I am very new in the single particle area. I recently working on a SPA dataset, and I want to use Topaz to optimize the particle selection. My particle is like a ring, I tried to use a blob picker and several rounds template picking, however, I don’t think I have got all particles, and many particles are off centered. While I am using topaz on 500 micrographs, I did not get a good curve in the precision, and I have tried several rounds, it stayed at low precision range (below 0.1). I am wondering what parameters I can tune and any suggestings tips will be super helpful!
My protein complex diameter: 190A in diameter, 140A in height. It has a preferred orientation (top and bottom views), and have some side views.
The 2D template using attached below and Topaz parameters I also attached along with the 2 rounds of training.
@alexjamesnoble might have some ideas?
Average-Precision values are relative numbers, meaning that you shouldn’t worry about what the number is, just that the curve generally increases monotonically. Have you tried extracting particles?
A couple things to note: 1) The expected number of particles is an important parameter that you should try to optimize. You can use cross-validation to optimize it. 2) The average-precision hasn’t plateaued yet. Try increasing the number of epochs to get a better model. Also, you can turn on pretrained initialization to decrease the training time with minimal affect on model performance.
Thanks Vamsee! abd Alex! SOrry for the delay in reply.
I have tried extract using the model I got, and I doubled size of my particles, and I got more different views other than just top/bottom-like views.
I will try to change parameters as you suggested.