Closed JatGreer closed 4 months ago
Assigning to myself. Plan of study is to produce 100 micrographs (1000x1000) image each with no particle(s) present in order to study how ice affects the thon rings. Two datasets will be produced, one via the parakeet GRF model, one via the parakeet slow ice (atomic simulation) model. Then run ctffind4 on both and determine whether thon rings extend to nyquist in both cases. Essentially we want to identify if there are any bugs/improvements which could be made in Parakeet or roodmus which result in images with this super-high SNR ratio. James Parkhurst suggests that the high SNR may even just be a symptom of lack of movement of particles generated via roodmus. The slow ice method may be VERY slow, so need to run a pilot study on how long 1 image takes to generate for for those parameters. Need to evaluate whether to continue on this path for paper 1 or leave as a possible subject for paper 2. Perhaps future studies could add Gaussian noise to the ice micrographs to quantise how this affects the thon rings - but I'm not sure of whether this is actually useful from a theoretical point of view yet - please advise!
Update on this - artefacts were found in atomistic noise ugraphs (in contact w JParakhurst) and grf implementation has not been found in Parakeet src code. Next steps are running through multem tutorial to get a fuller understadning of how noise and simulation are implemented within Parakeet/multem before asking specifc questions to JP and doign follow up study
I think we realised recently that it is not so much the noise level we should be concerned with when it comes to seeing Thon rings at high spatial frequency, but rather the envelope of the CTF, which is determined by the radiation damage model applied. This radiation damage model however is only correctly applied if we simulate fractionated images, which we're planning for the next version
Considering this "solved" as we have fewer problems with high resolution information in our "movie" datasets which include radiation damage (which it was realised we did not include in our "micrograph" datasets
Issue 35 from gitlab simulate a small data set (~100 micrographs of 1000x1000) without molecules in them and check if the Thon rings extend to Nyquist to judge if the noise model of Parakeet is correct.