Multireference analysis implements simultaneous alignment and classification. From the operative point of view, a Multirefence analysis is implemented by a special type of Dynamo multireference projects.
- One of the possible caveats of PCA is that it requires the particles to have been correctly aligned in the first place. If heterogeneity drives alignment to fail, PCA may not be capable to classify the particles.
- PCA is prone to fail to detect features that are present in small subsets of the data.
- PCA is difficult to implement efficiently for large data sets.
Multireference analysis is switched on on the dcp GUI when our project has more than one reference channels (i.e., it is a multireference project), and additionally the swap option is switched on (meaning that particles will be allowed to swap among channels during the iteration).
In any multireference project, the project processes different reference channels during the iteration step. All particles are compared to all available references and each reference channel is archived in its own table.
At each iteration, each particle is compared with every reference. At the end of an iteration, only the particles which best match a given reference are used in the averaging step. Informally, the particle chooses which class it belongs to.