I am doing Ab-initio with 2 classes and I found out when I exclude particles of the second class and do refinement of the larger class, I get a better resolution after refinement (resolution of 6 A) in comparison to when I do Ab-initio with one class and include all of those particles (resolution 8 A). However, when I run 2D classifications on the excluded particles, they are still my proteins. This causes a drastic decrease in the number of particles but resolution is better. Should I include those particles? Can anyone help me with this?
If you only include the 2D classes that look like your protein from the worse class, does your resolution still decrease?
If you haven’t tried it already, I generally notice I get better results (less misclassified particles; higher downstream resolution) if I throw all of the particles from an ab-initio run into a heterogeneous refinement, even if it doesn’t change the number of particles per class by much. What does a 3d refinement of the “bad class” particles look like? Does it seem like it’s a different conformation/ different oligomeric state etc.? If so you should refine them separately.
Like posertinlab said, you could also try to get rid of any junk particles from the bad 3d-class using 2d, and reintroduce the remaining particles.
Yes, always do this.
Also it’s a useful to ponder a question - can an estimate of the mean ever be improved by discarding samples?
Thank you! But, heterogeneous refinement needs 2 initial volumes as inputs. One of them is my Ab-initio, but what should I use for the next volume?
Thank you! What I have got is that heterogeneous refinement needs 2 initial volumes as inputs. One of them is my Ab-initio, but what should I use for the next volume?
I suggest using 1-2 copies of the good initial reference and then 2-4 additional bad references either from completed ab initio runs with multiple classes, or from an ab initio run which is killed and marked complete soon after the initial random volumes appear.