Is singscore usable on single cell data?

[kvittingseerup]

Is singscore usable on single cell data? If so are there specific parameters you recommend?

[bhuvad]

Hi,

We have not yet tested singscore on single-cell data though this is within our immediate plans. We have received great interest in applying singscore to single-cell data, therefore, we are very keen to test it out. At the moment, we do not have any recommendations and would prefer to avoid premature recommendations that are not strongly supported by evidence. We are keen to hear of any ideas/suggestions you may have if you wish to share them.

Cheers,
Dharmesh

[dylanmr]

Just curious if there has been any more thought given to this problem? I have been playing around with this in my single cell data, it is by far the fastest method out there. Thanks for your help!

[bhuvad]

Hi @dylanmr,

We are making progress on benchmarks and should have some more tangible results over the next few months too. Based on our preliminary results, we feel singscore works well for scRNA-seq data. We are currently trying to characterise the properties of scores and trying to see how they compare with other methods. Theoretically, the speed is expected since the computation is much simpler to perform and generally produces similar scores to other methods. It is good to get a measure of this from you (super-excited!) and I am really looking forward to finalising its application on scRNA-seq data!

We have an incredibly talented student (@Malvikakh) working on it so look forward to updates in a few months!

Cheers,
Dharmesh

[ZheFrench]

Hi, any news on this topic ? Would be curious to apply singscore to single cells too.
Congrats for the past work by the way.
Cheers,

[bhuvad]

Hi,

Sorry for the delay with responding to you and for the delay with our single-cell work. We have been able to perform preliminary benchmarks on scRNA-seq data from various technologies and our results thus far show that it is valid to use singscore on scRNA-seq data as is. We do recommend the standard filtering and QC required for scRNA-seq data prior to running singscore. We are still in the process of finalising the benchmark and will soon publish our results (the pandemic has slowed things down).

Our current results suggest that the zeros in scRNA-seq data are very powerful in distinguishing cells with specific phenotypes and singscore can harness this information. Apt QC and filtering are therefore important and standard practices such as those listed in the OSCA bioconductor book and the seurat vignettes suffice for a singscore analysis. For instance, T-cell specific genes being unexpressed in other immune cells provide a strong discriminating power. This aligns with previous findings in the field, that unexpressed genes provide great cell type discriminating power.

We are also exploring potential improvements on the idea of gene-set testing in scRNA-seq data; particularly, we are exploring alternatives to individual cell level analyses as we expect phenotypes to mostly manifest in a small (potentially rare) subset of cells rather than in a single cell out of thousands of cells. Our benchmarking framework is set up to allow such questions to be explored so we hope to report this as soon as possible (factoring in the pandemic of course!).

We appreciate your positive feedback regarding our package and hope to get the next version ready as soon as possible!

Cheers,
Dharmesh