DeLisa Research Group
93 posts
@DeLisaGroup
We are the DLRG. We engineer bacteria to solve problems in biology and medicine. @Cornell University.
Protein Language Model Hits Undruggable Targets, No Structure Required PepMLM generates binders to challenging #therapeutic targets across cancer and neurological disease using protein sequence and no structure @pranamanam @Penn #AI #proteins hubs.li/Q03CMPXh0
How are we going to develop more domestic manufacturing capability that can deliver medicines? #cellfree biotechnology can help. Check out our new manuscript on scalable cell-free protein synthesis w? @akarimlab @DeLisaGroup @aw_rochelle doi.org/10.1002/bit.28…
Our pLM-designed peptides, when attached to E3 ligases (aka uAbs), enable us to programmably degrade disease targets (think CRISPRi, but for proteins). But what if we want to STABILIZE useful proteins? In @NatureComms, we introduce duAbs! 🌟 📜: nature.com/articles/s4146… 🧬: addgene.org/232089/ CRISPR is so powerful because of it's programmability. 🧬 Design a guide RNA, and you can guide a Cas enzyme to almost any loci for editing, especially with our PAM-flexible Cas9s! And this is not just to edit, you can both activate (CRISPRa) and inhibit (CRISPRi) gene expression too! ⬆️⬇️ My lab's goal is to build an analogous system for programmable proteome editing—where we stabilize (CRISPRa-like) or degrade (CRISPRi-like) any protein using AI-generated "guide" peptides, like those from our peptide-generating pLMs: SaLT&PepPr, PepPrCLIP, and PepMLM! 💻 In the wet lab (yes, half of my lab is experimental! 👩🔬), we routinely design and clone guide peptides upstream of E3 ubiquitin ligase catalytic domains to create uAbs. As you've seen in our papers, because we don't use structure-based methods (i.e. RFDiffusion or AF-based methods), we've ubiquitinated and degraded diverse disease targets with uAbs, from stable to heavily disordered! 🧫 Here, instead of E3 ligases, we fused our peptides to the OTUB1 deubiquitinase catalytic domain (thus removing ubiquitin chains) — creating a fully programmable, genetically-encoded tool for targeted protein stabilization (TPS)! Our duAbs can work on pretty much any target! 🙌 Just design, clone, and transfect! 💻➡️🧬➡️🧫 We first showed that duAbs work through true DUB-dependent mechanisms—using catalytic mutants (🧬🔧) and pan-DUB inhibitors (🧪) to confirm that stabilization depends on active deubiquitinase function. We also confirmed, via whole cell proteomics, that duAbs exhibit low off-target effects. 🎯 Then, to show programmability... We designed duAbs (via all of our algorithms) and successfully stabilized both semi-structured regulatory proteins like β-catenin and WEE1, as well as heavily disordered, undruggable TF targets like FOXP3 and the fusion oncoprotein PAX3::FOXO1 in ARMS cells! 🍝 Finally, we turned our attention to the holy grail target for TPS: tumor suppressor p53! 🧬 With PepMLM peptides, we first showed strong stabilization via plasmid transfection. 👍 We then encapsulated our best p53-targeting duAb as mRNA in LNPs 💊🧬, delivered them to human cells 🧫, and observed potent p53 stabilization along with downstream apoptosis induction (via cleaved PARP-1) ☠️— a major step toward in vivo proteome editing! 🐁 This is the culmination of a two year-long, herculean effort of my brilliant, first-ever PhD student, @lauren_hong11! 🌟 She is the glue of our lab, and I am just so grateful for the sheer persistence and positivity she brings every. single. day. 🦸♀️ I literally cannot imagine the lab without her. 🥹 We're also super grateful to our collaborators in the @DeLisaGroup, and the students (most of whom @lauren_hong11 mentored, like Tian who performed all of the LNP work!) who made this possible! 🫶 Please take a read and try out our duAb system to stabilize whatever protein you want! We have deposited the plug-and-play cloning vector on @Addgene (#232089)! 🧬
