←
Andreas Gschwind61 posts
Andreas Gschwind avatar

Andreas Gschwind

@argschwind

Computational Biologist at @Stanford.

◷ Joined September 2015▧ 7 Photos and videos
Jesse Engreitz avatar
⟳ Andreas Gschwind retweeted
Jesse Engreitz@jengreitz·Aug 4

Update: scE2G is now online at @NatureGenetics! 📄 nature.com/articles/s41588-0… Since the preprint, we added a new benchmark on a fully held-out CRISPR dataset. scE2G continues to show state-of-the-art performance, reinforcing that it generalizes well 1/

Post attachment

Introducing scE2G: a new model to link enhancers to target genes using single-cell data. Excited that scE2G will enable building enhancer maps in hundreds of cell types in the human body! Wonderful collaboration with @robin_andersson @613weilin @mayayayas and others 👇

Kushal K Dey avatar
⟳ Andreas Gschwind retweeted
Kushal K Dey@kanishkadey·Jul 15

How to link a non-coding disease GWAS variant to its target genes in the right cell type? ENCODE-rE2G, now out in @Nature addresses this through optimal integration of chromatin activity and 3D contact data in 1,458 biosamples. A huge collaborative effort spanning 6+ years led by @argschwind @jengreitz nature.com/articles/s41586-0…

Andreas Gschwind avatar
Andreas Gschwind@argschwind·Jul 15

Our paper is now out in Nature! 🎉 Expanded to 92M+ enhancer–gene links across 1,458 biosamples, with new benchmarks and analyses. "An encyclopedia of human enhancer–gene regulatory interactions": nature.com/articles/s41586-0… Grateful to the whole team of this massive effort!

Excited to share our latest work on bioRxiv! biorxiv.org/content/10.1101/… A highly collaborative effort with @ksmualim @karbalayghareh, @mayayayas, @kanishkadey, @evelynjagoda,@WangXijhu, @LarsMSteinmetz, @anshulkundaje, @jengreitz and many others from @ENCODE_NIH! Thread👇

Andreas Gschwind avatar
Andreas Gschwind@argschwind·Jul 15

Our paper is now out in Nature! 🎉 Expanded to 92M+ enhancer–gene links across 1,458 biosamples, with new benchmarks and analyses. "An encyclopedia of human enhancer–gene regulatory interactions": nature.com/articles/s41586-0… Grateful to the whole team of this massive effort!

Andreas Gschwind avatar
Andreas Gschwind@argschwind·14 Nov 2023

This has been a wonderful community effort involving countless people from the @ENCODE_NIH consortium and beyond. I would like to thank everybody I had the pleasure to work with during this long and exciting journey! 17/17

Andreas Gschwind avatar
Andreas Gschwind@argschwind·14 Nov 2023

Summary: We present a new approach to map genome-wide enhancer-gene regulatory interactions, show how to link variants to genes, and interpret the model to learn about enhancer regulation. Our framework provides a resource for building future, increasingly accurate models. 16/n

Andreas Gschwind avatar
Andreas Gschwind@argschwind·14 Nov 2023

What assays do you need to build ENCODE-rE2G maps in a new cell type? In short, start with DNase-seq (which already provides very good performance), then add H3K27ac ChIP-seq or cell-type specific Hi-C to improve performance further: 14/n

Post attachment
Andreas Gschwind avatar
Andreas Gschwind@argschwind·14 Nov 2023

Regarding enhancer-enhancer interactions — We followed up to experimentally examine whether enhancers act additively or non-additively on gene expression in CRISPR data. Enhancers in close proximity can influence each others’ activity, leading to super-additive effects. 13/n

Post attachment
Andreas Gschwind avatar
Andreas Gschwind@argschwind·14 Nov 2023

Top features in the model: 1. ABC score (combination of enhancer activity and 3D E-P contact) 2. Promoter class: Housekeeping genes are less responsive to enhancers 3. Enhancer-enhancer interactions: Nearby enhancers increase enhancer activity. 12/n

Post attachment
Andreas Gschwind avatar
Andreas Gschwind@argschwind·14 Nov 2023

So, how does the ENCODE-rE2G model work? It’s a logistic regression model using (in the simplest version) 13 features derived from DNase-seq data to predict CRISPR results in hold-one-chromosome-out cross-validation. This allows us to look at which features are important: 11/n

Andreas Gschwind avatar
Andreas Gschwind@argschwind·14 Nov 2023

Example: we find a fine-mapped causal variant for mean corpuscular hemoglobin linked to the gene CPEB4 by ENCODE-rE2G+PoPs only in hematopoietic progenitors and not in other closely related blood cell types like B-cells, T-cells and myeloid progenitors. 10/n

Post attachment
Andreas Gschwind avatar
Andreas Gschwind@argschwind·14 Nov 2023

How can we use ENCODE-rE2G to identify target genes of GWAS variants? By combining ENCODE-rE2G with PoPs, we achieve high precision at linking non-coding GWAS variants to genes. This not only prioritizes disease genes, but also provides information on their regulation. 9/n

Post attachment
Andreas Gschwind avatar
Andreas Gschwind@argschwind·14 Nov 2023

By benchmarking ENCODE-rE2G against fine-mapped eQTL and GWAS variants, we show that the model generalizes beyond the CRISPR training set to make predictions in new cell types, and enables linking GWAS variants to known genes. 8/n

Andreas Gschwind avatar
Andreas Gschwind@argschwind·14 Nov 2023

Using CRISPR perturbation screens, we evaluated two versions of ENCODE-rE2G: one that uses features derived from DNase-seq only, and one that uses an extended feature set derived from many different assays. Result: ENCODE-rE2G performs very well compared to other methods. 7/n

Post attachment