my biggest advice to undergrads that ask for career advice is 'don't talk to the people you don't want to talk to'.

RT @andrewwhite01: The new @METR_Evals results show that experienced developers think their productivity is increasing, but it's actually d….

RT @ElkaFirmanda: 😍.

RT @debesh_man: Very cool.

RT @shanilpanara: ✨ Congrats on the release of BAGEL ✨. I've been behind the scenes watching the huge amount of work from @jakublala, Ayham….

RT @ayanobacter: 3 things I love here:. Open source. Programmable protein design. And BAGEL(s) ofc.

RT @chaitjo: A new approach to the multi-state design problem!.

RT @harshag1312: Very excited to finally see what @jakublala has been cooking for so long! This is a different way of approaching the prote….

If you'd like to collaborate, get in touch at jakublala@gmail.com, or DM directly. 🦸‍♂️. [9/8+1].

Also, huge thanks to the folks who made this possible - Stefano, Daniele, Shanil, Arnav, Harsh and Ayham! 💙. [8/7+1].

We wrote BAGEL with generality and community in mind - so please, feel free to design new energy terms, implement better structure predictors, or add new optimization & mutation protocols!.🧑‍🔬 Lastly, we are making an open call for potential experimental collaborators to.

To simplify dependency management across deep learning models, we’re also releasing a companion package, boileroom, which abstracts this away via @modal_labs integration. Code (boileroom) ⚙️:

The last application is especially interesting:.💡 Monte Carlo sampling in embedding space to generate novel enzyme variants while keeping the active site intact. Previously put out by @aminoadam, @amino_abhi (both at @AminoAnalytica), and @QTroillet, but given BAGEL’s.

We showcase BAGEL’s use case across four application domains:.🧬 de novo binder design.🌀 targeting disordered regions.🔍 selective binders via multi-state optimization.⚗️ enzyme variant generation w/ conserved active sites.All examples + templates are open-sourced on the GitHub

BAGEL has been built from ground up to be fully modular and extensible:.🔁 Plug in your own samplers (Monte Carlo, tempering, annealing…).🔀 Choose your mutation protocol - including a grand-canonical variant with insertions & deletions, inspired by molecular simulation methods

Instead of rigid protein design pipelines, BAGEL formalizes protein design as an energy optimization task - letting you express complex design goals using simple energy terms. One can think of arbitrary design goals, without the need for end-to-end differentiable models/metrics.

🚨 Excited to share the first major project of my PhD: BAGEL, a modular, model-agnostic, fully customizable Python framework for programmable protein design 🧬🥯.Code (BAGEL) 💻: Code (boileroom) ⚙️: Preprint 📜:

RT @AllThingsApx: pLM’s embeddings == “energy map” ⚡. Monte Carlo sampling swaps up to 90 % of residues while keeping the active site withi….

RT @DenisVfilms: Coming in 2026

RT @matthewclifford: As reported, I’ll be stepping back from my role as the Prime Minister’s Adviser on AI at the end of July. This will….

RT @NotinPascal: 🚨 New paper 🚨 RNA modeling just got its own Gym! 🏋️ Introducing RNAGym, large-scale benchmarks for RNA fitness and structu….