
This article uses materials kindly provided by Dr. Adam Simpkin and Dr. Ronan Keegan, whose help is greatly appreciated.
References