A wonderful article on PepMLM by @xiaofei_lin in @GENbio! 🌟 Really captures how we think about sequence vs. structure and how, via experimental collaboration, we've gotten biologists (like Ray Truant, @DeLisaGroup, and @AguilarVirology) to believe in sequence-based design. 💚

PepMLM is out in @NatureBiotech! 🎭Want a short binder for any target? Just input its sequence + desired binder length → PepMLM unmasks it. Simple? No way? The results speak for themselves. 👇 📜: nature.com/articles/s41587-0… 💻: github.com/programmablebio/p… 🤗: huggingface.co/ChatterjeeLab… A few years ago, I told my undergrad @SophieVincoff (now my PhD student) to try the obvious: given the target sequence, use autoregressive next-token prediction to generate the cognate peptide binder. After tons of work, we still couldn't get it to work too well. 😔Then, my Masters student (now Harvard PhD student) @LeoTZ03 said: what if we just attach the peptide sequence to the end of its target sequence and fully fine-tune an encoder (in this case, ESM-2-650M) to unmask it? Basically, a BERT-like take on conditional (short) sequence generation. I said to @LeoTZ03: sounds kind of silly.😅 Well, he did what any good student would do: he ignored me, trained the model on ~10k putative peptide-protein sequence pairs, and the in silico results were pretty amazing! PepMLM consistently outperformed RFDiffusion on held-out/structured targets, with a higher hit rate (38% to 29%) and low perplexities that closely matched real binders, with generated sequences showing target specificity even in stringent permutation tests! 🦾 But the proof was in the experiments, and we did a LOT of them! 🧪 In our lab, Zach showed that PepMLM peptides achieved nM binding affinity on disease-related receptor targets, like NCAM1 and AMHR2 (RFDiffusion peptides failed to bind). With the Truant Lab @McMasterU, we showed PepMLM peptides, when fused to E3 ubiquitin ligases (our uAb architecture), not only degraded MSH3 but completely eliminated Huntington protein in HD cells! 🧫 We showed similar results by identifying a PepMLM uAb that degraded MESH1, a protein controlling ferroptosis, in collaboration with @TsanJen's group @DukeU. Again, no hits for RFDiffusion there. And with @AguilarVirology (led by @MadeleineDumas2), in collaboration with @DeLisaGroup @Cornell, PepMLM-derived peptides (in uAbs) bound and reduced levels of viral phosphoproteins from Nipah, Hendra, and HMPV, and in live HMPV infection models, they almost completely cleared the P protein! 🦠 I guess sometimes the "silly" ideas win. 😂 Even though my lab is theoretically-inclined (discrete diffusion/flow matching/Schrodinger bridges) to now design more specialized/optimized/specific molecules @Penn, PepMLM is still a go-to model for my experimentalists. Remember @lauren_hong11's duAbs in @NatureComms that stabilized THE tumor suppressor p53? Yep, PepMLM. 💁‍♂️ In fact, PepMLM has averaged ~600 downloads a month on @huggingface over the last year! It's so easy to use (takes seconds to run on Colab, with just a few lines of transformers code), you don't need to input (or go through) structure, you can enter almost any target sequence (disordered/stable, short/long), and you get a binder. And again, don't believe me, look at the diversity of proteins we've gone after! If you're an experimentalist, it's super worth it to try PepMLM. 😇 The code/model is now fully open-sourced for both academia and industry! 🔓 Btw, this was an incredibly collaborative undertaking: 5 labs, across multiple amazing universities, led by co-first authors @LeoTZ03 (who led all model design, training, and validation), and Zach (@DukeU), Madeleine (@Cornell), Christina (@McMasterU ) who led each of the experimental efforts!! I'm so proud and grateful for everybody's belief that yes, sequence can be all you need for binder design! 😉

I'm usually not too emotional about paper acceptances, but this one justifies it. 🥹 It doesn't feel real, but PepPrCLIP is now published at @ScienceAdvances! Let me tell you its story. 📕 📜: science.org/doi/10.1126/scia… 💻: huggingface.co/ubiquitx/pepp… PepPrCLIP (🌶️📎) began as Cut&CLIP, before I even came to @DukeU in 2022, where my first @Harvard undergrad mentee @kalyanmpalepu (now at @DEShawResearch ) would cut ✂️peptides from interacting partner proteins, and throw them thru a trained peptide-protein CLIP 📎 model to predict which ones were specific to the target (cool use of DALL E 2's architecture, right?). We had good initial experimental results, but they weren't robust enough for publication. 🤔 While the team tried to get our CLIP model better for experimental testing, my amazing other @Harvard undergrad @garykbrixi came up with a really unique way of cutting peptides from the ESM-2-predicted binding sites of partners (our SaLT&PepPr model🧂). We changed the focus to SnP (get it, SnP ✂️?), and spent over a year through numerous good/bad review processes, and FINALLY got it out last October in @CommsBio: nature.com/articles/s42003-0… 🥳 @bhat_suhaas, a @Harvard undergrad, later a Rhodes Scholar (so proud of you!! 🥹🫵), and @kalyanmpalepu's roommate, who stuck with me as I transitioned to @DukeU, kept working on it with Kalyan, and not only trained a beautiful CLIP model on peptides and proteins with ESM-2 embeddings (goodbye ESM-1b and MSA transformer!), but also devised an incredibly clever way of generating new peptides via Gaussian perturbation of peptide embeddings for screening with CLIP, creating PepPrCLIP! 🌶️📎 The experimental validations were beautiful, and it became our first de novo peptide model in the lab! 👩‍🔬 We went into submission at another top journal, and worked hard to iron out IP issues, but at the end of the day, it was just too difficult to convince structure-focused reviewers how this was better than RFDiffusion 😑, even though we showed strong data showing PepPrCLIP-generated peptides allowed us to extend to conformationally disordered targets that RFDiffusion/structure-based methods could not access (we even did better on structured targets than RFDiffusion in vitro!). Frustrating. 😒 Still, we had incredible data! As you can see in our manuscript, PepPrCLIP-generated peptides can serve as inhibitory peptides to enzymes (i.e. biotin ligases, like UltraID) as well as degraders to transcription factors (β-catenin) and EVEN heavily-disordered fusion oncoproteins (SS18-SSX1)! 🧫 We showed extensive binding, inhibition, and degradation data in endogenous cellular settings and have now extended PepPrCLIP to more undruggable targets! And after over 2 years since our first submission, @ScienceAdvances, who published my first first-author paper back in graduate school, decided to accept it! 🌟 Full circle. ⭕️ Incredible. 😭 To the most important part: I am forever, forever grateful to @bhat_suhaas and @kalyanmpalepu's incredible dedication to this work, even after I moved to @DukeU and even after they graduated from @Harvard. 🫂 I love you guys. 🫶 And I couldn't be more thankful to our incredible collaborators @SoderlingLab @DeLisaGroup and @anideshpandelab (as well as the amazing experimentalists in my lab) who believed in our algorithm and did the painstaking experiments to prove it out. 🙏 Just an INCREDIBLE team-wide effort to get this over the finish line. I am just so lucky and so grateful -- I don't deserve such an amazing collaborators. 🥲 Finally, with the generous support of my company @UbiquiTxINC, we have made the code FREELY AVAILABLE to academics after signing a non-commercial license! I can guarantee you that PepPrCLIP peptides will work for your target! 😉 Feel free to read our paper and provide your thoughts! 💡We just got multiple other big paper acceptances and will be sharing those shortly as they come online! 🪇

Designing proteins is all fun and games, and yes, I believe sequence-based models do it better. 😅 BUT, at the end of the day, it only matters if we can actually have a meaningful impact with the proteins we design. And so I am SUPER EXCITED to show our new collaborative data on how we've done exactly that! 😉 If you remember, we previously developed the SaLT&PepPr (SnP) algorithm, where we first fine-tuned ESM-2 to predict binding interfaces on target-interacting proteins and then subsequently cut (SnP) these interfaces to make target-binding peptides (of course, with no structural input). ✂️ We showed that SnP-derived peptides, when attached to E3 ubiquitin ligase domains, induce highly potent and specific degradation of diverse target proteins in cells. 🙌 And that's amazing, but can we translate these degraders into an actual biologic? 🤔 In this paper, we decided to do exactly that for β-catenin, a key transcription factor that regulates the critical Wnt signaling pathway in cells but in its dysregulated form, can drive cancer. You see, regular, happy β-catenin is bound to E-cadherin at the cell membrane, but dysregulated β-catenin is found in the cytosol. Thus, we used SnP to "SnP" peptides from E-cadherin's interface, which only interact with cytosolic β-catenin! ✂️ We show that these SnP-derived peptides, when attached to E3 ubiquitin ligase domains, preferentially degrade cytosolic β-catenin in colorectal cancer cells, significantly reducing their viability and proliferation. We then encapsulated these degraders as mRNA in lipid nanoparticles (LNPs), delivered them intravenously into mice, and demonstrated cytosol-specific β-catenin degradation in vivo! 🐁 To the best of our knowledge, this is the first example of a language model-derived intracellular biologic working in animals! 🌟 Crazy stuff!! 🤯 This is the result of an amazing continued collaboration with the @DeLisaGroup @Cornell -- we have integrated our technologies together to hopefully have real clinical impact! 🤝 I mean, who could have guessed that language models would produce molecules with therapeutic effects in vivo? Well, I guess it's now possible! We're moving ahead on MANY other clinically-relevant, conformationally-diverse targets. 🤗 Please take a read of our paper and let us know what you think! And for those of you still thinking structure-based drug design is the only way to go, I hope that this data can convince you otherwise. 🥰 📜Paper: biorxiv.org/content/10.1101/… 🧂SaLT&PepPr Paper: nature.com/articles/s42003-0… 💻SaLT&PepPr Code: huggingface.co/ubiquitx/salt…

Exciting news from the lab! 👀 We have updated PepMLM on the @arxiv with new computational and experimental benchmarking results! 🥳 Just as a reminder: we trained PepMLM via a novel span masking strategy that positions peptide binder sequences at the C-terminus of their target protein sequences, and then fine-tuned ESM-2 to fully reconstruct the binder region. PepMLM's generated peptides outperform known peptides to target proteins with a ~40% hit rate, higher than RFDiffusion (~30%), while only using the target sequence! 🤯 In this new update, we experimentally demonstrate that PepMLM peptides bind potently to target proteins, and when fused to E3 ubiquitin ligase domains, exhibit degradation of emergent pandemic viral phosphoproteins and undruggable Huntington's disease-driving proteins (MSH3 and mHTT, itself)! 🧫 I can't say enough about the incredible wizardry of my little Master's student @LeoTZ03 🪄 (who's become a celebrity with his paper reviews!) as well as the wonderful ongoing collaboration with @DeLisaGroup and @AguilarVirology at @Cornell and the Truant Lab @McMasterU! Together, we're a strong team! 🤝 The paper is making it through review as we speak, but please check out the newest version of our preprint and also try out PepMLM on @huggingface! 🤗 Paper: arxiv.org/abs/2310.03842 Model: huggingface.co/ChatterjeeLab…