Sign in

Shenzhi Chen

@shenzhichen1999.bsky.social
38 followers 42 following 19 posts

PhD student in Stark group@IMP Vienna, Machine Learning, Gene Regulation

PostsRepliesMedia
Shenzhi Chen @shenzhichen1999.bsky.social · 25/08/2026
Grateful for the amazing guidance from @alex-stark.bsky.social and Vincent Loubiere to help me make my first PhD project come ture! Huge thanks to all co-authors and collaborators, especially Evgeny @evgenykvon.bsky.social and Jacob @jmschreiber91.bsky.social , for making this work possible. 🙏
050 0
Shenzhi Chen @shenzhichen1999.bsky.social · 25/08/2026
Interestingly, tissue-specific accessible regions alone gave performance approaching that of models fine-tuned on VISTA. This suggests a practical route to enhancer design in tissues where experimentally validated enhancer datasets are scarce.
131 0
Shenzhi Chen @shenzhichen1999.bsky.social · 25/08/2026
What if no in vivo–validated enhancer dataset is available? We benchmarked inferred enhancer labels based on: • distal accessible regions without CTCF • H3K4me1/H3K27ac-overlapping accessible regions • tissue-specific accessible regions
120 0
Shenzhi Chen @shenzhichen1999.bsky.social · 25/08/2026
In vivo validation was the critical test: ❤️ heart: 5/5 🦵 limb: 5/5 🧠 midbrain: 5/5
130 0
Shenzhi Chen @shenzhichen1999.bsky.social · 25/08/2026
To move from prediction to design, we optimized sequences de novo for both accessibility and tissue-specific enhancer activity using Ledidi, a model-guided, gradient-based sequence design method: www.biorxiv.org/content/10.1...
biorxiv.org
120 0
Shenzhi Chen @shenzhichen1999.bsky.social · 25/08/2026
The models also captured interpretable regulatory logic, including tissue-specific TF motifs and motif combinations/grammar. This suggests that part of the mammalian cis-regulatory code is sufficiently learnable to support predictive sequence design.
120 0
Shenzhi Chen @shenzhichen1999.bsky.social · 25/08/2026
We used transfer learning: first pretraining sequence models on genome-wide chromatin accessibility, then fine-tuning them on experimentally validated tissue-specific enhancer activity from VISTA.
111 0
Shenzhi Chen @shenzhichen1999.bsky.social · 25/08/2026
Excited to share our paper in @natgenet.nature.com on the de novo design of tissue-specific mammalian enhancers that function in vivo in mouse embryos. 🧬🐭 www.nature.com/articles/s41...
35826 0
Shenzhi Chen @shenzhichen1999.bsky.social · 24/12/2025
We use compact CNNs and transfer learning (pre‑trained on genome‑wide ATAC‑seq and fine‑tuned on validated enhancers from VISTA enhancer browser) to program tissue‑specific enhancers. (2/6)
000 0
Shenzhi Chen @shenzhichen1999.bsky.social · 24/12/2025
Transfer learning reveals clear shifts in motif importance between sequence‑to‑accessibility and sequence‑to‑activity models, highlighting the specific features that drive enhancer function. (3/6)
000 0
Shenzhi Chen @shenzhichen1999.bsky.social · 24/12/2025
100% (15/15) of designed enhancers for heart, limb, and CNS are active in their intended tissues in mouse embryos. 🎯 (4/6)
000 0
Shenzhi Chen @shenzhichen1999.bsky.social · 24/12/2025
We hope this approach will help enable the design of more versatile and precise synthetic enhancers across mammalian tissues, cell types, and dynamic cell states. (5/6)
000 0
Shenzhi Chen @shenzhichen1999.bsky.social · 24/12/2025
Grateful for the amazing guidance from @alex-stark.bsky.social and Vincent Loubiere to help me make my first PhD project come ture! Huge thanks to all co-authors and collaborators, especially Evgeny @evgenykvon.bsky.social and Jacob @jmschreiber91.bsky.social, for making this work possible. 🙏 (6/6)
000 0
Shenzhi Chen @shenzhichen1999.bsky.social · 24/12/2025
🚀 Excited to share our latest work now on bioRxiv: the computational de novo design of tissue‑specific mammalian enhancers that work in vivo in the mouse embryo. 🐭 More details in the preprint and code: bioRxiv: biorxiv.org/content/10.6... (1/6)
biorxiv.org
Predictive design of tissue-specific mammalian enhancers that function in vivo in the mouse embryo
Enhancers control tissue-specific gene expression across metazoans. Although deep learning has enabled enhancer prediction and design in mammalian cell lines and invertebrate systems, it remains uncle...
500 0
Shenzhi Chen @shenzhichen1999.bsky.social · 24/12/2025
We use compact CNNs and transfer learning (pre‑trained on genome‑wide ATAC‑seq and fine‑tuned on validated enhancers from VISTA enhancer browser) to program tissue‑specific enhancers. (2/5)
000 0
Shenzhi Chen @shenzhichen1999.bsky.social · 24/12/2025
Transfer learning reveals clear shifts in motif importance between sequence‑to‑accessibility and sequence‑to‑activity models, highlighting the specific features that drive enhancer function. (2/5)
000 0
Shenzhi Chen @shenzhichen1999.bsky.social · 24/12/2025
100% (15/15) of designed enhancers for heart, limb, and CNS are active in their intended tissues in mouse embryos. 🎯 (3/5)
000 0
Shenzhi Chen @shenzhichen1999.bsky.social · 24/12/2025
We hope this approach will help enable the design of more versatile and precise synthetic enhancers across mammalian tissues, cell types, and dynamic cell states. (4/5)
000 0
Shenzhi Chen @shenzhichen1999.bsky.social · 24/12/2025
Grateful for the amazing guidance from @alex-stark.bsky.social and Vincent Loubiere to help me make my first PhD project come ture! Huge thanks to all co-authors and collaborators, especially Evgeny @evgenykvon.bsky.social and Jacob @jmschreiber91.bsky.social, for making this work possible. 🙏 (5/5)
000 0