Replying to
@yihui
After a inspirational introduction on minimalism, Yihui presents
litedown: R Markdown Reimagined
https://yihui.org/litedown/
Remote
Laurent Gatto
@lgatto@fosstodon.org
mastodon 4.7.3Inclusive, open and reproducible research, computational biology, omics, emacs, R, running, parenting.
Software/rstats package development (mostly for omics/proteomics/mass spec) are public on my own github or on RforMassSpectrometry: https://github.com/RforMassSpectrometry/
I am also interesting in teaching and pedagogy - see this post for my latest thoughts on the topic https://lgatto.github.io/ungrading/.
Prof at UCLouvain in Brussels.
1245 Followers
629 Following
19 Posts
Joined November 06, 2022
github:
Pronouns:
he/him
Replying to
scplainer models biological and technical sources of variability concurrently, effectively decoupling biological information from batch effects. It also includes cell normalisation during data modelling.
Replying to
Before modelling, sclpainer applies a minimal data processing approach. Most data processing steps are interdependent and involve an estimation step. Therefore, applying them sequentially may lead to artefacts. scplainer, instead, transfers most steps during the modelling step.
Replying to
Not at all! We can find peptides that mostly consist of biological variation and we do find peptides with consistent biological effects despite important batch effects. Strong batch effects do not harm data quality as long as these effects are thoroughly modelled.
Replying to
For instance, vimentin is more abundant in melanoma cells and may explain the increased cell adherence compared to monocytes. Since we model the data at the peptide level, we could see that the increase is consistent across most vimentin peptides.
Replying to
scplainer will be (very) soon available in scp so that you can try it yourself. Need data? All data used in this work were retrieved using our scpdata package! (both available on @bioconductor@genomic.social)
https://github.com/UCLouvain-CBIO/scp/
https://github.com/UCLouvain-CBIO/scpdata/
Replying to
Thank you to Christophe Vanderaa for his fantastic work! This work concludes his outstanding contributions to single-cell data analysis 👏 !
Replying to
Single-cell proteomics (SCP) data are complex. To better understand these data, scplainer relies on linear models for interpretable and explorable results.
How does it work?
Replying to
The scplainer model is adapted to each peptide based on the patterns of missing values, removing the need for excessive missing value filtering or missing data imputation.
Peptides for which there are fewer observations than model parameters are ignored, alleviating the need to remove highly missing peptides based on an arbitrary proportion of missing values.
Replying to
sclpainer then explores the model results through analysis of variance, differential analysis and component analysis. Let's demonstrate this in practice.
Replying to
Analysis of variance quantifies the amount of information captured by each descriptor in the model. Applying scplainer to several SCP data set revealed that these data mainly consist of technical sources of variation and little bioligical information. Are these data useless?
Replying to
Differential abundance analysis delves deeper into the exploration and understanding of the biological effects. We found many peptides differentially abundant in the nPOP data set from the @nslavov@mstdn.science explaining morphological and functional differences between monocytes and melanoma cells.
Replying to
The component analysis condenses the data into a few dimensions, enabling the exploration of cellular patterns. scplainer leverages the APCA+ framework to assess the quality of batch correction and to explore cellular heterogeneity.
Replying to
We seamlessly retrieved the melanoma subpopulation that has been validated by @AndrewLeduc@genomic.social . After remodelling the data upon clustering, we found evidence for increased oxidative phosphorylation and decreased glycolysis processes in the melanoma subpopulation.
Replying to
While scplainer models data at the peptide level, we provide tools to also generate results at the protein level. It also provides functionality to generate batch corrected for further downstream processing or interactive visualisation, such as with @bioconductor@genomic.social and @csoneson@fosstodon.org iSEE.
Replying to
scplainer can be adapted to any SCP experimental design. Checkout the SCP.replication website for other example.
https://uclouvain-cbio.github.io/SCP.replication/