Calculate distances between diffusion profiles

Distances between two sets of stored profiles, e.g. drugs against diseases. Lower means more similar. Ruiz et al. rank drug-disease pairs by the correlation distance on the multiscale interactome and by Canberra when comparing against gene expression signatures.

Usage

calculate_profile_distances(
  object,
  from,
  to,
  metric = c("correlation", "canberra", "l1", "l2", "cosine")
)

Arguments

object

A DiffusionProfiles object with generated profiles.

from

Character vector. Seeds for the rows of the result.

to

Character vector. Seeds for the columns of the result.

metric

Character. One of "correlation", "canberra", "l1", "l2" or "cosine". Defaults to "correlation".

Value

A numeric matrix of length(from) x length(to) distances.

Details

Every profile sums to 1 and the seed keeps roughly 1 - alpha of the mass on itself. What each metric weights:

• "correlation": Pearson on the full vector. Driven by the large entries, i.e. the immediate neighbourhood of the seed and hubs.

• "cosine": Without centring. Since all profiles sum to 1 and are mostly near zero, it ranks almost identically to "correlation".

• "l1": Twice the total variation distance between the two distributions. Less dominated by the top entries than the above.

• "l2": Dominated by the largest entries; favours big, well-connected nodes.

• "canberra": Relative differences per node, so the many near-zero entries count as much as the large ones. Useful when the profile is compared against a signature on a different scale.

References

Ruiz, Zitnik and Leskovec, Identification of disease treatment mechanisms through the multiscale interactome, Nat Commun 2021.