Filippo Stocco

PhD in Protein Design @ Ferruz Lab | Research Intern @ NVIDIA

[Filippo Stocco @Filippo_Stocco_]

Watch your protein parameters go brrr 🚀
Check out our preprint on applying reinforcement learning (RL) to optimize Protein Language Models: https://t.co/OuZgLJ12Hz

With @mariartigues, Andi, @talaldotpdb, @marcguellc and @ferruz_noelia

[Tudor-Stefan Cotet @CotetTudor]

I'm super excited to share what we've been working on at Adaptyv: a new installment of our Protein Design Competitions, generously co-sponsored by Anthropic. We'll present 5 challenges across different protein targets, with new criteria such as conditional binding and binding specificity. The first one starts September 28th!

Our ambition was always to build a living benchmark that scales in scope and complexity the same way the field does. I've been amazed by the progress in protein design in just 2 years, from a more than 5-fold increase in hit rates across the 2 rounds of EGFR to methods winning on hard targets such as RBX1. And agents are now successfully orchestrating tools end to end. I think this competition will define the next frontier for binder design, and I wonder where we'll be 2 years from now.

Huge thanks to the Anthropic team for advancing the protein engineering frontier so decisively, and especially to @amirshanehsaz for carefully curating the 5 challenges.

Register here → https://t.co/cq1pzXQZLf

[Sergey Ovchinnikov @sokrypton]

Finally a more intuitive way to learn pLDDT/pAE? 😎
https://t.co/1YWv3AK1cU

(Character idea from @HannesStaerk & Alex Waldherr)

[sri @srivsesh]

really excited to share our work is in press @NatureComms today. when i met with @romerolab1 back in sep 2022, hoping to rotate in the lab, one of my first questions was “have you thought about RLHF?” his eyes lit up and he said “YES!”. fast forward to sep 2026, i can truly say this project challenged me in every way possible. this was a real labor of love. i learned so much about myself, the field, and where it’s heading working on this with @NathanielBlalo2 and the rest of my amazing colleagues. i hope you find this work insightful if you happen to give it a read.

https://t.co/M4ZpVAafgU

[Filippo Stocco @Filippo_Stocco_]

Upload your CSV, and it will fine-tune, apply RL, and generate new sequences from your experimental data!

ProtRL: https://t.co/fKC2Y7kR6m
Updated preprint: https://t.co/OuZgLJ0uS1

Thanks to @infinitycrab2, @ferruz_noelia, @mariartigues, @talaldotpdb, @marcguellc and Andi.

[Filippo Stocco @Filippo_Stocco_]

ProtRL just got an upgrade!

We’ve added:
1) Distributed training for scaling up protein RL
2) Easier support for custom loss
3) New experimental results in the updated preprint
Now you can make your proteins go BRRRR at scale 🚀
Want to get started? 👉 https://t.co/dRXNMBc9h5

[Filippo Stocco @Filippo_Stocco_]

Thanks to everyone who supported me and helped make this possible, including to @ferruz_noelia and team, @CRGenomica and @BecariosFLC for their support and flexibility.

[Filippo Stocco @Filippo_Stocco_]

I am in San Francisco for the next few months!
Starting a research internship at @NVIDIA (🤯) and very excited about what I’ll learn from the people around me.

If you’re around and up for a chat about what you’re working on, or (even better) a hike, let’s connect!

[Talal Widatalla @talaldotpdb]

Excited to share the final version of ProteinDPO is out today in @naturemethods! New in the paper is application of the model to stabilization of the pre-fusion state of hemagglutinin, the primary component of flu vaccines. 1/8

🔗https://t.co/Hp3m3FJFdF
📑https://t.co/mVggm42WEI

[Paola Laurino @PaolaLaurino]

We're seeking a motivated postdoc candidate to join at OIST to work on the structural and functional characterisation of plant enzymes that break down polysaccharides, exploring sequence space to understand the emergence of enzyme function using AI tools

[Michele @infinitycrab2]

🧬 Introducing ProtGPT3, a new open-source family of protein language models, from 112M to 10B parameters, now live on Hugging Face! @Filippo_Stocco_ @LasseMiddendorf @ferruz_noelia

Thread 🧵👇

[Noelia Ferruz @ferruz_noelia]

Very excited to share ProtGPT3! After ProtGPT2, we release open Hugging Face models up to 10B, with MSA prompting, DPO alignment, and experimentally expressed, soluble defluorinase designs. Congratulations to @infinitycrab2, @LasseMiddendorf, @Filippo_Stocco_, and the rest of the team. Read the details below :)

[Nora Martin @Nora_S_Martin]

🧵 New preprint:
In 2023, Papkou et al. reported an empirical fitness landscape with a striking property: >500 local optima ("peaks"), spanning both high and low fitness. Despite this ruggedness, >75% of simulated adaptive walks reached the top ~14% of peaks.

[Lasse Middendorf @LasseMiddendorf]

AI is everywhere in biology right now, but how do these models actually hold up once they hit the lab?

Our latest review on generative AI for enzyme design cuts through the hype to see what’s actually working in the lab🧪

⬇️LINK

[Synbio Lab @Synbio_Lab_UPF]

📣 SynBYSS 2026 Registration and Abstract Submission are Open! 🎉

Join us at @the_prbb, 8-11 June 2026 for the The 2nd International SynBYSS Conference!

🌐 More information: https://t.co/OECVyiaVVa

@JCVenterInst @UPFbiomed @Moon_Synth_Bio @marcguellc

[Noelia Ferruz @ferruz_noelia]

Rate our designs at the @adaptyvbio competition! 30 seconds is literally all it takes to support this group of ridiculously talented designers!🪄✨⭐️

[Lucia Urcelay @luciaurcelayg]

We at @ferruz_noelia lab participated in the @adaptyvbio Nipah binder competition, testing several design strategies! 🧵
⭐️ your favourite designs here

[Tudor-Stefan Cotet @CotetTudor]

You can now vote for your favourite submissions on Proteinbase!!

[Ge Liu @GeLiuSaber]

1-line code change. Better RL fine-tuning. Introducing ADRPO at #NeurIPS2025. Core Insight: Instead of a fixed $\beta$ (divergence weight), let the advantage guide it. A plug-n-play solution w/ strong results across diffusion T2I, LLMs & Audio Reasoning:
✅ 5× reward over GRPO + emergent escape from local optima via active exploration
✅ Finetuned 2B SD3 outperforms FLUX (12B) & SANA (4.8B) on ALL metrics, no diversity collapse
✅ 7B audio model beats Gemini 2.5 Pro & GPT-4o Audio

[Biology+AI Daily @BiologyAIDaily]

Guiding Generative Models for Protein Design: Prompting, Steering and Aligning

1. This review explores recent advances in guiding generative AI models for protein design, categorizing methods into train-time modifications and inference-time control. The focus is on steering models to produce proteins with rare or desired properties not well-represented in training data.

2. Train-time methods like supervised fine-tuning and reinforcement learning modify model parameters to align with specific design objectives. Fine-tuning adapts models to curated datasets, while RL explores novel solutions through reward-based optimization.

3. Inference-time control offers flexibility by influencing generation without altering model weights. Techniques include prompt engineering, retrieval-augmented generation, Bayesian guidance, and sampling controls, enabling rapid repurposing of models for diverse tasks.

4. The review highlights the potential of combining physics-based scoring with RL fine-tuning to move beyond natural protein distributions, addressing biases in training data and improving out-of-distribution generalization.

5. Laboratory automation and experimental validation are emphasized as crucial for reliable protein design, with lab-in-the-loop frameworks integrating generative models and experimental testing to accelerate innovation.

6. Future prospects include diversifying sequence sampling during pretraining and leveraging evolutionary context to improve model training efficiency and generalizability. The goal is to achieve truly controllable and creative protein design.

📜Paper: https://t.co/3c1rDRA92G
#ProteinDesign #GenerativeAI #MachineLearning #ComputationalBiology

[Julian Englert @julian_englert]

Today we’re releasing real-world experimental data for over 1000 novel AI-designed proteins on our new platform @proteinbase!