Faster Singscore implementation

[ivokwee]

Hi.

We just wanted to inform that we have a faster implementation of Singscore which is 100x faster and much more memory efficient. You can download and install the R package from https://github.com/bigomics/plaid

A preprint is available here https://www.biorxiv.org/content/10.1101/2025.06.14.659661v1

[bhuvad]

Hi Ivo,

Great to see a fast re-implementation of our method! Can I just check which function you used for the benchmarking, simpleScore or multiScore? The former was meant to be a more user friendly function with additional checks and reporting when scoring with 1 gene-set. The latter was developed to deal with the case where 1000s of gene-sets are tested simultaneously, and we removed some of the message/warning reporting to speed-up operations. This is purely out of curiosity from my end. Great work on the new package and good to see computational optimisation on all the awesome scoring algorithms!

Cheers,
Dharmesh

[ivokwee]

Hi Dharmesh.

We used multiScore. Here is the code part we used: https://github.com/bigomics/plaid/blob/3a18befeec7b7ddd1cd3a82f657b417208758d14/experiments/R/functions.R#L166-L180

The wrapper to simulate singscore is plaid::replaid.sing() . See https://github.com/bigomics/plaid/blob/3a18befeec7b7ddd1cd3a82f657b417208758d14/R/plaid.R#L213

Would be great if you could check if we used Sing correctly. Surely there is some overhead creating the gene set collections but mostly it seems to be fast compared to the actual computation.

Ivo