SNF Schemes

Download a copy of the vignette to follow along here: snf_schemes.Rmd

This vignette describes how individual input dataframes are converted into a final fused network in different ways using the snf_scheme parameter. Ultimately, variation across schemes gives the user access to a broader (but still sensible) space of similarity networks and clustering solutions.

There are currently 3 available schemes.

(1) “Individual”

snf_scheme = 1:

Each individual input dataframe provided in the data list is combined into a final network by SNF directly. This is identical to treating your data list like the WALL parameter in the original SNF function.

The limitation (or strength) of this approach is that the integration is biased towards information sources (domains) that have the most representation within the list. For example, if you happened to have 5 demographic dataframes but 10 neuroimaging dataframes, you will get an integration that is biased towards neuroimaging data.

(2) “Two-step”

snf_scheme = 2:

Input dataframes are combined within user-specified data domains by one round of SNF and then combined across domains by a second round of SNF.

This scheme was intended to reduce some of the weighting issues in the individual approach. The final integration is done by combining a single demographic similarity matrix with a single neuroimaging similarity matrix, resulting in a balanced integration at the domain level.

There is no “three-step” or greater SNF scheme accounting for dataframe subdomains or subsubdomains at this time, but ultimately this approach should serve well to increase the space of clustering solutions you have access to.

(3) “Domain”

snf_scheme = 3:

Input dataframes are combined within domains by concatenation and then combined into a single fused network by SNF. This in some way is also quite similar to this spirit of the original SNF examples, but where each initial distance matrix is done an entire domain at a time. This approach can lead to the greatest information loss during the distance matrix calculation step, but still offers an additional way to generate distinct cluster solutions.

Custom SNF schemes

There is no simple interface to add custom SNF schemes at this time. If there is a different scheme you would like to use, feel free to submit a suggestion in the package’s issues page or to try poking around with the snf scheme source code yourself!