Docker Demoprojekt

Docker Demoprojekt
Default BOINC server banner displayed on Docker Demoprojekt
Bacteriophage T4 lysozyme (PDB 184L), one of the five benchmark protein-ligand complexes docked in the project's GNINA test applications
Project
Status	Active; Temporary alpha test server - scheduled to be deleted
Category	Software testing; Biochemistry / Drug Discovery (test workload)
Compute	CPU & GPU
Requires	BOINC client 8.2+, Podman (4.1+ for GPU) or Docker, NVIDIA driver 570+ and NVIDIA Container Toolkit (for GPU tasks)
Development
Developer	Martin Blenn (koschi)
Author	Martin Blenn (boinc-stack and boinc-apps); David Ryan Koes et al. (GNINA)
Sponsor	Planet 3DNow! and Rechenkraft.net e.V.
Maintainer	Martin Blenn (koschi)
Initial release	October 28, 2026  (0 years ago)
Completed	No
Repository	https://github.com/MBlenn/boinc-stack (server stack) https://github.com/MBlenn/boinc-apps (container apps)
Software
Written in	Shell, Python, C++ (server stack and wrapper); C++, CUDA, Python (GNINA)
Operating system	Linux (x86_64)
Size	~4.62 GB (GNINA container image); ~410 KB (shared test inputs); ~15 GB BOINC disk quota recommended
BOINC statistics
Stats as of	November 5, 2026  (0 years ago)
Performance	2.43 TFLOPS
Active users	5
Total users	5
Active hosts	7
Total hosts	7
Analytics
GPU performance	25,790 GigaFLOPS
CPU performance	4 GigaFLOPS
Metadata
Website	https://boinc.kerbodyne.com/dockerdemo/
License	MIT (boinc-stack and boinc-apps); Apache-2.0 / GPL-2.0 (GNINA); LGPL-3.0-or-later (BOINC)

Docker Demoprojekt (German for Docker Demo Project) is an alpha-stage volunteer distributed computing test project running on the BOINC (Berkeley Open Infrastructure for Network Computing) platform. Developed and administered by German volunteer computing contributor Martin Blenn (known on forums as koschi and on GitHub as MBlenn), the project is part of a joint infrastructure initiative by the German distributed computing communities Planet 3DNow! and Rechenkraft.net e.V. (the non-profit association behind Yoyo@home and RNA World (beta)).

Much like dedicated quality-assurance servers such as BOINC Alpha Test, Albert@Home, and RALPH@home, Docker Demoprojekt does not perform novel production research of its own. Instead, it serves as an ephemeral technical proving ground for boinc-stack and boinc-apps, an open-source framework for hosting multiple containerized BOINC servers and distributing containerized CPU and GPU scientific applications via Podman and Docker. To exercise the complete computational pipeline under realistic conditions, the server distributes short GPU verification tasks alongside benchmark protein-ligand docking workunits powered by GNINA, a deep-learning molecular docking suite. According to project developers and testers, Docker Demoprojekt is a temporary testbed intended for small-scale alpha verification and is scheduled to be deleted once testing of the container workflow is complete.

History and background

Collaboration between Planet 3DNow! and Rechenkraft.net

Docker Demoprojekt grew out of a multi-year partnership between two of Germany's longest-running volunteer computing communities: Rechenkraft.net e.V., a registered non-profit scientific association headquartered in Marburg, and the distributed computing team of Planet 3DNow! (including its community hardware collective, the P3D-Cluster). In February 2026, Rechenkraft.net chairman Dr. Michael H.W. Weber reported that a new CitizenPharma project server sponsored by Planet 3DNow! had completed hardware stress-testing (including a world-record GPU run on PrimeGrid's AP26/27 search) and that an initial prototype of the latest BOINC server software had been brought up inside a Docker/Podman container.

On 13 October 2026, Planet 3DNow! coordinator thorsam opened a call for Podman and Docker alpha testers on the Planet 3DNow! distributed computing forum, outlining two concrete goals for the joint container infrastructure:

• CitizenPharma biomedical campaigns: Rechenkraft.net and Planet 3DNow! prepared to transition into the operational phase of CitizenPharma, an open-source citizen-science drug discovery and biomedical computing platform. In this division of labor, the P3D-Cluster supplies server hardware in Hannover after a year-long reliability burn-in test, while Rechenkraft.net develops the scientific software stack, which explicitly includes GNINA, AutoDock Vina, Smina, and OpenMM. Two initial research campaigns were announced: one focused on stomach cancer (Magenkrebs) and another targeting Lyme disease (Borreliose).
• Hosting and containerizing SiDock@home: The joint RKN and P3D team also arranged to provide reliable European Union server hosting in Hannover (with a backup system planned in Braunschweig) for the Slovenian drug-discovery project SiDock@home, migrating its infrastructure onto Podman containers running on a dual AMD EPYC 7V12 server (128 cores / 256 threads, 512 GB ECC DDR4 RAM, and hardware RAID-1 NVMe/SAS SSD arrays) as SiDock@home expanded from SARS-CoV-2 screening toward Ebola research.

Development of boinc-stack and alpha launch

On 25 October 2026, developer Martin Blenn (koschi) evaluated the official upstream BOINC/boinc-server-docker images and noted that their underlying software stack (Ubuntu 18.04, BOINC Server 1.2.0, MariaDB 10.3.8, and PHP 7.0 from 2018) was too outdated for modern public internet deployment. Rather than isolating multiple BOINC projects inside heavy full virtual machines, Blenn used AI-assisted software engineering (Claude) to design a lightweight, reproducible multi-project container stack from scratch, publishing the initial version of boinc-stack under the MIT License on 26 October 2026 and the companion boinc-apps repository on 28 October 2026.

Blenn stood up Docker Demoprojekt on a personal test machine linked to a virtual private server (boinc.kerbodyne.com) and ran the first internal container workunits on 27 October 2026. After Planet 3DNow! member Stony666 completed the first external alpha tests, Blenn published step-by-step Linux host setup instructions and opened https://boinc.kerbodyne.com/dockerdemo/ to fellow team alpha testers on 1 November 2026. Over the next four days, testers including Stony666, FritzB, and thorsam exercised the project across diverse Linux distributions and multi-GPU systems, uncovering several upstream edge cases in BOINC's docker_wrapper and Podman's Container Device Interface (CDI) handling that were patched in rapid application releases up to version 1.09 (gnina and gnina_cpu) and version 1.10 (gpucheck).

Discovery by the wider BOINC community

Because several alpha testers simultaneously ran the non-CPU-intensive monitoring project WUProp@Home on their Linux hosts, WUProp@Home automatically cataloged the new project name (Docker Demoprojekt) and its three application titles, propagating them to third-party aggregation sites such as Free-DC. On 2 November 2026, user mmonnin asked about the mysterious project on the WUProp@Home forums after noticing GNINA docking (CPU, test), GNINA docking (test), and GPU check listed without public project URLs, followed by a URL inquiry from veteran cruncher mikey on 3 November 2026. On 4 November 2026, user toggleton clarified the project's private alpha nature:

As the title says it is so far a test of setting up the docker tasks for a later project by planet3dnow and rechenkraft teams. Don't think it is for the public yet(the tasks are just docker image setup und very short test compute). The project will be deleted once the test finishes. So ignore it for now. Would guess it will have an announcement when more beta tester are wanted.

— toggleton on the WUProp@Home message boards, 4 November 2026

Timeline

Date	Event
4 February 2026	Rechenkraft.net e.V. reports that the Planet 3DNow!-sponsored CitizenPharma server is operational and running a prototype containerized BOINC server.
13 October 2026	Coordinator thorsam announces the next phase of the Rechenkraft.net and Planet 3DNow! collaboration (stomach cancer, Lyme disease, and SiDock@home Podman migration) and recruits alpha testers.
26 October 2026	Martin Blenn (MBlenn / koschi) publishes the initial open-source release of boinc-stack on GitHub.
27 October 2026	First internal test workunit (gpucheck) is executed on the Docker Demoprojekt server.
28 October 2026	Companion repository MBlenn/boinc-apps is created with the containerized gnina docking test application and RMSD validator.
1 November 2026	https://boinc.kerbodyne.com/dockerdemo/ opens to Planet 3DNow! alpha testers with gpucheck, gnina, and gnina_cpu applications.
2 November 2026	Updated application versions gpucheck 1.10 and gnina / gnina_cpu 1.09 are deployed; project apps appear in WUProp@Home and Free-DC listings.
4 November 2026	Community note on WUProp@Home confirms the server is a temporary pre-beta container test by Planet 3DNow! and Rechenkraft.net that will be deleted after testing.

Server and container architecture

Comparison between running an application directly on a host operating system versus inside an isolated container sharing the host Linux kernel, the execution model tested by Docker Demoprojekt.

Multi-project Podman server stack

Docker Demoprojekt runs BOINC Server release 1.6.2 (database schema version 27030) built from official BOINC sources atop Debian 13 (Trixie). In the boinc-stack design, a single Linux server can host $1$ to $N$ independent BOINC projects without root-level cross-access:

• Unprivileged user isolation: Each project runs under its own Linux account (boinc-<project>) managed via rootless Podman and systemd Quadlet container units, optionally backed by per-project ZFS datasets providing transparent compression, snapshots, and quotas.
• Three disposable containers per project: Every hosted project is split into <project>-db (MariaDB 12.3 LTS), <project>-web (Apache HTTP Server, PHP, BOINC scheduler, and upload handler), and <project>-daemons (feeder, transitioner, file_deleter, validators, and assimilators; visible as host dockerdemo-daemons on the status page).
• VPS edge proxy and WireGuard tunnel: For Docker Demoprojekt, public HTTPS traffic to boinc.kerbodyne.com terminates at a Caddy reverse proxy on an external virtual private server (handling automatic Let's Encrypt TLS certificates) and forwards requests across an encrypted Tailscale (WireGuard) tunnel to a second Caddy instance on the backend BOINC host. This topology avoids opening inbound ports on residential or laboratory firewalls while allowing multiple projects to share a single public IP address.
• Offloaded container image delivery: Continuous integration workflows on GitHub Actions build the base image (ghcr.io/mblenn/boinc-app-base:trixie) and scientific application images (ghcr.io/mblenn/boinc-app-gnina:1.3.3-2, roughly 4.62 GB) and publish them publicly on the GitHub Container Registry (GHCR). Volunteer clients pull the multi-gigabyte container layers anonymously from GHCR once per application version, sparing the BOINC project server's storage and outbound bandwidth.

Official logo of Docker. Docker Demoprojekt tests BOINC's native container application architecture using rootless Podman and Docker on Linux hosts.

BOINC wrapper and server patches

Whereas BOINC Central introduced Docker support through BOINC Universal Docker App (BUDA) workflows, Docker Demoprojekt specifically tests native BOINC 1.6 container delivery using docker_wrapper on rootless Podman Linux clients. Alpha testing uncovered several subtle interactions between BOINC, rootless Podman, and multi-GPU Linux systems, resulting in ten custom patches maintained in boinc-stack:

1. Zero-RUN client Dockerfiles: Because the boinc-client daemon runs as an unprivileged system service without a systemd user login session, executing RUN instructions during podman build on volunteer hosts fails with D-Bus authorization errors. Client Dockerfiles therefore contain only FROM, WORKDIR, and CMD directives layered atop pre-built GHCR images.
2. Podman 4.x CDI GPU passthrough: Older Podman releases such as version 4.9 on Ubuntu 24.04 and Linux Mint 22 rejected Docker's --gpus flag; docker_wrapper was patched to pass NVIDIA GPUs via Container Device Interface (--device nvidia.com/gpu=all).
3. Per-task GPU pinning on multi-GPU hosts: Early wrapper builds allowed concurrent GPU tasks on dual-GPU machines to pile onto GPU 0. Patch 0007 exports CUDA_VISIBLE_DEVICES inside docker_wrapper to match the specific GPU index assigned by the BOINC client (tested on mixed GeForce RTX 4070 Ti and RTX 5070 Ti hosts alongside projects such as Moo! Wrapper and Eternity II).
4. Accurate /proc memory and CPU accounting: Under rootless Podman, podman stats reported the combined memory footprint of all active BOINC containers together, causing 64 GB workstations (such as an AMD Ryzen 9 7950X host) to falsely stall tasks with "Waiting for memory" after 8 to 10 tasks. Patches 0002 and 0010 read per-container CPU time and resident working-set size (WSS) directly from /proc.
5. Fail-fast startup and validator fixes: Additional patches detect failed container launches immediately instead of looping until timeout (0009), strip trailing whitespace in BOINC 1.6.2's script_validator --init_script parser (0001), and automatically prune superseded container images (0004).

Applications and scientific methods

Schematic and three-dimensional surface illustration of protein-ligand molecular docking, in which a small-molecule ligand is sampled and scored inside a target protein's binding pocket.

As of November 2026, Docker Demoprojekt hosts three Linux x86_64 container applications managed via the boinc-app deployment tool:

Application title	Internal name	Version	Plan class	Resources per task	Opt-in beta	Average computing (5 Nov 2026)
GPU check	gpucheck	1.10	docker_nvidia	1 NVIDIA GPU + CPU	No	0 GigaFLOPS
GNINA docking (GPU, test)	gnina	1.09	docker_nvidia_cpu1	1 NVIDIA GPU + 1 CPU core	Yes	25,790 GigaFLOPS
GNINA docking (CPU, test)	gnina_cpu	1.09	docker_cpu1	1 CPU core (no GPU reserved)	Yes	4 GigaFLOPS

GPU check

GPU check (gpucheck) is a lightweight diagnostic application distributed automatically to newly attached hosts (MAX_IN_PROGRESS=1). Completing in roughly three seconds once the base container is cached, it executes nvidia-smi inside the container and returns the detected GPU model, driver version, VRAM capacity, CUDA compute capability, and BOINC-assigned device index. If a host returns "nvidia-smi: not available", the container ran successfully on the CPU, but host-to-container GPU passthrough (NVIDIA Container Toolkit or CDI configuration) requires adjustment before enabling scientific GPU workloads.

GNINA molecular docking test applications

Three-dimensional ribbon structure of human glutathione S-transferase P1-1 (PDB entry 10GS), one of the five reference receptor-ligand systems used by Docker Demoprojekt's GNINA test suite.

Once a volunteer verifies their setup with GPU check and enables "Run test applications?" in their project preferences, the scheduler dispatches tasks for GNINA docking (GPU, test) (gnina) and GNINA docking (CPU, test) (gnina_cpu). Both applications package GNINA 1.3.3 (linked with CUDA 12 and cuDNN 9 libraries extracted from NVIDIA runtime wheels), an open-source deep-learning molecular docking program created by David Ryan Koes and collaborators as a fork of smina and AutoDock Vina.

Whereas earlier BOINC drug-screening projects such as Docking@Home, SiDock@home, Rosetta@home, and World Community Grid relied primarily on classical force fields or empirical scoring functions, GNINA augments Markov chain Monte Carlo (MCMC) conformational sampling on the CPU with ensembles of three-dimensional convolutional neural networks (CNNs) that evaluate voxelized atomic density grids of the protein-ligand binding site. On a multi-core workstation with a single NVIDIA GPU, the two distinct plan classes allow BOINC to run one GPU-accelerated gnina task (using CNN-intensive refinement, metrorescore, or metrorefine modes) alongside CPU-only gnina_cpu tasks (using --no_gpu and final pose rescore mode) on the remaining processor cores.

Each batch generated by jobs.py draws a random integer seed $s\in \{1,2,\dots ,2^{31}-1\}$, generates up to nine binding poses (--num_modes 9 --cpu 1) across five well-characterized benchmark systems from GNINA's test suite (184l, bacteriophage T4 lysozyme; 10gs, human glutathione S-transferase P1-1; 3rod; grid; and noelem), and distributes three distinct job types:

• Redocking (~60% of tasks): Docks a native crystal ligand back into its cognate receptor structure (refinement mode on GPU, rescore mode on CPU) so accuracy against the known experimental crystal structure can be verified automatically.
• Cross-docking (~30% of tasks): Docks a ligand from system $A$ into receptor $B$ using randomly selected built-in GNINA 1.3 CNN models (such as crossdock_default2018, general_default2018_1, redock_default2018_1_3, or knowledge-distilled 1.3 ensembles) and weighted GPU scoring modes (refinement 50%, metrorefine 30%, metrorescore 20%).
• Intentional error-handling tasks (~10% of tasks): Labeled _broken_ in the workunit name, these tasks supply a truncated ligand file (broken_lig.sdf) with receptor 10gs (max_total_results = 1) and verify that the container wrapper properly detects non-pose output and exits with error code 2 rather than validating corrupt output.

Shared test input files total approximately 410 KB and are marked <sticky/> so each host downloads them only once, while output structure-data files (docked.sdf) are gzip-compressed inside the container (shrinking typical 20 to 45 KB outputs by more than two-thirds) before upload.

Mathematical formulation and result validation

Every returned docking result is inspected on the server by gnina_validate.py via BOINC's script_validator. First, for a ligand containing $N$ heavy (non-hydrogen) atoms with Cartesian coordinates ${\mathbf{𝐫}}_i=(x_i,y_i,z_i)$, the validator computes the geometric centroid of each predicted pose:

${\mathbf{𝐜}}_{\mathrm{p}\mathrm{o}\mathrm{s}\mathrm{e}}=\frac{1}{N}\sum _{i=1}^N{\mathbf{𝐫}}_i$

To pass general plausibility validation (applied to both redock and cross tasks), every pose in docked.sdf must preserve the exact heavy-atom count $N$ of the input ligand, stay inside the binding region within a Euclidean centroid radius $d_{\max }=12.0$ Ångströms around the reference box ligand centroid ${\mathbf{𝐜}}_{\mathrm{b}\mathrm{o}\mathrm{x}}$, and carry finite empirical affinity ($E_{\mathrm{a}\mathrm{f}\mathrm{f}}$, in kcal/mol) and neural pose confidence ($\mathrm{C}\mathrm{N}\mathrm{N}\mathrm{s}\mathrm{c}\mathrm{o}\mathrm{r}\mathrm{e}$) values within valid intervals:

$\Vert {\mathbf{𝐜}}_{\mathrm{p}\mathrm{o}\mathrm{s}\mathrm{e}}-{\mathbf{𝐜}}_{\mathrm{b}\mathrm{o}\mathrm{x}}\Vert \le d_{\max },E_{\mathrm{a}\mathrm{f}\mathrm{f}}\in [-50,100],\mathrm{C}\mathrm{N}\mathrm{N}\mathrm{s}\mathrm{c}\mathrm{o}\mathrm{r}\mathrm{e}\in [0,1]$

For redocking workunits (gnina_redock_* and gnina_cpu_redock_*), the validator additionally enforces crystallographic pose accuracy. Because symmetric functional groups (such as benzene rings or carboxylate oxygens) can be numbered equivalently in different orders, gnina_validate.py partitions heavy atoms by chemical element $e$ (with $n_e$ atoms of element $e$ and $\sum _e{n}_e=N$) and solves the minimum-cost bipartite matching ${\pi }_e$ via the Hungarian algorithm in ${𝒪}(n_e^3)$ time to obtain the symmetry-matched root-mean-square deviation (RMSD) without coordinate superposition:

$\mathrm{R}\mathrm{M}\mathrm{S}\mathrm{D}(A,B)=\sqrt{\frac{1}{N}\sum _e\underset{{\pi }_e}{\min }\sum _{i=1}^{n_e}\Vert {\mathbf{𝐫}}_{A,e,i}-{\mathbf{𝐫}}_{B,e,{\pi }_e(i)}{\Vert }^2}$

At least one output pose must satisfy $\mathrm{R}\mathrm{M}\mathrm{S}\mathrm{D}(A,B)\le {\mathrm{R}\mathrm{M}\mathrm{S}\mathrm{D}}_{\max }$, where the per-system acceptance limit ${\mathrm{R}\mathrm{M}\mathrm{S}\mathrm{D}}_{\max }$ is calibrated from measured GNINA 1.3.3 baselines: 2.0 Å for 184l (baseline 0.5 to 0.9 Å), 3rod (0.5 Å), and grid (0.6 Å); 3.0 Å for 10gs (0.8 to 2.3 Å); and 3.5 Å for noelem (2.2 to 2.6 Å).

Finally, because the BOINC client aborts any task whose elapsed runtime exceeds the ratio of the workunit's floating-point operation bound to the processor's peak floating-point rate $P_{\mathrm{p}\mathrm{e}\mathrm{a}\mathrm{k}}$,

$t_{\max }=\frac{\mathrm{r}\mathrm{s}\mathrm{c}\mathrm{\_}\mathrm{f}\mathrm{p}\mathrm{o}\mathrm{p}\mathrm{s}\mathrm{\_}\mathrm{b}\mathrm{o}\mathrm{u}\mathrm{n}\mathrm{d}}{P_{\mathrm{p}\mathrm{e}\mathrm{a}\mathrm{k}}}$

and a modern GPU's $P_{\mathrm{p}\mathrm{e}\mathrm{a}\mathrm{k}}$ (tens of TFLOPS) is several hundred times larger than that of a single CPU core ($\approx 5\times 10^{10}$ FLOPS), the job generator assigns separate floating-point budgets so initial tasks have sufficient headroom to pull the 4.62 GB container image:

• CPU tasks (gnina_cpu): $F_{\mathrm{e}\mathrm{s}\mathrm{t}}=3\times 10^{12}$ FLOPs and $F_{\mathrm{b}\mathrm{o}\mathrm{u}\mathrm{n}\mathrm{d}}=1000\times F_{\mathrm{e}\mathrm{s}\mathrm{t}}=3\times 10^{15}$ FLOPs.
• GPU tasks (gnina): $F_{\mathrm{e}\mathrm{s}\mathrm{t}}=10^{16}\times f_{\mathrm{m}\mathrm{o}\mathrm{d}\mathrm{e}}$ FLOPs and $F_{\mathrm{b}\mathrm{o}\mathrm{u}\mathrm{n}\mathrm{d}}=200\times F_{\mathrm{e}\mathrm{s}\mathrm{t}}$ FLOPs (yielding $F_{\mathrm{b}\mathrm{o}\mathrm{u}\mathrm{n}\mathrm{d}}=2\times 10^{18}$ FLOPs when the scoring-mode factor $f_{\mathrm{m}\mathrm{o}\mathrm{d}\mathrm{e}}=1.0$).

System requirements and participation

Docker Demoprojekt is currently an alpha infrastructure test rather than an open public crunching campaign, and volunteers not actively testing Linux container setups are advised that the project will be removed once verification concludes. Tasks run for only a few seconds (gpucheck) to a few minutes (gnina and gnina_cpu), do not implement checkpointing, and display only approximate progress percentages inside BOINC Manager.

For Linux testers participating in the alpha phase, the documented host requirements are:

• Operating system and BOINC client: 64-bit Linux (x86_64) running BOINC client 8.2 or newer as a system service (tested on Ubuntu 24.04, Ubuntu 26.04, and Linux Mint 21.3 / 22.3; Windows WSL and macOS are not yet supported).
• Container runtime: Rootless Podman (recommended; version 4.1 or newer required for CDI GPU tasks) configured with subordinate UID/GID ranges (500000-565535) and cgroup_manager = "cgroupfs" for the boinc service account, or Docker.
• Disk space: Approximately 15 GB of free disk quota permitted to BOINC on /var/lib/boinc (individual GNINA jobs request up to a 12 GB disk quota alongside the ~4.62 GB container image).
• Graphics hardware (for GPU apps): An NVIDIA GPU of the Pascal microarchitecture or newer with CUDA compute capability $\ge 6.0$ (GeForce GTX 10-series, Turing RTX 20-series, Ampere RTX 30-series, Ada Lovelace RTX 40-series, and Blackwell RTX 50-series; older Maxwell 5.x cards are unsupported), paired with NVIDIA Linux driver 570 or newer (575+ recommended) and the NVIDIA Container Toolkit (/etc/cdi/nvidia.yaml). AMD GPUs are not yet supported.

Server statistics

The following operational figures were recorded from the Docker Demoprojekt server status, applications, and host breakdown pages on 5 November 2026 at 16:43 UTC:

Metric	Value
BOINC server software version	1.6.2 (database schema 27030)
Users with credit / active users	5 / 5 (4 registered in past 24 hours)
Computers with credit / active hosts	7 / 7 (100% Linux x86_64)
Connected BOINC client versions	8.2.9 (3 hosts, 96.94% of recent credit), 8.2.13 (3 hosts, 3.06%), 8.2.15 (1 host, <0.01%)
Tasks ready to send	54 (45 gnina, 9 gpucheck, 0 gnina_cpu)
Tasks in progress	2 (1 gnina, 1 gpucheck)
Current floating-point speed (server_status.php)	2,427.73 GigaFLOPS (2.43 TFLOPS)
Application average throughput (apps.php)	25,794 GigaFLOPS total (25,790 GigaFLOPS GPU; 4 GigaFLOPS CPU)

Scientific publications

Publications by the project

As of 5 November 2026, no peer-reviewed scientific papers have been published directly from Docker Demoprojekt computations. Because the server was created purely to validate containerized BOINC infrastructure using established protein-ligand benchmark complexes ahead of production CitizenPharma and SiDock@home deployments, its test results are discarded after automated validation.

Foundational publications for GNINA and docking methods

The containerized docking applications tested on Docker Demoprojekt (gnina and gnina_cpu) are built directly on the following peer-reviewed scientific publications describing GNINA, its convolutional neural network scoring ensembles, and its underlying molecular docking engines:

• McNutt, Andrew T.; Li, Yanjing; Meli, Rocco; Aggarwal, Rishal; Koes, David Ryan (2025). "GNINA 1.3: the next increment in molecular docking with deep learning". Journal of Cheminformatics 17 (1): 28.
• McNutt, Andrew T.; Francoeur, Paul; Aggarwal, Rishal; Masuda, Tomohide; Meli, Rocco; Ragoza, Matthew; Sunseri, Jocelyn; Koes, David Ryan (2021). "GNINA 1.0: molecular docking with deep learning". Journal of Cheminformatics 13 (1): 43.
• Ragoza, Matthew; Hochuli, Joshua; Idrobo, Elisa; Sunseri, Jocelyn; Koes, David Ryan (2017). "Protein-Ligand Scoring with Convolutional Neural Networks". Journal of Chemical Information and Modeling 57 (4): 942-957.
• Koes, David Ryan; Baumgartner, Matthew P.; Camacho, Carlos J. (2013). "Lessons Learned in Empirical Scoring with smina from the CSAR 2011 Benchmarking Exercise". Journal of Chemical Information and Modeling 53 (8): 1893-1904.
• Trott, Oleg; Olson, Arthur J. (2010). "AutoDock Vina: Improving the speed and accuracy of docking with a new scoring function, efficient optimization, and multithreading". Journal of Computational Chemistry 31 (2): 455-461.

Docker Demoprojekt runs on the BOINC volunteer computing platform.

See also

• BOINC projects
• SiDock@home
• Yoyo@home
• RNA World (beta)
• BOINC Central
• BOINC Alpha Test
• Albert@Home
• RALPH@home
• Docking@Home
• Rosetta@home
• GPUGRID
• WUProp@Home
• Main Page

External links

• https://boinc.kerbodyne.com/dockerdemo/ (official project website)
• https://boinc.kerbodyne.com/dockerdemo/server_status.php (live server status)
• https://boinc.kerbodyne.com/dockerdemo/apps.php (project applications)
• https://github.com/MBlenn/boinc-stack (boinc-stack containerized BOINC server stack on GitHub)
• https://github.com/MBlenn/boinc-apps (boinc-apps containerized GNINA applications on GitHub)
• https://github.com/MBlenn/boinc-stack/blob/main/docs/volunteer-setup.md (Linux Podman and Docker volunteer host setup guide)
• https://www.rechenkraft.net/wiki/CitizenPharma (CitizenPharma initiative on the Rechenkraft.net Wiki)
• https://github.com/gnina/gnina (GNINA upstream deep-learning molecular docking repository)