Iakov Davydov

What are the clinically meaningful ways of looking at tumor RNA-Seq?

Well, there are too many to count. But we show that RNA-Seq could be a good proxy for CD8+ T-cell localization within a tumor. (poster for #bc2 is here: https://bc2.ch/posters/gallery/169)

Why is this localization important? Immune phenotypes or spatial distribution of CD8+ T cells in the tumor microenvironment turns out to be predictive of treatment outcome in many indications (more here: https://doi.org/10.1016/j.immuni.2019.12.011).

(thread)

Join the conference!

The [BC]2 Basel Computational Biology Conference is Switzerland's main event in the domain of Computational Biology and one of the major events of its kind in Europe. The conference unites scientists from academia, industry and healthcare.

Sep 4, 2023 at 8:25amWeb
Show threadIakov DavydovSep 4, 2023
We used tumor RNA-Seq and IHC data from 2023 patients across 14 Phase I–III clinical trials to train an ElasticNet model. The model seems to distinguish immune phenotypes surprisingly well, given that bulk RNA-Seq doesn't have spatial resolution.
Show threadIakov DavydovSep 4, 2023
In addition, our predictions are relevant for the clinical outcome in cancer immunotherapy (OAK clinical trial).
Show threadIakov DavydovSep 4, 2023
Here’s where scRNA-Seq comes in.Predictions are almost as good when instead of genes, we only use a small set (92) of cell-type specific signatures previously derived from public single-cell RNA-seq data (more here: https://github.com/bedapub/besca). Something to keep in mind when training new RNA-Seq classifiers.
GitHub - bedapub/besca: BESCA (Beyond Single Cell Analysis) offers python functions for single-cell analysis

BESCA (Beyond Single Cell Analysis) offers python functions for single-cell analysis - GitHub - bedapub/besca: BESCA (Beyond Single Cell Analysis) offers python functions for single-cell analysis

GitHub
Show threadIakov DavydovSep 4, 2023
Finally, we used batch-correction to be able to run our model on RNA-Seq from TCGA. We now can see immune phenotype distribution across different indications. We see that predicted immune phenotypes affect overall survival. Hazard ratio across cancer types:
0.88 [0.80-0.97], but there are notable outlier indications. Stay tuned for the full paper.
Show threadIakov DavydovSep 4, 2023
You can already grab the R package from here: https://github.com/bedapub/cd8ippred
Before predicting, make sure that your data is comparable to what we used in the training; some batch correction might be required.
Many colleagues contributed to this study, please check the authors list on the poster.
GitHub - bedapub/cd8ippred: R package to predict immune phenotypes from RNA-Seq

R package to predict immune phenotypes from RNA-Seq - GitHub - bedapub/cd8ippred: R package to predict immune phenotypes from RNA-Seq

GitHub
Show threadIakov DavydovSep 4, 2023
In case you cannot access the link, here's the full poster.