Molecular Dynamics
RELION v5.0 Beta Now Supported in Exxact Systems
November 2, 2023
4 min read
What’s new with RELION v5.0 Beta?
As a long time, life science solutions provider for cryo-EM applications, Exxact offers the new RELION v5.0 Beta release in our Relion for Cryo-EM Systems. Below is a brief summary of what's new in RELION v5.0 Beta. For the full documentation of the release please refer to the full release notes located at:
Blush Regularization
Dari Kimanius has developed a new method to incorporate more prior knowledge into the cryo-EM refinement process than the one typically used (which merely assumes smoothness in real-space, or limited power in Fourier-space). This method is called Blush regularisation and it uses a denoising convolutional neural network inside the iterative refinement algorithm of Class3D, Refine3D or MultiBody jobs. The effects of this are largest when the signal is weak and standard refinement in RELION would overfit (as for example visible from streaky artefacts in the solvent region). Using Blush reglarisation, Dari successfully refined a data set of a 40 kDa protein: RNA complex to 2.5A. The same data set was intractable in standard RELION or CryoSPARC.
DynaMight - Model Continuous Structural Heterogeneity
Johannes Schwab developed a method called DynaMight that ‘explores protein Dyna-mics, and Might improve your map’. It is based on a variational auto-encoder for modelling continuous structural heterogeneity that predicts 3D deformations of a Gaussian model for the consensus map, and a deformed backprojection algorithm that attempts to “un-do” these deformations to reconstruct an improved consensus map.
ModelAngelo - Automated Atomic Model Building
Kiarash Jamali developed a machine-learning approach for automated atomic model building and identification of unknown proteins in cryo-EM maps. ModelAngelo will build most of your automatically, provided the resolution extends beyond 3.5-4.0 Angstroms. Goodbye to months in the dark graphics room!
Select Subsets of Filaments Using Dendrograms
David Li developed a useful utility to select subsets of filament particles that belong to the same structural class. It has been implemented on the Helix tab of the Subset selection job type.
Support for AMD and Intel GPUs (HIP/ROCm and SYCL)
Suyash Tandon from AMD and Jason Do from Intel, together with their colleagues, have contributed code for GPU acceleration of Relion v5.0 in HIP/ROCm and SYCL, respectively. This means that the relion_refine program can now also be run efficiently on AMD and Intel GPUs. (Previously, Topaz, Blush, class_ranker, DynaMight and ModelAngelo only worked on existing CUDA implementation. Vectorized CPU-acceleration still work too.)
Subtomo-gram Averaging Pipeline
Alister Burt, Euan Pyle, Sjors Scheres and others have developed a new pipeline for sub-tomogram averaging that starts with serialEM mdoc files and raw movies, and potentially ends with automated model building by ModelAngelo. You can access it by launching relion --tomo from the command line. However, please do note that this part of the code is not yet well tested and have not yet been able to write an explanatory tutorial for this, so please be patient. Until the RELION team has finished the documentation and testing, you can still access and experiment with the code, but they cannot yet provide any feedback…
Closing Remarks
RELION v5.0 remains proudly open-source and is completely free for all users. Download the code from their GitHub.
Read its new documentation here!
For those of you who have installed RELION before, a lot has stayed the same, but please note the additional step of making a conda environment, prior to compiling RELION.
Molecular Dynamics
RELION v5.0 Beta Now Supported in Exxact Systems
November 2, 20234 min read
What’s new with RELION v5.0 Beta?
As a long time, life science solutions provider for cryo-EM applications, Exxact offers the new RELION v5.0 Beta release in our Relion for Cryo-EM Systems. Below is a brief summary of what's new in RELION v5.0 Beta. For the full documentation of the release please refer to the full release notes located at:
Blush Regularization
Dari Kimanius has developed a new method to incorporate more prior knowledge into the cryo-EM refinement process than the one typically used (which merely assumes smoothness in real-space, or limited power in Fourier-space). This method is called Blush regularisation and it uses a denoising convolutional neural network inside the iterative refinement algorithm of Class3D, Refine3D or MultiBody jobs. The effects of this are largest when the signal is weak and standard refinement in RELION would overfit (as for example visible from streaky artefacts in the solvent region). Using Blush reglarisation, Dari successfully refined a data set of a 40 kDa protein: RNA complex to 2.5A. The same data set was intractable in standard RELION or CryoSPARC.
DynaMight - Model Continuous Structural Heterogeneity
Johannes Schwab developed a method called DynaMight that ‘explores protein Dyna-mics, and Might improve your map’. It is based on a variational auto-encoder for modelling continuous structural heterogeneity that predicts 3D deformations of a Gaussian model for the consensus map, and a deformed backprojection algorithm that attempts to “un-do” these deformations to reconstruct an improved consensus map.
ModelAngelo - Automated Atomic Model Building
Kiarash Jamali developed a machine-learning approach for automated atomic model building and identification of unknown proteins in cryo-EM maps. ModelAngelo will build most of your automatically, provided the resolution extends beyond 3.5-4.0 Angstroms. Goodbye to months in the dark graphics room!
Select Subsets of Filaments Using Dendrograms
David Li developed a useful utility to select subsets of filament particles that belong to the same structural class. It has been implemented on the Helix tab of the Subset selection job type.
Support for AMD and Intel GPUs (HIP/ROCm and SYCL)
Suyash Tandon from AMD and Jason Do from Intel, together with their colleagues, have contributed code for GPU acceleration of Relion v5.0 in HIP/ROCm and SYCL, respectively. This means that the relion_refine program can now also be run efficiently on AMD and Intel GPUs. (Previously, Topaz, Blush, class_ranker, DynaMight and ModelAngelo only worked on existing CUDA implementation. Vectorized CPU-acceleration still work too.)
Subtomo-gram Averaging Pipeline
Alister Burt, Euan Pyle, Sjors Scheres and others have developed a new pipeline for sub-tomogram averaging that starts with serialEM mdoc files and raw movies, and potentially ends with automated model building by ModelAngelo. You can access it by launching relion --tomo from the command line. However, please do note that this part of the code is not yet well tested and have not yet been able to write an explanatory tutorial for this, so please be patient. Until the RELION team has finished the documentation and testing, you can still access and experiment with the code, but they cannot yet provide any feedback…
Closing Remarks
RELION v5.0 remains proudly open-source and is completely free for all users. Download the code from their GitHub.
Read its new documentation here!
For those of you who have installed RELION before, a lot has stayed the same, but please note the additional step of making a conda environment, prior to compiling RELION.
