Reference Motion correction no trajectory

Hello,

The reference motion correction job finished but I don’t see any trajectories in the final output. It seems that nothing has changed. I used all the default parameters.

image

Am I doing anything in an incorrect way? What did I miss?

Thanks,
Chen

Hi @csun,

You probably aren’t doing anything wrong, this can sometimes happen. Since the per-particle residual motion after patch motion correction is usually small, the hyperparameter set that produces the best trajectory estimates tends to be one where the regularization is fairly “strong”. If you look at this figure from the cryosparc guide, you’ll notice the position of the cross-validation score optimum is very close to zero trajectory activity. In some datasets, either the optimization is too coarse, or the CV score doesn’t provide a clear enough distinction between the zero-trajectory regime and the small-but-accurate trajectory regime.

You could try using the “extensive” mode for the hyperparameter search, or you could eyeball the plots from the hyperparameter search and manually select a set of hyperparameters from earlier in the search where there was some activity, even if the measured CV score doesn’t say those trajectories are the best. I might be able to advise more if you share your “Trajectory activity vs CV score” plot from the last iteration of hyperparameter optimization.

– Harris

1 Like

Hi Harris,

Thank you so much for your reply. I have deleted the previous reference motion correction job since it doesn’t seem to have changed anyting and rerun it with another local refinement job output.
This time some particles have short linear trajectories but some are just random circles. Is it because they have low signal to noise ratio?
image

Below is the Trajectory activity vs CV score" Plot from the last iteraction of hyperparameter optimization. Is it a good curve? Which hyperparameter will be choosen?

image

Thanks,
Chen