3️⃣ The Flywheel: Science-as-a-Service The true power lies in the Community-Scale Scientific Flywheel. It’s a coupled online-offline loop that drives continuous improvement: 🔁 Online Execution → Agents execute at scale, generating rich execution traces and validation outcomes. 🔁 Offline Refinement → Traces propagate backward to update models, orchestration policies, and knowledge bases. Execution improves knowledge. Knowledge improves execution. This is the path to Science-as-a-Service: where scientific production is continuously improved by shared infrastructure. 📄 Read the full manifesto: j1q.cn/aD1aHSES #Bohrium #SciMaster #ScienceAsAService #DeepModeling

📄 #POTD: Today’s paper presents DPDispatcher, the task-scheduling engine behind @DPTechnology and modern AI-for-Science workflows. It is the middleware that keeps large-scale modeling, active learning, QM labeling, and MD campaigns reliable across heterogeneous HPC systems. DPDispatcher provides a clean Python framework for launching and managing thousands of jobs that span GPUs and CPUs. It removes the fragility of ad-hoc scripts and handles the scheduler complexity that often slows down scientific automation. ✨ What stands out ✅ Client-side metascheduler design (Context/Machine split) ✅ Persistent job state with automatic retry & resume ✅ Unified interface for Slurm, PBS, LSF, SGE, SSH ✅ Safe handling of large job ensembles across clusters ✅ Proven as the backbone for 10 scientific workflows (DP-GEN, CatFlow, DP-TI, ChecMatE, APEX, RiD-kit, mech2d…) For modern AI-for-Science, workflow reliability is as critical as the models themselves. DPDispatcher is now a key part of both the #Bohrium platform and the DP Technology ecosystem. 📑 Full paper below 👇 #AIforScience #HPC #DeepModeling #ComputationalChemistry #DPTechnology #JCIM

Built by Bohrium, DeepModeling and global researchers, SciencePedia fosters an open contributor ecosystem: 🔬Experts ensure rigor 🌍Community advances knowledge ⭐Credits track contributors’ work 💻Open source keeps science transparent 🔧Join us: github.com/deepmodeling/scie…

🚀 New Insights from DeePMD-kit! 🚀 Check out the latest Nature Chemistry article where DeePMD-kit reveals a dual-gated mechanism of proton transport in water, advancing our understanding of this vital process. 🔗 Read more: doi.org/10.1038/s41557-024-0… #DeepModeling #DeePMD #AI4S

🚨 Update on our Summer 2024 #DataDriven #ManyBody Tour 🚨 Today, Francesco visited @TheDPTechnology to learn about the latest developments in @DeepModeling #machinelearning models and discuss the connections between #neuralnetwork potentials and our MB-pol and MB-nrg #datadriven #manybody potentials. Key references about MB-pol: 👉 doi.org/10.1021/ct400863t 👉 doi.org/10.1021/ct500079y 👉 doi.org/10.1021/ct5004115 👉 doi.org/10.1021/acs.jctc.3c0… 👉 doi.org/10.1021/ct501131j 👉 doi.org/10.1021/jacs.6b00893 👉 doi.org/10.1021/jacs.7b03143 👉 doi.org/10.1021/acs.jpclett.… 👉 doi.org/10.1021/acs.jpclett.… 👉 nature.com/articles/s41467-0… Key references about MB-nrg: 👉 doi.org/10.1021/acs.jctc.6b0… 👉 doi.org/10.1063/1.4993213 👉 nature.com/articles/s41557-0… 👉 doi.org/10.1021/acs.jpclett.… 👉 doi.org/10.1021/acs.jctc.9b0… 👉 doi.org/10.1021/acs.jpcb.0c0… 👉 doi.org/10.1021/acs.jpcb.2c0… 👉 doi.org/10.1021/acs.jpcb.2c0… 👉 doi.org/10.1021/acs.jctc.2c0… 👉 doi.org/10.1021/acs.jctc.3c0… @UCSanDiego @UCSDPhySci @UCSDChemBiochem @HDSIUCSD @SDSC_UCSD

🚀 We're thrilled to launch the #OpenLAM initiative! 🌟 With the new DPA-2 model, we're ready to "Conquer the Periodic Table". Join us in this exciting journey! Learn more: aissquare.com/openlam #DeepPotential #opensource #DeepModeling

Excited to share that Uni-Mol has joined the DeepModeling community! 🚀 Our 3D molecular representation framework is revolutionizing molecular design. Check out our GitHub for more details and to contribute: github.com/deepmodeling/Uni-… #unimol #deepmodeling #opensource

Check out our new @J_A_C_S paper. @DeepModeling based simulations and theoretical analyses show a trend nanobubble dissolution akin to experimental observations, highlighting the restricted dissolution of bulk nanobubbles in alkaline conditions.

Ever wondered if we could model #water to autoionize and correctly predict pH = 7? 🤔 Well, now we can! 😎 In our latest @ChemRxiv preprint, we introduce a #deepneuralnetwork potential trained on density-corrected #DFT that predicts the autoionization constant of water to be Kw = 1.23 × 10^-14! 🚀 Read more: doi.org/10.26434/chemrxiv-20… What’s new? 👉 Building on our previous work with DC-SCAN (nature.com/articles/s41467-0…), we’ve trained a @DeepModeling potential on a diverse dataset of neutral and ionized #water configurations calculated at the DC-r2SCAN level of theory. 👉 Using enhanced sampling #compchem simulations with @plumed_org, we’ve mapped the free-energy landscape of water autoionization, which has allowed us to calculate Kw. Our analyses highlight the critical role of nuclear quantum effects and the Grotthuss mechanism in setting the pH of neutral water to 7. Not yet a fully transferable #datadriven #manybody potential like MB-pol, but stay tuned! @UCSanDiego @UCSDPhySci @UCSDChemBiochem @HDSIUCSD @SDSC_UCSD

👉If you like Domain-Driven Design but Model-Driven Design remains an enigma or simply theoretical, feel free to take your ticket to become a specialist. 🎯With @kenny_bass, we are preparing a dream training session for you. lnkd.in/eYz2k3EZ #dddesign #deepmodeling #suppledesign #modeldrivendesign #strategicdesign #tacticaldesign

I am happy to share this good news with my friend @kenny_bass . We'll speak about Deep Dive into Model-Driven Design! #dddesign #deepmodeling #stuppledesign

DeepModeling posted OpenLAM 2024 Q0 report: deepmodeling.com/blog/openla…

We are preparing an unprecedented session with @kenny_bass and me, illustrating #suppledesign and #deepmodeling live. These two subjects are covered at length in the blue book. We are excited to present this ambitious subject to #kandddinsky #DDDesign.

Exciting in-person AI for Science event by @DeepModeling on @HKUniversity campus @ May 25, check it out here -- deepmodeling.typeform.com/hk…

Dozens of #AI developer communities, including #Linux, Red Hat and MindSpore, attended the Global AI Developer Conference 2023 in #Shanghai. LF AI & Data China Community, CSDN and DeepModeling are among the first batch of open-source AI communities landed in the Lingang New area.

Accepted in @JChemPhys! We explore the prospect of combining the accuracy of our #datadriven #manybody MB-pol potential with the efficiency of @DeepModeling #NeuralNetworks to scale up #compchem simulations of water over the whole phase diagram. 👉 doi.org/10.1063/5.0142843 1/6

Congratulations Pablo!

Nice work, Pablo and team! Congratulations 👏

Our paper on ice nucleation using first principles MD simulations driven by an ML model just appeared online in PNAS. 300,000 atoms with ab initio accuracy! Thanks @snsf_ch @doescience @CcscCsi @DeepModeling @plumed_org for making this possible. pnas.org/doi/full/10.1073/pn…

And thanks @zkgoldsmith for a wonderful tutorial on Quantum Espresso!