Join GitHub today
GitHub is home to over 28 million developers working together to host and review code, manage projects, and build software together.Sign up
eman2 - initial model
This protocol wraps EMAN2 e2initialmodel.py program. EMAN2  currently applies Monte Carlo algorithm for initial model generation that uses 2D class averages as input. The program is designed to work with low-symmetry structures (with C or D symmetry).
The user must identify as broad a set of different particle views as possible without including projections of other compositional or conformational states of the particle. Input class-averages are then treated as if they were particles used for high resolution refinement. The class-averages are compared to a set of uniformly distributed projections of an initial model, which is used to identify the orientation of each class-average, which is then in turn used to make a new 3-D model. The initial model is produced by low-pass filtering to 2/5 Nyquist and masking to 2/3 of the box size, pure Gaussian noise, independently for each of N different starting models. The goal of this process is to produce random density patterns roughly the same size as the particle. After refining each of these random starting models, the final set of projections is compared against the class-averages as a self-consistency test, and the results are presented to the user in order of agreement between class-averages and projections .
Additional documentation for this program can be found on EMAN2 web site.
You should consider the following when using this protocol (from 2016 Summer Tutorial for EMAN2.12 June 2016):
For most structures, there are a number of ‘local minima’ in the energy space. What that means is, there are a number of incorrect structures which can still appear to agree fairly well (but not as well as the correct structure) with the input data. So, some fraction of the answers you get out are likely to be bad starting models. On the bright side, such bad starting models are usually quite obvious. The severity of this problem varies considerably with the shape of the molecule and the amount of orientation coverage you have. The B-gal structure is reasonably good in this respect, but there are still some wrong answers you may see. Interestingly, particles like ribosomes, generally viewed as 'difficult' have virtually no local minima, and will produce a usable starting model most of the time.
If you are uncertain about the quaternary structure of your molecule, tilt validation is a critical test. You collect pairs of images at 0 degrees and typically 10-20 degrees tilt, box out tilt pairs of particles, then run them through a tilt validation procedure against your final 3-D map.
Heterogeneity is a potential issue. If you have a particle that is highly heterogeneous, the EMAN initial model strategy will produce a model, but it will not be unique (since there isn’t one). Single particle tomography may offer the best solution towards understanding the heterogeneity in your specimen.
Poor angular distribution. If your particles have a single, strongly preferred orientation, especially if this is combined with a low symmetry, mathematically there may not be enough information to produce an unambiguous starting model. However, it is also important to note that in this situation, even if you get a good starting model, refinement will also tend to degrade rather than improve the model. To perform a proper 3-D reconstruction, you must have a reasonable number of particles in orientations covering at least one great circle around the unit sphere.
If you do have a difficult structure, single particle tomography is one good solution. Random Conical Tilt is another possibility
[[]] Tang et al. (2007). EMAN2: an extensible image processing suite for electron microscopy. JSB, 157 (1): 38 - 46.
[[]] Bell J.M. et al. (2016). High resolution single particle refinement in EMAN2.1. Methods, 100: 25 - 34.