#sc-velocity

2020-09-05

Kevin Rue-Albrecht (07:09:57): > @Kevin Rue-Albrecht has joined the channel

Kevin Rue-Albrecht (07:09:57): > set the channel description: Discussions about the analysis of expression dynamics in single cell RNA seq data

Aaron Lun (07:10:07): > @Aaron Lun has joined the channel

2020-09-14

Aaron Lun (18:33:47): > PAARRTYY

Aaron Lun (18:34:11): > ooh wait, the partey started last week.

Aaron Lun (18:35:13): - File (PNG): image.png

2020-09-17

Charlotte Soneson (04:49:40): > @Charlotte Soneson has joined the channel

2020-09-28

Kevin Rue-Albrecht (11:23:44): > tryingautoscale=Falsenow

Kevin Rue-Albrecht (11:32:50): > looks good! thanks!

Charlotte Soneson (11:34:25): > :tada:

Kevin Rue-Albrecht (11:55:01): > I’m still intrigued by the warning > > Warning: file 'velociraptor/configure' did not have execute permissions: corrected > > But on my Mac it says: > > -rwxr-xr-x 1 kevin staff 71B 7 Jun 08:06 configure > -rwxr-xr-x 1 kevin staff 87B 7 Jun 08:06 configure.win >

Kevin Rue-Albrecht (11:55:32): > (figuring out how to check file permission on Windows)

Kevin Rue-Albrecht (11:59:12): > looks executable there too

2020-10-04

Kevin Rue-Albrecht (09:32:32): > https://github.com/Bioconductor/Contributions/issues/1632#issuecomment-703177829woohoo:party_parrot::bioc-load:

2020-10-10

Aaron Lun (13:07:06): > if you subset to the HVGs to be used withsubset.row=inscvelo(), you shouldn’t need to subset the cells.

Aaron Lun (13:22:53): > oh wait, we do that already.

Aaron Lun (13:23:37): > vignette should say “we assume that quality control, normalization and feature selection have already been eprformed”

Aaron Lun (13:38:45): > @Charlotte Sonesoncan you cast your eye overhttps://github.com/Bioconductor/OrchestratingSingleCellAnalysis/pull/44

Aaron Lun (13:39:03): > mostly the text in that section

Charlotte Soneson (13:58:24): > yes, will do

Aaron Lun (14:02:07): > Thanks, You might want to just pull the branch and look at the entire section, the PR doesn’t actually cover all of it

2020-10-11

Charlotte Soneson (08:34:27): > Ok, checked and PR’ed to thevelocitybranch

Aaron Lun (13:49:54): > ooops, i forgot to push my latest changes before you looked at it

Aaron Lun (13:50:15): > well, I just merged your PR, but you can give a final look at my PR before I merge it

Aaron Lun (13:54:54): > @Kevin Rue-Albrechtalso have some changes in velociraptor

Aaron Lun (14:22:16): > Anyway, I can’t say I really understand the difference between the velocity pseudotime and the latent time.

Aaron Lun (14:28:08): > reading the corresponding section of the biorxiv ms is completely cryptic to me.

Kevin Rue-Albrecht (14:29:09): > i pushed velociraptor (and iSEE) upstream

2020-10-12

Aaron Lun (20:10:54): > I’m going to assume that the chapter was all good.

2020-10-13

Charlotte Soneson (02:35:12): > Oops, sorry, thought I had replied; yes, looked good.

2020-11-19

David Dittmar (08:32:55): > @David Dittmar has joined the channel

Aaron Lun (19:29:56): > I’m still a little nervous about how we’re plugging into scvelo, e.g., feeding it normalized expression values and precomputed PCs. Can someone reassure me that we’re doing the right thing? It would be nice if we could show that we get “close enough” results to the same analysis run purely with the Python sequence of calls.

2020-11-20

Charlotte Soneson (03:14:40) (in thread): > I’ll look into this and make a direct comparison.

Aaron Lun (03:26:55) (in thread): > thanks. If you come up with a few succinct scripts, we can stick them intoinst/so that we can re-run them in future builds.

2020-11-23

Charlotte Soneson (13:38:54) (in thread): > A bit bare-bones, but something like this? - File (HTML): compare_velociraptor_runs.html

Aaron Lun (13:40:15) (in thread): > oh yeah

Aaron Lun (13:40:38) (in thread): > probably throw down some silent stopifnots so that I pay attention when it breaks

Charlotte Soneson (13:44:38) (in thread): > Right. Format would be ok like this - just an Rmd that runs stuff and makes some plots?

Charlotte Soneson (13:51:32) (in thread): > And is there a safe way to get the path to the environment thatvelociraptoris using? Right now I’m providing it manually toreticulate::use_virtualenv()(in a hidden chunk) so that I can compare to running the commands directly in python chunks. Otherwise I guess I’ll just leave those comparisons out, I just wanted to make sure we get the same things with the wrapper.

Aaron Lun (14:11:05) (in thread): > Yeah, Rmarkdown is fine.

Aaron Lun (14:11:38) (in thread): > Path to environment - should be possible with something frombasilisk.utils

Aaron Lun (14:13:57) (in thread): > Well, not pretty but it should work robustly: > > basilisk:::.obtainEnvironmentPath(velociraptor:::velo.env) >

Aaron Lun (14:14:31) (in thread): > Maybe I should export that.

Charlotte Soneson (14:15:33) (in thread): > It’s doing the right thing in this case at least:slightly_smiling_face:

2020-11-24

Charlotte Soneson (03:00:03) (in thread): > Ok, PR here:https://github.com/kevinrue/velociraptor/pull/27. I tried just swapping the HVGs from scran to those from scVelo, and the results were even more similar, so I guess that’s where the biggest difference comes from.

Aaron Lun (03:07:49) (in thread): > great, thanks. will try to look at it in a few dayas

2020-12-01

Charlotte Soneson (03:43:05): > Any idea what may be going on here (it’s been failing for several builds):http://bioconductor.org/checkResults/3.13/bioc-LATEST/velociraptor/riesling1-buildsrc.html?zellkonverterandbasiliskare fine here, so it doesn’t seem to be a general conda issue.

Aaron Lun (11:17:20): > it’s probably a sporadic windows error, where the conda environment installation fails but is still cached (so it won’t clear up in subsequent builds). I protect the main conda installation from mysterious windows install failures but it occurs to me that the same is not done for conda environments. It might help to bump the version.

2020-12-05

Alexander Toenges (17:34:42): > @Alexander Toenges has joined the channel

2020-12-12

Huipeng Li (00:38:25): > @Huipeng Li has joined the channel

2020-12-13

Kelly Eckenrode (13:42:11): > @Kelly Eckenrode has joined the channel

2021-01-01

Bernd (14:06:44): > @Bernd has joined the channel

2021-01-22

Annajiat Alim Rasel (15:45:28): > @Annajiat Alim Rasel has joined the channel

2021-02-12

Alexander Toenges (14:56:40): > https://twitter.com/volkerbergen/status/1360309761628139525?s=21 - Attachment (twitter): Attachment > scVelo goes multi-core, with another 10fold speed up on my instance. Thanks M Klein, P Weiler, Y Schaelte for this terrific work! http://scvelo.org

Aaron Lun (15:03:38): > hmm. Troublesome.@Kevin Rue-Albrechtwe’ll have to make sure we clamp down on unintended parallel processing.

Kevin Rue-Albrecht (15:07:14) (in thread): > I may be able to find some time next week to look at things. Got a week gap in my teaching.

Aaron Lun (15:07:55) (in thread): > note, this is for the next version, so it’s not entirely relevant to the current version.

Aaron Lun (15:08:33): > Need to avoid noisy neighbor issues when packages parallelize by default. Hopefully the default is still to use a single core.

Kevin Rue-Albrecht (15:18:38) (in thread): > ah right

2021-02-13

Aaron Lun (19:16:34) (in thread): > you should probably get Mike Stadler’s PR in.

2021-02-19

Aaron Lun (00:14:26): > God, that was cray cray

Aaron Lun (00:14:30): > https://community-bioc.slack.com/archives/CEQ04GKEC/p1613262381158600?thread_ts=1612926724.147400&cid=CEQ04GKEC - Attachment: Attachment > @Hervé Pagès was able to reproduce the timeout, and apparently the fix is: > > export KMP_DEVICE_THREAD_LIMIT=2 > export KMP_TEAMS_THREAD_LIMIT=2 > I have ABSOLUTELY NO IDEA why this works.

2021-02-22

Aaron Lun (19:44:52): > @Kevin Rue-Albrechtwhy is the docker push at the same location as the build?

Aaron Lun (19:45:06): > Shouldn’t it be a scheduled job so as to ensure dependencies are always up to date?

2021-02-23

Kevin Rue-Albrecht (11:39:56): > the build isn’t working anyway, as the dockerhub instance runs out of memoryhttps://hub.docker.com/repository/registry-1.docker.io/kevinrue/velociraptor/builds/475f52eb-d5b1-4755-aeac-c8a99f03036ealso, GHA runs intohttps://github.com/kevinrue/velociraptor/runs/1960479532#step:28:318 > > one node produced an error: could not find function "activateEnvironment" >

Aaron Lun (11:41:07): > Should be there inbasilisk.utils.

Kevin Rue-Albrecht (11:41:25): > it only fails on ubuntu it seems

Kevin Rue-Albrecht (11:44:15): > there’s this weird thing toohttps://github.com/kevinrue/velociraptor/runs/1960479532#step:28:270 > > -- RMarkdown error ------------------------------------------------------------- > /bin/bash: /github/home/.cache/basilisk/1.3.6/velociraptor-1.1.5/env/lib/libtinfo.so.6: no version information available (required by /bin/bash) > /bin/bash: /github/home/.cache/basilisk/1.3.6/velociraptor-1.1.5/env/lib/libtinfo.so.6: no version information available (required by /bin/bash) >

Kevin Rue-Albrecht (11:44:34): > it’s all happening duringpkgdown::build_site()it seems

Kevin Rue-Albrecht (11:47:07): > also > > Quitting from lines 57-66 (compare_velociraptor_runs.Rmd) > > points to the chunk that includes > > zellkonverter::writeH5AD(sce, X_name = "counts", > file=file.path(tmpdir, "anndata.h5ad")) >

Aaron Lun (11:47:50): > is this the same problem, or a bunch of different problems?

Kevin Rue-Albrecht (11:55:36): > they’re all problems in the same build, maybe related maybe not

2021-04-18

Aedin Culhane (17:36:03): > @Aedin Culhane has joined the channel

2021-05-11

Megha Lal (16:45:50): > @Megha Lal has joined the channel

2021-05-20

Aaron Lun (13:27:22): > @Charlotte SonesonI’m re-reading your velocity preprocessing paper again; with bussparse set to separate, how exactly are reads assigned to introns vs exons? are they double counted?

Charlotte Soneson (14:19:56): > Yes, in the ‘include’ approach they are captured for both the intronic and exonic counting.

Aaron Lun (14:22:08): > hm. that sounds kinda bad. or is there some EM magic that decides whether a read is assigned to one (spliced/unspliced) transcript over another (spliced/unspliced) transcript?

Charlotte Soneson (14:27:25): > There are different ways of counting (https://bustools.github.io/manual), including an EM algorithm, but as far as I understand the read (or BUS entry) is still included in both the intronic and exonic capture (which are counted separately).

2021-06-30

Somesh (05:43:46): > @Somesh has joined the channel

2021-07-26

Wes W (08:56:02): > @Wes W has joined the channel

2021-08-04

shristi shrestha (13:58:35): > @shristi shrestha has joined the channel

2021-09-06

Eddie (08:23:32): > @Eddie has joined the channel

Eddie (09:10:42): > @Eddie has left the channel

2021-09-09

Julien Roux (01:59:31): > @Julien Roux has joined the channel

2021-10-18

Qirong Lin (19:35:07): > @Qirong Lin has joined the channel

2021-10-19

Michael Love (09:19:04): > @Michael Love has joined the channel

Michael Love (09:33:27): > question for velocity folks: if we have an SCE with spliced (S), unspliced (U) and ambiguous (A) read counts, do you see any problem with passing a cell along with zero counts for U and/or A? To me, this shouldn’t pose a problem for downstream applications, but just want to check CC@Wes W

Wes W (09:50:37): > @Charlotte Sonesoncan correct me if I am wrong but my understanding is that if there is a non-positive value in there thenvelociraptoron defaults will throw a hissy fit at the on normalization throughscuttle. However if you passuse.theirs=TRUEthen there is no problems and let scVelo use its own normalization and feature selection. At first glance throwing out an entire cell with no Unspliced seems… not great… but I dont know scVelo’s source well enough to know if its doing that already and a ratio of X:0 is a deal breaker for it.

Charlotte Soneson (12:10:55): > I thinkscVelowill get around it here:https://github.com/theislab/scvelo/blob/1805ab4a72d3f34496f0ef246500a159f619d3a2/scvelo/preprocessing/utils.py#L622-L623(basically, don’t allow 0 column sums; and yes, I guess that should be fine downstream). By default, they won’t do any filtering (withuse.theirs=TRUE, the size of the output fromvelociraptor::scvelo()is the same as that of the input), and the normalization is done separately for the two matrices. Note that you can define size factors for spliced and unspliced matrices manually withscvelo(so you can set the size factors for the unspliced matrix to be the same as those for the spliced one).

Michael Love (12:13:39): > I don’t really get why someone would want different size factors for the same cell, but is that the default for some pipeline?

Michael Love (12:14:48): > I agree with this: > > so you can set the size factors for the unspliced matrix to be the same as those for the spliced one) > and hence I don’t think someone should throw out a cell just bc it may have no unspliced or ambiguous counts

Michael Love (12:15:27): > this may not be a big deal, maybe the counts for S and U or A are comparable so there aren’t many cells to worry about

Charlotte Soneson (12:25:05): > > this may not be a big deal, maybe the counts for S and U or A are comparable so there aren’t many cells to worry about > I’d say the ratio of S to U is perhaps in the range of 4-15 or so in typical data sets (you have to put the A in the S and U bins somehow before runningscvelo). As far as I know,scVelowill normalize the matrices separately by default.

Michael Love (12:43:05): > huh. it (norm separately) seems strange to me bc it’s just a technical artifact (seq depth) that should be the same across each assay

Michael Love (12:43:41): > I guess if you have a cell with a count of 20 and then it has a 0 for U it’s really not a big deal to toss that cell

Michael Love (12:44:03): > ok thanks, i think it’s not something we have to worry about — people can toss these cells if they want:slightly_smiling_face:

Peter Hickey (17:24:36): > @Peter Hickey has joined the channel

2021-10-20

Michael Love (10:40:16): > ok and just an advert,@Dongze He@Wes Wand@Steve Lianoglouhave worked onloadFry()infishpondpackage which will be in upcoming release, it outputs a SCE

Dongze He (10:40:24): > @Dongze He has joined the channel

Steve Lianoglou (10:40:24): > @Steve Lianoglou has joined the channel

2021-10-28

Julien Roux (05:16:34): > Hi everyone! I am runningvelociraptorto runscveloon a dataset and understand better how different parameters and setups could affect the velocity results (last time I tried that I usedreticulateto runscvelofunctions directly). > If I’m not mistaken there is no wrapper for thescv.tl.rank_velocity_genesfunction which aims at extracting the genes contributing most the velocity signal across clusters. Looking at their help page (https://scvelo.readthedocs.io/scvelo.tl.rank_velocity_genes/) it says: > > This applies a differential expression test (Welch t-test with overestimated variance to be conservative) on velocity expression, to find genes in a cluster that show dynamics that is transcriptionally regulated differently compared to all other clusters (e.g. induction in that cluster and homeostasis in remaining population). > So if I understand well to perform a similar task I could simply run thefindMarkersfunction on thevelocityassays of thescvelo()output, providing my cluster assignments as argument. Do you think this makes sense?

Kevin Rue-Albrecht (06:32:58): > Hi@Julien RouxThanks for usingvelociraptor! > From a brief look at the python function, your reasoning sounds right:https://github.com/theislab/scvelo/blob/master/scvelo/tools/rank_velocity_genes.py#L194L323, the velocity layer is extracted > L340, the means and variances are compared > I’d certainly be interested to hear whether the results that you get out of findMarkers are similar enough to the original python output.

2021-10-29

Julien Roux (08:32:29): > I am using the argumenttop.genesin thescvelo()function but in the slotrowData(velo.out)$velocity_genesnot all values areTRUE. I guess not all genes given as input are used for the fitting…

Julien Roux (08:33:30): > Do you know how these velocity genes are selected?

Julien Roux (08:34:54): > My best guess is that in this function:https://scvelo.readthedocs.io/scvelo.tl.recover_dynamics/the parameter****var_names******** ****interferes with the selection of genes (it operates a selection among the HVGs I give as input), maybe I should control that with thescvelo.paramsargument ofscvelo()? I could force scvelo to use all genes given as input the function

Julien Roux (08:36:05): > Did anyone look into this before?

Charlotte Soneson (08:55:04): > Velocity genes are defined here:https://github.com/theislab/scvelo/blob/1805ab4a72d3f34496f0ef246500a159f619d3a2/scvelo/tools/velocity.py#L92-L98- basically, they need to have at least a certain R2 (I think 0.01 by default)

Julien Roux (10:32:08) (in thread): > Thanks Charlotte! It’s funny how changes in parameters can actually change the velocity genes!

2021-11-08

Paula Nieto García (03:29:33): > @Paula Nieto García has joined the channel

2022-01-14

Michael Love (08:52:23): > glad to see this acknowledged by the larger community: > > Besides this inherent noise, preprocessing decisions in the alignment of spliced and unspliced reads impact the final velocity estimate [Soneson 2021].

Michael Love (08:52:48): > https://doi.org/10.1038/s41592-021-01346-6 - Attachment (Nature): CellRank for directed single-cell fate mapping > Nature Methods - CellRank infers directed cell state transitions and cell fates incorporating RNA velocity information into a graph based Markov process.

2022-01-28

Megha Lal (11:14:41): > @Megha Lal has left the channel

2022-02-15

Gene Cutler (12:01:33): > @Gene Cutler has joined the channel

2022-03-08

Andrew McDavid (13:56:51): > @Andrew McDavid has joined the channel

2022-03-21

Pedro Sanchez (05:02:37): > @Pedro Sanchez has joined the channel

2022-04-28

Jared Andrews (17:34:39): > @Jared Andrews has joined the channel

2022-07-15

Ashley Robbins (15:18:29): > @Ashley Robbins has joined the channel

2022-08-11

Rene Welch (17:16:16): > @Rene Welch has joined the channel

2022-08-15

Michael Kaufman (13:13:48): > @Michael Kaufman has joined the channel

2022-11-06

Sherine Khalafalla Saber (11:21:23): > @Sherine Khalafalla Saber has joined the channel

2022-12-12

Carlos José Ferreira da Silva (18:58:09): > @Carlos José Ferreira da Silva has joined the channel

Carlos José Ferreira da Silva (18:58:13): > @Carlos José Ferreira da Silva has left the channel

2022-12-20

Jennifer Foltz (10:41:29): > @Jennifer Foltz has joined the channel

2023-05-12

Aaron Lun (13:33:17): > @Aaron Lun has left the channel

2023-06-19

Pierre-Paul Axisa (05:12:21): > @Pierre-Paul Axisa has joined the channel

2023-07-12

Axel Klenk (19:33:47): > @Axel Klenk has joined the channel

2023-07-28

Benjamin Yang (15:58:56): > @Benjamin Yang has joined the channel

2023-07-31

Chenyue Lu (17:51:16): > @Chenyue Lu has joined the channel

2023-09-13

Christopher Chin (17:05:03): > @Christopher Chin has joined the channel

2023-09-20

Alik Huseynov (04:51:16): > @Alik Huseynov has joined the channel

2023-11-18

Michael Love (09:18:12): > @Michael Love has left the channel

2024-04-24

Vince Carey (14:11:14): > @Vince Carey has joined the channel

Vince Carey (14:21:23): > @Rob PatroI would like to simplify use of splici indices which can be time-consuming to build. Are the instructions athttps://combine-lab.github.io/alevin-fry-tutorials/2021/alevin-fry-velocity/up to date? I think I would start by building the indices for mm10 and GRCh38 and putting them in compressed tars in open storage buckets. Then add a Bioconductor package that knows how to find and cache them for downstream use. Are the same indices used for piscem? - Attachment (combine-lab.github.io): An introduction to RNA-velocity using alevin-fry > Recently, RNA-velocity estimation has becomes increasingly popular tool in single-cell RNA seq analysis. In this post, we will discuss an additional advantage brought by the Unspliced-Spliced-Ambiguous (USA) mode introduced in alevin-fry 0.3.0 and later. That is, the solution presented in that approach for controlling the spurious mapping to spliced transcripts of sequenced fragments arising from introns (in the absence of full decoy) basically gives us the preprocessing results we need to perform an RNA-velocity analysis “for free”. Here we provide an end-to-end tutorial describing how to perform an RNA-velocity analysis for a 10x Chromium dataset. In this tutorial, we will show the whole analysis pipeline, starting from the raw FASTQ files to the gorgeous velocity plots (generated by scVelo) that you may like to include in your next analysis or paper.

Rob Patro (14:21:26): > @Rob Patro has joined the channel

stefano mangiola (20:48:46): > @stefano mangiola has joined the channel

2024-08-27

Rob Patro (10:07:47): > Hi@Vince Carey— sorry that I seem to have completely missed this call out! So the overall strategy is the same, yes, but it’s way easier now that we recommendsimpleafas the way to run alevin-fry.

NILESH KUMAR (11:37:24): > @NILESH KUMAR has joined the channel

2025-01-09

Ammar Sabir Cheema (11:40:36): > @Ammar Sabir Cheema has joined the channel