Diego E. Kleiman
@kleimandiego.bsky.social
Ph.D. candidate in Biophysics Shukla Group | University of Illinois | Molecular dynamics simulations 🧬 + ML 🤖
7/ I'd like to thank everyone in @shuklagroup.bsky.social as well as The Molecular Sciences Software Institute for supporting this work!
August 28, 2025 at 7:47 PM
7/ I'd like to thank everyone in @shuklagroup.bsky.social as well as The Molecular Sciences Software Institute for supporting this work!
6/ A handy visualization tool allows you to get an interactive view of the results on a browser.
August 28, 2025 at 7:47 PM
6/ A handy visualization tool allows you to get an interactive view of the results on a browser.
5/ The code openly available on GitHub (t.co/R2n5G5rJzb) and a Colab Notebook is available to run online (t.co/aF6eD8F0X6)
August 28, 2025 at 7:47 PM
5/ The code openly available on GitHub (t.co/R2n5G5rJzb) and a Colab Notebook is available to run online (t.co/aF6eD8F0X6)
4/ ⚡ Key highlights:
• Predicts dynamic contacts in seconds
• Matches/outperforms other methods in its task
• Generalizes to unseen systems (membrane proteins, de novo designs, multimers)
• Enables automatic CV selection for MSM construction
• Predicts dynamic contacts in seconds
• Matches/outperforms other methods in its task
• Generalizes to unseen systems (membrane proteins, de novo designs, multimers)
• Enables automatic CV selection for MSM construction
August 28, 2025 at 7:47 PM
4/ ⚡ Key highlights:
• Predicts dynamic contacts in seconds
• Matches/outperforms other methods in its task
• Generalizes to unseen systems (membrane proteins, de novo designs, multimers)
• Enables automatic CV selection for MSM construction
• Predicts dynamic contacts in seconds
• Matches/outperforms other methods in its task
• Generalizes to unseen systems (membrane proteins, de novo designs, multimers)
• Enables automatic CV selection for MSM construction
3/ 🧠 Built on ESMFold, ESMDynamic is trained on both experimental structure ensembles and MD simulations. It captures transient interactions across diverse protein systems.
August 28, 2025 at 7:47 PM
3/ 🧠 Built on ESMFold, ESMDynamic is trained on both experimental structure ensembles and MD simulations. It captures transient interactions across diverse protein systems.
2/ Most structure prediction models output static conformations, but proteins are dynamic. We introduce ESMDynamic, a deep learning model that predicts dynamic contact probability maps directly from sequence, no MSAs required.
August 28, 2025 at 7:47 PM
2/ Most structure prediction models output static conformations, but proteins are dynamic. We introduce ESMDynamic, a deep learning model that predicts dynamic contact probability maps directly from sequence, no MSAs required.