Pith. sign in

REVIEW 5 major objections 5 minor 53 references

Automating MD simulations for Proteins using Large language Models: NAMD-Agent

T0 review · 5 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read NAMD-Agent turns short natural-language prompts into working protein simulation setups, completing five of seven test systems and preparing them 2-4 times faster than an experienced human.

desk verdict Competent engineering with a genuinely new membrane-builder automation, but the '71.4% accuracy' is really just a no-crash rate; deserving of peer review with validation fixes. read the letter →

arxiv 2507.07887 v1 pith:FQRCSPJJ submitted 2025-07-10 cs.CL cs.CEq-bio.BM

Achuth Chandrasekhar , Amir Barati Farimani This is my paper · ORCID
classification cs.CLcs.CEq-bio.BM
keywords moleculardynamicssimulationautomationlargelanguagemodelsagenticAIretrieval-augmentedgenerationCHARMM-GUINAMDproteinReActagents
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

NAMD-Agent is an automated pipeline that takes a short natural-language prompt about a protein system, generates the configuration files needed to run NAMD, launches a production simulation, and returns standard analyses such as RMSD, RMSF, SASA, radius of gyration, and hydrogen-bond counts. The paper's central claim is that a retrieval-augmented large language model, acting through a ReAct reasoning-and-acting loop and browser automation, can carry out this whole workflow with negligible human intervention. Across seven solution and membrane protein systems, five runs completed successfully, a 71.4% success rate, and the agent's setup time was 2-4 times shorter than an experienced human's hands-on time. The authors present this as evidence that LLM-supervised agents can act as reliable co-pilots for routine molecular dynamics preparation, reducing manual error and enabling parallel setup of multiple systems.

What carries the argument

The load-bearing mechanism is the ReAct loop, in which the LLM alternates between reasoning steps, labeled 'Thought,' and tool-using actions, labeled 'Action,' generating, executing, and revising Python scripts until the workflow completes. A code-aware retrieval-augmented generation stage grounds these scripts in tested templates, API patterns, and parameter settings from a curated repository, which the authors argue reduces hallucinated parameters and speeds up prototyping. Around this loop sits browser automation that operates CHARMM-GUI's web forms, PDBFixer for structure cleanup, NAMD3 as the simulation engine, and trajectory-analysis libraries for RMSD, RMSF, SASA, radius of gyration, and hydrogen-bond counting. This combination lets the model compose and run each step of the pipeline from a natural-language prompt.

What would settle it

Take the same seven PDB systems, run NAMD-Agent, and independently prepare reference systems with the same CHARMM-GUI inputs and NAMD parameters by hand. Compare the automatically generated input decks and the resulting trajectories: if the agent's runs fail to match the reference RMSD, RMSF, SASA, radius of gyration, and hydrogen-bond profiles within tolerances, or if the pipeline passes systems that fail standard checks such as energy-minimization convergence or box-size adequacy, the 71.4% accuracy claim is not supported.

Watch

Extended reading notes

Core claim

The central discovery claimed is that an end-to-end agentic pipeline, not just a chatbot that suggests commands, can physically prepare and execute MD simulations. The agent interprets a user's prompt, uses PDBFixer to repair the input structure, drives CHARMM-GUI through a browser automation session to build a solvated solution or POPC bilayer system, organizes the resulting topology, coordinate, and parameter files, runs a 1 ns NAMD production simulation in the NPT ensemble, and post-processes the trajectory. The paper reports that five of seven attempted setups, two solution and three membrane, produced stable trajectories whose structural profiles matched expected equilibrated behavior, while two membrane runs failed: one from an atypical RMSD trace (1J4N) and one from an undersized membrane XY dimension (1K4C). The authors interpret this 71.4% success rate and the 2-4x setup speedup as evidence that LLM-supervised agents are a feasible route to automating routine MD tasks.

Load-bearing premise

The benchmark's definition of 'successful' is that a 1 ns NAMD run completes and returns RMSD, RMSF, SASA, radius of gyration, and hydrogen-bond plots that look typical of equilibrated systems, with no independent reference set of known-correct inputs or trajectories against which the generated topologies, coordinates, and parameters are checked.

Editorial extensions

If this is right

  • A researcher could hand NAMD-Agent a protein PDB identifier and a plain-English request, and receive a finished NAMD input deck, a completed short simulation, and analysis plots without manually filling input files.
  • The 2-4x setup time reduction makes it practical to prepare several protein systems in parallel on modest hardware, supporting high-throughput studies of mutations, lipid compositions, or solvent conditions.
  • The same agentic architecture should transfer to other CHARMM-GUI builders and other MD engines, since the pipeline's dependencies are the web interface and a code repository, not the specific builder.
  • Until automatic geometry and stability validation is added, occasional failures like the undersized membrane box (1K4C) and the unstable 1J4N run are expected; pre-run checks for box dimensions and equilibration metrics would make the workflow fully hands off.
  • The modular design means a failed run exposes the failing step directly, allowing fast error identification and recovery in future iterations.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The paper's success criterion is internally generated 'typical' profiles, not agreement with reference simulations; a fair accuracy test would compare NAMD-Agent's inputs and outputs against validated CHARMM-GUI/NAMD runs for the same PDBs and quantify profile differences.
  • The 71.4% accuracy is based on seven systems, and two failures were geometric or stability issues a human would likely catch; a larger benchmark with diverse proteins, membrane sizes, and force fields is needed to estimate the true failure rate.
  • The 2-4x speed comparison covers hands-on setup time only; wall-clock comparisons including LLM API calls, CHARMM-GUI server queues, and browser-automation retries may narrow the advantage for short simulations.
  • A concrete reproducibility test would rerun the same prompts several times with controlled randomness; if the model's code revisions vary across runs, the pipeline's outputs may not be reproducible enough for strict method reporting.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 5 minor

Summary. The paper introduces NAMD-Agent, a pipeline that combines the Gemini 2.0 Flash LLM with LlamaIndex-based retrieval-augmented generation (RAG), Python scripting, and Selenium web automation to automate the preparation of NAMD molecular dynamics input files through the CHARMM-GUI web interface. The agent is claimed to accept a short natural-language prompt, generate YAML configuration files, run CHARMM-GUI solution and membrane builders, launch NAMD simulations, and perform post-processing analyses (RMSD, RMSF, SASA, radius of gyration, hydrogen bonds). The authors report 5 successful runs out of 7 setups (71.4%), describe two failures, and claim the agent is 2–4 times faster than an experienced human. The paper is framed as a feasibility demonstration of LLM-driven automation for biomolecular simulation workflows.

Significance. If the claims are properly substantiated, the work would be a useful proof-of-concept for applying LLM agents to complex, multi-step scientific web workflows: it combines code generation, iterative error correction, and browser automation in a domain where input preparation is notoriously error-prone. The paper's strengths include its honest reporting of two failed runs, its clear acknowledgment of limitations (web-interface dependency, hallucination risk, engine specificity), and the public availability of the code. However, the significance is currently limited by the weak validation: the success criterion is operational rather than based on independent ground truth, the benchmark is small and self-selected, and the human-time comparison is under-specified. As a pilot demonstration the contribution is interesting, but the central claims about 'valid' input files and 'negligible human intervention' need stronger evidence.

major comments (5)
  1. [Section 5, Table 1, Section 6] The success criterion is defined operationally as producing RMSD, RMSF, SASA, radius of gyration, and hydrogen-bond profiles 'consistent with typical behavior for equilibrated systems.' This is a self-referential check: a 1 ns trajectory that completes without crashing and yields plausible-looking plots is counted as success, regardless of whether the underlying topology, force field parameters, solvation box size, or equilibration protocol are actually correct. The conclusion in Section 6 that the agent 'generated valid topologies, coordinates, and parameter files' is therefore not supported. I recommend adding an independent validation step: compare the generated inputs against a manually prepared CHARMM-GUI reference (e.g., checksums or content verification of topology/parameter files, matching force-field assignments, and box dimensions), and/or compare simulation observables against a known-good reference simulation or experimental data over a longer production run. Without such ground truth, the 71.4% figure measures 'ran without obvious failure' rather than input correctness.
  2. [Section 6 and Limitations] The abstract and Section 6 claim the pipeline operates 'with negligible human intervention' and is 'fully autonomous,' but the Limitations section states that 'occasional hallucinated parameters (e.g. unsupported thermostat keywords) still arise and require manual oversight,' and item 5 says 'The user is responsible for ensuring that the simulation parameters conform with reality.' These statements are in direct tension. The paper should either quantify the human oversight time per run and include it in the time comparison, or revise the autonomy claims to 'assisted automation with occasional oversight.' As written, the headline claim is stronger than the evidence and the authors' own limitations allow.
  3. [Table 2 and Section 5] The runtime comparison in Table 2 reports a single time for each system with no description of how the human times were measured or the human's level of expertise. Was the human using the same auto_cgui scripts or preparing inputs manually? How many trials were averaged, and what is the run-to-run variance of the agent (the two 1AFO runs differ by 10 minutes: 19 vs 29 minutes)? Without a defined protocol, multiple replicates, and reported medians/ranges, the claim of being '2-4 times faster' is not statistically grounded. I recommend specifying the human baseline (e.g., a graduate student with a given amount of MD experience) and reporting distribution statistics over several agent runs.
  4. [Section 3.5 and Section 3.3] The RAG component, which appears in the paper's title and conclusions, is described only at a high level. The manuscript does not specify the contents of the retrieval corpus (beyond 'curated repository' and the auto_cgui repository), the retrieval mechanism (dense, sparse, or hybrid), the number of retrieved chunks per query, the embedding model, or how the retrieved code is incorporated into the LLM prompt. Since the paper claims that RAG 'reduces hallucinations' and 'grounds the output in real, tested examples,' the contribution is not verifiable or reproducible without these details. I recommend providing concrete implementation details and an ablation showing the effect of removing RAG.
  5. [Section 5, Table 1] The benchmark consists of seven runs, five of which succeed, giving 71.4%. With n=7 and binary outcomes, the 95% confidence interval is approximately 29% to 96%, so the phrase 'a solid accuracy benchmark' is overly strong. Additionally, two runs use the same protein (1AFO) with slightly different box sizes, so the number of distinct systems is effectively six. For a feasibility study this sample size is acceptable, but the claim should be framed as a pilot demonstration rather than a benchmark. Reporting confidence intervals and adding more diverse systems (various sizes, oligomeric states, protein–ligand complexes) would substantially strengthen the paper.
minor comments (5)
  1. [Section 4] The equations for RMSF and Rg contain formatting defects (e.g., the RMSF formula in Section 4.2 appears garbled, lacking proper angle-bracket and power notation; the Rg equation in Section 4.4 has mismatched braces). Please re-typeset them carefully.
  2. [Section 5 and Supplemental Information] There are repeated typos such as 'dimesnions' in Sections 6.3–6.5, and the figure references in Section 5 ('Figures 3–6 and 8') do not match the actual numbering of the main-text figures (Figures 1–3) and the supplemental figures (Figures 4–9). Please renumber all figure references.
  3. [Section 3.6] The code repository URL should be verified and ideally accompanied by a versioned release or DOI. The text refers to 'NAMD_AGENT' in the URL but 'NAMD-Agent' in the title; ensure consistency for reproducibility.
  4. [Section 3.3] The paper does not state the CHARMM-GUI version or the Selenium/browser versions used. Because the automation depends on hard-coded HTML elements and download paths, specifying these versions is essential for reproducing the workflow.
  5. [Section 2.3] The comparison with MDCrow and AutoSolvateWeb is qualitative. Adding a short table comparing supported systems, required human oversight, success rates, and wall-clock time would make the contribution of NAMD-Agent clearer.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: empirical pipeline paper; validation weakness is not a derivation loop.

full rationale

NAMD-Agent is an empirical systems/automation paper rather than a derivation chain: no quantity is predicted from another equation, and the central claim (an LLM agent can drive CHARMM-GUI and NAMD end-to-end) is evaluated by direct execution. The 71.4% accuracy is a literal count of 5/7 runs that completed and produced structural profiles, not a fitted parameter presented as a prediction. The self-referential character of the success criterion—judging input validity by whether the resulting 1 ns trajectory looks 'typical'—is a genuine external-validity weakness, but it does not make the paper's claim equivalent to its inputs by construction; the outputs come from external tools (CHARMM-GUI, Auto CGUI, NAMD3, PDBFixer) that the agent orchestrates. Self-citations [10]-[14] and [39]-[40] are background examples of prior agent/RAG work and are not load-bearing for the pipeline's operation, benchmark, or any forced uniqueness claim. No ansatz is imported via citation and no known result is renamed. Score 0.

Assumptions & free parameters 0 free parameters · 3 assumptions · 0 invented entities

The central claim is an engineering feasibility claim; it rests on the stability of third-party web services and the reliability of an LLM and standard MD tools, not on fitted parameters or new entities.

assumptions (3)
  • domain assumption CHARMM-GUI's web interface and NAMD3 behave as documented and remain stable during automation.
    The whole pipeline is built on Selenium-driving CHARMM-GUI; the paper's Limitation 1 admits hard-coded HTML elements and download paths can break with interface changes.
  • domain assumption A 1 ns trajectory with 'typical' RMSD, RMSF, SASA, Rg, and H-bond profiles indicates a correctly prepared system.
    Used as the success criterion for 5 of 7 runs in Section 5; no independent ground truth or longer equilibration is provided.
  • domain assumption Gemini-2.0-Flash, with the provided RAG corpus, produces sufficiently correct code, and iterative refinement corrects errors.
    The agentic loop's reliability is assumed; the paper's Limitation 3 says hallucinated parameters still arise and require manual oversight.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Automating MD simulations for Proteins using Large language Models: NAMD-Agent." pith.science (2026). https://pith.science/paper/FQRCSPJJ

@misc{pith2026250707887,
  author       = {Pith},
  title        = {Pith review of: Automating MD simulations for Proteins using Large language Models: NAMD-Agent},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FQRCSPJJ}},
  note         = {Machine review of arXiv:2507.07887}
}
read the original abstract

Molecular dynamics simulations are an essential tool in understanding protein structure, dynamics, and function at the atomic level. However, preparing high quality input files for MD simulations can be a time consuming and error prone process. In this work, we introduce an automated pipeline that leverages Large Language Models (LLMs), specifically Gemini 2.0 Flash, in conjunction with python scripting and Selenium based web automation to streamline the generation of MD input files. The pipeline exploits CHARMM GUI's comprehensive web-based interface for preparing simulation-ready inputs for NAMD. By integrating Gemini's code generation and iterative refinement capabilities, simulation scripts are automatically written, executed, and revised to navigate CHARMM GUI, extract appropriate parameters, and produce the required NAMD input files. Post processing is performed using additional software to further refine the simulation outputs, thereby enabling a complete and largely hands free workflow. Our results demonstrate that this approach reduces setup time, minimizes manual errors, and offers a scalable solution for handling multiple protein systems in parallel. This automated framework paves the way for broader application of LLMs in computational structural biology, offering a robust and adaptable platform for future developments in simulation automation.

Figures

Figures reproduced from arXiv: 2507.07887 by the authors.

Figure 1
Figure 1. ReAct-based workflow for molecular simulation setup: The AI agent reasons and performs ac [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Workflow diagram illustrating the automated molecular dynamics simulation pipeline managed by [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Overview of a retrieval-augmented generation (RAG) pipeline for molecular dynamics simulation: [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Structural analysis of the simulation system showing (a) RMSD, (b) RMSF, (c) SASA, (d) Radius [PITH_FULL_IMAGE:figures/full_fig_p027_4.png]
Figure 5
Figure 5. Figure 5: Structural analysis of the simulation system showing (a) RMSD, (b) RMSF, (c) SASA, (d) Radius [PITH_FULL_IMAGE:figures/full_fig_p028_5.png]
Figure 6
Figure 6. Figure 6: Structural analysis of the simulation system showing (a) RMSD, (b) RMSF, (c) SASA, (d) Radius [PITH_FULL_IMAGE:figures/full_fig_p030_6.png]
Figure 7
Figure 7. Figure 7: Structural analysis of the simulation system showing (a) RMSD, (b) RMSF, (c) SASA, (d) Radius [PITH_FULL_IMAGE:figures/full_fig_p031_7.png]
Figure 8
Figure 8. Figure 8: Structural analysis of the simulation system showing (a) RMSD, (b) RMSF, (c) SASA, (d) Radius [PITH_FULL_IMAGE:figures/full_fig_p033_8.png]
Figure 9
Figure 9. Figure 9: Structural analysis of the simulation system showing (a) RMSD, (b) RMSF, (c) SASA, (d) Radius [PITH_FULL_IMAGE:figures/full_fig_p034_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

53 extracted references · 38 canonical work pages

  1. [1]

    Vaswani, N

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, I. Polosukhin, Attention is all you need, Advances in neural information processing systems 30 (2017). 19

    work page 2017

  2. [2]

    McTear, Conversational ai: Dialogue systems, conversational agents, and chatbots, Springer Nature, 2022

    M. McTear, Conversational ai: Dialogue systems, conversational agents, and chatbots, Springer Nature, 2022

    work page 2022

  3. [3]

    Y. Yang, Y. Tan, J. Min, Z. Huang, Automatic text summarization for government news reports based on multiple features, The Journal of Supercomputing 80 (3) (2024) 3212–3228

    work page 2024

  4. [4]

    Van Veen, C

    D. Van Veen, C. Van Uden, L. Blankemeier, J.-B. Delbrouck, A. Aali, C. Bluethgen, A. Pareek, M. Polacin, E. P. Reis, A. Seehofnerová, et al., Adapted large language models can outperform medical experts in clinical text summarization, Nature medicine 30 (4) (2024) 1134–1142

    work page 2024

  5. [5]

    Team, et al., Scaling neural machine translation to 200 languages, Nature 630 (8018) (2024) 841

    N. Team, et al., Scaling neural machine translation to 200 languages, Nature 630 (8018) (2024) 841

    work page 2024

  6. [6]

    Durante, Q

    Z. Durante, Q. Huang, N. Wake, R. Gong, J. S. Park, B. Sarkar, R. Taori, Y. Noda, D. Terzopoulos, Y. Choi, et al., Agent ai: Surveying the horizons of multimodal inter- action, arXiv preprint arXiv:2401.03568 (2024)

    arXiv 2024

  7. [7]

    S. Jo, T. Kim, V. G. Iyer, W. Im, Charmm-gui: a web-based graphical user interface for charmm, Journal of computational chemistry 29 (11) (2008) 1859–1865

    work page 2008

  8. [8]

    J. Lee, X. Cheng, S. Jo, A. D. MacKerell, J. B. Klauda, W. Im, Charmm-gui input generator for namd, gromacs, amber, openmm, and charmm/openmm simulations using the charmm36 additive force field, Biophysical journal 110 (3) (2016) 641a

    work page 2016

Show all 53 references
  1. [9]

    URLhttps://github.com/SeleniumHQ/selenium

    SeleniumHQ, Contributors, SeleniumHQ/selenium: A suite of tools for web browser automation, accessed: 2025-01-12 (2025). URLhttps://github.com/SeleniumHQ/selenium

    2025

  2. [10]

    Jadhav, P

    Y. Jadhav, P. Pak, A. B. Farimani, Llm-3d print: large language models to monitor and control 3d printing, arXiv preprint arXiv:2408.14307 (2024). 20

    2024

  3. [11]

    J. Ock, T. Vinchurkar, Y. Jadhav, A. B. Farimani, Adsorb-agent: Autonomous identifi- cation of stable adsorption configurations via large language model agent, arXiv preprint arXiv:2410.16658 (2024)

    2024 arXiv

  4. [12]

    Raman, C

    A. Raman, C. Merrill, A. George, A. B. Farimani, Llm-drone: Aerial additive manufacturing with drones planned using large language models, arXiv preprint arXiv:2503.17566 (2025)

    2025 arXiv

  5. [13]

    Chaudhari, J

    A. Chaudhari, J. Ock, A. B. Farimani, Modular large language model agents for multi- task computational materials science (2025)

    2025

  6. [14]

    T. Zeng, S. Badrinarayanan, J. Ock, C.-K. Lai, A. B. Farimani, Llm-guided chemi- cal process optimization with a multi-agent approach, arXiv preprint arXiv:2506.20921 (2025)

    2025

  7. [15]

    A. A. Yekeen, O. A. Durojaye, M. O. Idris, H. F. Muritala, R. O. Arise, Chaperong: A tool for automated gromacs-based molecular dynamics simulations and trajectory analyses, Computational and structural biotechnology journal 21 (2023) 4849–4858

    2023

  8. [16]

    Singh, A

    H. Singh, A. Raja, A. Prakash, B. Medhi, Gmx_qk: an automated protein/protein– ligand complex simulation workflow bridged to mm/pbsa, based on gromacs and zenity- dependent gui for beginners in md simulation study, Journal of Chemical Information and Modeling 63 (9) (2023) 2603–2608

    2023

  9. [17]

    Suplatov, Y

    D. Suplatov, Y. Sharapova, V. Švedas, Easyamber: A comprehensive toolbox to au- tomate the molecular dynamics simulation of proteins, Journal of Bioinformatics and Computational Biology 18 (06) (2020) 2040011

    2020

  10. [18]

    M. P. Baumgartner, H. Zhang, Building admiral, an automated molecular dynamics and analysis platform, ACS Medicinal Chemistry Letters 11 (11) (2020) 2331–2335. 21

    2020

  11. [19]

    Carvalho Martins, E

    L. Carvalho Martins, E. A. Cino, R. S. Ferreira, Pyautofep: An automated free energy perturbation workflow for gromacs integrating enhanced sampling methods, Journal of Chemical Theory and Computation 17 (7) (2021) 4262–4273

    2021

  12. [20]

    Campbell, S

    Q. Campbell, S. Cox, J. Medina, B. Watterson, A. D. White, Mdcrow: Au- tomating molecular dynamics workflows with large language models, arXiv preprint arXiv:2502.09565 (2025)

    2025 arXiv

  13. [21]

    R. S. Gadde, S. Devaguptam, F. Ren, R. Mittal, L. Dong, Y. Wang, F. Liu, Chatbot- assisted quantum chemistry for explicitly solvated molecules, Chemical Science (2025)

    2025

  14. [22]

    Gridach, J

    M. Gridach, J. Nanavati, K. Z. E. Abidine, L. Mendes, C. Mack, Agentic ai for scien- tific discovery: A survey of progress, challenges, and future directions, arXiv preprint arXiv:2503.08979 (2025)

    2025 arXiv

  15. [23]

    Liu, Llamaindex

    J. Liu, Llamaindex. URLhttps://github.com/jerryjliu/llama_index

  16. [24]

    S. Yao, J. Zhao, K. N. Yu, Y. Cao, React: Synergizing reasoning and acting in language models, arXiv preprint arXiv:2210.03629 (2022)

    2022 arXiv

  17. [25]

    T. Chen, J. Li, K. Lin, V. Jain, et al., Agents: An open-source framework for building llm-based agents, arXiv preprint arXiv:2308.08155 (2023)

    2023 arXiv

  18. [26]

    Eastman, O

    P. Eastman, O. Team, Pdbfixer,https://github.com/openmm/pdbfixer, accessed: 2025-06-10 (2024)

    2024

  19. [27]

    E. J. Wu, X. Cheng, S. Jo, H. Rui, K. Song, W. Im, Charmm-gui membrane builder toward realistic biological membrane simulations, Journal of Computational Chemistry 35 (27) (2014) 1997–2004

    2014

  20. [28]

    Kern, auto_cgui: Automatic charmm-gui browser interaction with python,https: //github.com/nk53/auto_cgui, version accessed on 2025-06-09, MIT license (2025)

    N. Kern, auto_cgui: Automatic charmm-gui browser interaction with python,https: //github.com/nk53/auto_cgui, version accessed on 2025-06-09, MIT license (2025). 22

    2025

  21. [29]

    J. Lee, D. S. Patel, J. Ståhle, S. J. Park, N. R. Kern, S. Kim, J. Lee, X. Cheng, M. A. Valvano, O. Holst, et al., Charmm-gui membrane builder for complex biological membrane simulations with glycolipids and lipoglycans, Journal of Chemical Theory and Computation 15 (1) (2019) 775–786

    2019

  22. [30]

    J. C. Phillips, R. Braun, W. Wang, J. Gumbart, E. Tajkhorshid, E. Villa, C. Chipot, R. D. Skeel, L. Kale, K. Schulten, Scalable molecular dynamics with namd, Journal of Computational Chemistry 26 (16) (2005) 1781–1802

    2005

  23. [31]

    T. A. Trautman, G. Fiorin, S. Patel, D. J. Hardy, J. E. Stone, R. D. Skeel, J. C. Phillips, K. Schulten, Namd goes quantum: An integrative suite for hybrid simulations, Journal of Chemical Theory and Computation 18 (5) (2022) 3062–3079

    2022

  24. [32]

    Humphrey, A

    W. Humphrey, A. Dalke, K. Schulten, Vmd: visual molecular dynamics, Journal of Molecular Graphics 14 (1) (1996) 33–38

    1996

  25. [33]

    R. T. McGibbon, K. A. Beauchamp, M. P. Harrigan, C. Klein, J. M. Swails, C. X. Hernández, C. R. Schwantes, L.-P. Wang, T. J. Lane, V. S. Pande, Mdtraj: a modern open library for the analysis of molecular dynamics trajectories, Biophysical journal 109 (8) (2015) 1528–1532

    2015

  26. [34]

    Eastman, J

    P. Eastman, J. Swails, J. D. Chodera, R. T. McGibbon, Y. Zhao, K. A. Beauchamp, L.-P. Wang, A. C. Simmonett, M. P. Harrigan, C. D. Stern, et al., Openmm 7: Rapid de- velopment of high performance algorithms for molecular dynamics, PLoS computational biology 13 (7) (2017) e1005659

    2017

  27. [35]

    Lewis, E

    P. Lewis, E. Perez, A. Piktus, F. Petroni, V. Karpukhin, N. Goyal, D. Kiela, V. Moi- seev, S. Riedel, V. Stoyanov, Retrieval-augmented generation for knowledge-intensive nlp tasks, in: Advances in Neural Information Processing Systems, Vol. 33, 2020, pp. 9459–9474. 23

    2020

  28. [36]

    Schick, Y

    T. Schick, Y. Dwivedi-Yu, S. Hosseini, D. Sorokin, N. Houlsby, S. Gelly, Toolformer: Language models can teach themselves to use tools, arXiv preprint arXiv:2302.04761 (2023)

    2023 arXiv

  29. [37]

    S. Lu, D. Liu, W. Wang, et al., Codexglue: A benchmark dataset and open challenge for code intelligence, Empirical Methods in Natural Language Processing (EMNLP) (2021)

    2021

  30. [38]

    Mialon, T

    G. Mialon, T. Scialom, P. Gallinari, J. Staiano, Augmented language models: a survey, arXiv preprint arXiv:2302.07842 (2023)

    2023 arXiv

  31. [39]

    Chandrasekhar, J

    A. Chandrasekhar, J. Chan, F. Ogoke, O. Ajenifujah, A. B. Farimani, Amgpt: A large language model for contextual querying in additive manufacturing, Additive Manufac- turing Letters 11 (2024) 100232

    2024

  32. [40]

    Chandrasekhar, O

    A. Chandrasekhar, O. B. Farimani, O. T. Ajenifujah, J. Ock, A. B. Farimani, Nanogpt: A query-driven large language model retrieval-augmented generation system for nan- otechnology research, arXiv preprint arXiv:2502.20541 (2025)

    2025 arXiv

  33. [41]

    R. S. Gadde, S. Devaguptam, F. Ren, et al., Chatbot-assisted quantum chemistry for explicitly solvated molecules, Chemical Science (2025)

    2025

  34. [42]

    M. Chen, J. Tworek, H. Jun, et al., Evaluating large language models trained on code, in: NeurIPS, 2021

    2021

  35. [43]

    Zhang, et al., Code llama: Open foundation models for code, arXiv preprint arXiv:2308.12950 (2023)

    C. Zhang, et al., Code llama: Open foundation models for code, arXiv preprint arXiv:2308.12950 (2023)

    2023 arXiv

  36. [44]

    Kabsch, A solution for the best rotation to relate two sets of vectors, Acta Crystal- lographica Section A 32 (1976) 922–923.doi:10.1107/S0567739476001873

    W. Kabsch, A solution for the best rotation to relate two sets of vectors, Acta Crystal- lographica Section A 32 (1976) 922–923.doi:10.1107/S0567739476001873

    1976 doi

  37. [45]

    Karplus, J

    M. Karplus, J. A. McCammon, Molecular dynamics simulations of biomolecules, Nature Structural Biology 9 (2002) 646–652.doi:10.1038/nsb0902-646. 24

    2002 doi

  38. [46]

    Amadei, A

    A. Amadei, A. B. Linssen, H. J. C. Berendsen, Essential dynamics of proteins, Pro- teins: Structure, Function and Genetics 17 (4) (1993) 412–425.doi:10.1002/prot. 340170408

    1993 doi

  39. [47]

    Seeliger, B

    D. Seeliger, B. L. de Groot, Protein thermostability calculations using alchemical free energy simulations, Biophysical Journal 98 (10) (2010) 2309–2316.doi:10.1016/j. bpj.2010.01.051

    2010 doi

  40. [48]

    B. Lee, F. M. Richards, The interpretation of protein structures: Estimation of static accessibility, Journal of Molecular Biology 55 (3) (1971) 379–400.doi:10.1016/ 0022-2836(71)90324-X

    1971

  41. [49]

    Shrake, J

    A. Shrake, J. A. Rupley, Environment and exposure to solvent of protein atoms. lysozyme and insulin, Journal of Molecular Biology 79 (2) (1973) 351–371.doi: 10.1016/0022-2836(73)90011-9

    1973 doi

  42. [50]

    P. J. Flory, Statistical Mechanics of Chain Molecules, Interscience, New York, 1969

    1969

  43. [51]

    J. E. e. Kohn, Random-coil behavior and the dimensions of chemically unfolded proteins, Proceedings of the National Academy of Sciences USA 101 (34) (2004) 12491–12496. doi:10.1073/pnas.0403643101

    2004 doi

  44. [52]

    E. N. Baker, R. E. Hubbard, Hydrogen bonding in globular proteins, Progress in Biophysics and Molecular Biology 44 (1984) 97–179.doi:10.1016/0079-6107(84) 90007-5

    1984 doi

  45. [53]

    I. K. McDonald, J. M. Thornton, Satisfying hydrogen bonding potential in proteins, Journal of Molecular Biology 238 (5) (1994) 777–793.doi:10.1006/jmbi.1994.1334. 25 Supplemental Information All of these experiments were run for 1ns at a pH of 7.0 using the NPT ensemble. 6.1. ...

    1994

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.