Dear All，
I am writing to ask for your advice regarding an issue I am encountering with 3D classification in RELION.
In my dataset, 3D classification consistently produces one dominant class that appears to contain a mixture of multiple structural types. When I take this mixed class and perform further rounds of 3D classification, many of the same structural types that were already separated in the first round can be recovered again. This gives the impression that the initial classification was not able to fully separate the heterogeneity, even though the information seems to be present in the data.
I have tested different values of the regularisation parameter T (including lower and higher values) as well as different numbers of classes (K). However, increasing K or changing T does not fundamentally resolve the issue; the mixed class persists and continues to split into previously observed classes upon further classification.
This makes me wonder whether the limitation is mainly due to factors such as mask definition, residual alignment freedom during classification, or the nature of the heterogeneity itself (e.g. continuous variability or registry shifts), rather than the choice of T or K alone.
I would greatly appreciate your thoughts on what typically causes this behaviour in RELION 3D classification, and what strategies you would recommend to achieve a more complete and stable separation of classes.
Thank you very much for your time and advice.
Best regards,
GKW
Dear All，
I am writing to ask for your advice regarding an issue I am encountering with 3D classification in RELION.
In my dataset, 3D classification consistently produces one dominant class that appears to contain a mixture of multiple structural types. When I take this mixed class and perform further rounds of 3D classification, many of the same structural types that were already separated in the first round can be recovered again. This gives the impression that the initial classification was not able to fully separate the heterogeneity, even though the information seems to be present in the data.
I have tested different values of the regularisation parameter T (including lower and higher values) as well as different numbers of classes (K). However, increasing K or changing T does not fundamentally resolve the issue; the mixed class persists and continues to split into previously observed classes upon further classification.
This makes me wonder whether the limitation is mainly due to factors such as mask definition, residual alignment freedom during classification, or the nature of the heterogeneity itself (e.g. continuous variability or registry shifts), rather than the choice of T or K alone.
I would greatly appreciate your thoughts on what typically causes this behaviour in RELION 3D classification, and what strategies you would recommend to achieve a more complete and stable separation of classes.
Thank you very much for your time and advice.
Best regards,
GKW