🙏 Huge thanks to my amazing co-authors Gesa Voigt & @ajfaure.bsky.social for their key contributions, and to @benlehner.bsky.social for his great guidance!
📝 Grateful to the reviewers for their thoughtful feedback, and everyone @science.org for helping bring this work to light.

🔬 Our findings suggest models of protein evolution must account for energy couplings and allosteric constraints.
🚀 These insights could accelerate protein engineering—for example, guiding resurfacing to reduce immunogenicity via smarter directed evolution.

🧪 But stability isn’t the whole story—what about function?
🎯 Not all stable core variants could bind the FYN ligand.
🧿 Our findings suggest allostery is to blame: core mutations can subtly impact function—and become catastrophic when they pile up.

🧩 Many amino acid combinations from homologs cores worked when “transplanted” into FYN-SH3—but some didn’t.
🛠️ For the toughest cases, suppressor mutations outside the core rescued stability—thanks to energetic couplings across the protein.

🤖 We fed our large combinatorial mutagenesis datasets into an AI that trains fully interpretable energy models to predict protein variant stability.
🧮 These models accurately distinguished sequence combinations found in nature—from homologs that diverged billions of years ago.

🎲 We randomized the core and surface of the human FYN kinase SH3 domain using reduced amino acid alphabets.
🎰 Thousands of amino acid combinations retained the domain’s stability.
🧱 Even load-bearing amino acids at the core were highly malleable.

🌌 A small protein has as many possible sequences as atoms are in the Universe: 10^78.
🔭 How can we explore and model such a vast universe of possibilities?
🦎 Does that help understanding how evolution found so many stable, functional sequences?

🔗 You can read the paper here: www.science.org/doi/10.1126/...