iiinova — where your digital self keeps evolving

Krishna Shrinivas @shrinivaslab.bsky.social · 07/07/2026

This is now published in PNAS! And double-treat, we also made the cover :) www.pnas.org/doi/10.1073/...

Krishna Shrinivas @shrinivaslab.bsky.social · 08/06/2026

Design 3: localization in living cells 🔬 We designed IDPs to selectively enrich (client) or deplete (excluder) from FUS LC condensates & tested in living cells with @sneadlab.bsky.social. Great work by co-first authors Neha Tyagi & Jackson Boodry, with Vita Chou & Wilton Snead!

020

Krishna Shrinivas @shrinivaslab.bsky.social · 08/06/2026

Design 2: emergent condensate properties 🫧 By jointly optimizing IDP networks, we designed multicomponent mixtures that form condensates with: • layered architectures (like the nucleolus) • compositional specificity (who's in/out) • RNA-dependent reorganization

120

Krishna Shrinivas @shrinivaslab.bsky.social · 08/06/2026

Design 1: IDPs as complex physical signal processors 🧮 We designed IDPs that sense/compute over cues — a single sequence can act as: • threshold detector (phosphorylation counting) • bandpass filter (respond only to intermediate signals) • logic gate (XOR, OR, NAND) over 2 cues

100

Krishna Shrinivas @shrinivaslab.bsky.social · 08/06/2026

Our key idea: use ML to invert emerging physics-based models (Dimer, FINCHES, RPA) for de novo IDP design. Our framework uses gradient-based sequence design for target properties and is: fast: proteome-scale (10,000 seq) < 1 min on GPU flexible: change loss, change property

100

Krishna Shrinivas @shrinivaslab.bsky.social · 08/06/2026

Intrinsically disordered proteins (IDPs) flit between diverse shapes and interact in fuzzy ensembles whose collective properties shape function. Q: Can we rationally design emergent properties of IDPs? Our newest preprint takes a stab at this 🧵 biorxiv.org/content/10.6...

151

Krishna Shrinivas @shrinivaslab.bsky.social · 10/10/2025

Our framework: We bridge machine learning & statistical physics to directly invert molecular simulations to design IDPS and engineer examples that: 🌀 form loops & linkers with tuned flexibility ⚡ sense salt, temperature, or phosphorylation stimuli 🤝 bind disordered targets like FUS or Whi3

020

Krishna Shrinivas @shrinivaslab.bsky.social · 10/10/2025

The problem: AI tools like AlphaFold & ProteinMPNN accelerate design of stable protein folds by inverting the sequence-structure map. But IDPs don't have 1 shape - they occupy a huge ensemble of shapes. Physics simulations are good models to generate ensembles but hard to design/invert over!

120

Krishna Shrinivas @shrinivaslab.bsky.social · 10/10/2025

Happy to share our latest in @natcomputsci.nature.com led by (amazing) Ryan Krueger + colab w M. Brenner! We introduce a framework to directly design intrinsically disordered proteins (IDPs) from physics-based simulations. 🧬 doi.org/10.1038/s435... 📰 www.mccormick.northwestern.edu/news/article...

2268

Krishna Shrinivas @shrinivaslab.bsky.social · 22/09/2025

Led by the amazing Aidan Zentner, with contribs from Ethan Halingstad, and in collab with Cameron Chalk, Michael Brenner, @amurugan.bsky.social, and Erik Winfree. For a more fun overview, see Erik's version of the abstract www.dna.caltech.edu/DNAresearch_... :) (2/2)

040

Krishna Shrinivas @shrinivaslab.bsky.social · 22/09/2025

Preprint! Inspired by condensates that form on specific DNA, we ask: can we design multicomponent fluids to form distinct condensates on diff. surfaces? i.e. perform classification by condensation ⚛️ 💻 exploiting phase transitions beyond compartmentalization! arxiv.org/abs/2509.08100 (1/2)

1166

Krishna Shrinivas @shrinivaslab.bsky.social · 12/09/2025

Our work highlighted in @science.org by L. Bryan Ray! www.science.org/doi/10.1126/...

0120

Krishna Shrinivas @shrinivaslab.bsky.social · 04/09/2025

Another surprise: core & shell proteins don’t mix well (they’re immiscible, like oil & water). Putting these observations together in simulations suggests 🖥️⚛️: competition for RNA + immiscibility naturally push proteins to form different layers, even if they individually like the same parts of RNA.

100

Krishna Shrinivas @shrinivaslab.bsky.social · 04/09/2025

We use paraspeckles as a model to study this question. Paraspeckles are built around a non-coding RNA NEAT1 whose middle regions are in the core and 5’/3’ ends on the shell layer - each layer also recruiting different proteins. (2/6)

100