Open research questions in Scientific Computing and Data Management

What the literature leaves open

• While fully autonomous scientific writing and generating conference-ready papers remain notoriously difficult open problems for the entire field, SAGE successfully produces significantly more reliable and higher-quality scientific artifacts.

  One Reflection Is Not Enough: Self-Correcting Autonomous Research via Multi-Hypothesis Failure Attribution · 2026

• • Algebra of sufficient regime specifications — role independence, minimality, equivalence, dominance/refinement, category/lattice, regime-change sufficiency, machine-checkability, and inquiry termination. • Root of demand / coherence-of-posing — whether bounded inquiry itself forces stable verdict demand, possibly by treating inquiry as an identity-bearing process. • Whether r alone fixes tolerance / aggregation / governance / preservation relation / conflict poset. • Stochastic, nonstationary, hybrid, and broader general-topological capacity. • Route B (the Archimedean hinge) · continuous conflict beyond finite Birkhoff. • A formal theory of regime compression · empirical validation · standards · independent verification. END OF ORIENTATION · VERSION 4 What Licenses Sameness Through Change?

  A Short Orientation to the Identity-Persistence Program · 2026 · DOI

• In particular, sufficient motivation has not yet been established for AIGC issuers to actively publish structured and contextualized AIGCs using our framework.

  A Prompt-Aware Structuring Framework for Reliable Reuse of AI-Generated Content in the Agentic Web · 2026 · DOI

• Across these settings, NEI competence does not transfer reliably: models trained on shortcut-prone constructions fail to recognize semantically related insufficient evidence, and mixed-construction training narrows but does not close the gap.

  Evidence Absence Is Not Evidence Insufficiency: Diagnosing NEI Construction Artifacts in Fact Verification · 2026

• It addresses a fundamental gap in the current landscape: while content provenance standards describe what was generated, and legal remedies operate after harm is done, no standardized mechanism exists for determining whether AI-generated content is authorized before it is produced.

  Prompt Protocol · 2026 · DOI

• Despite this expansion, there exists no standardized, cryptographically verifiable infrastructure capable of proving the provenance, reproducibility, model origin, training lineage, or verification status of AI-generated artifacts.

  INTERGET v2.0: The Provenance and Autonomous Coordination Layer of the Sovereign Web — Intelligent Network for Trusted Execution of Recorded Generative Experiences and Trajectories · 2026 · DOI

• (cid:136) The chyren-cim, chyren-ternary, and chyren-vision crates are early-stage; their APIs should be treated as unstable. (cid:136) The MYELIN Holonomic Clustering is currently implemented as k-means in embedding space; a proper E8-rooted clustering requires further development. (cid:136) The Lean 4 verification covers core axioms; full verification of the 20-crate system is a multi-year effort. (cid:136) Agent mesh (MQTT-based, in the cursor/integration-hardening branch) is in progress; production readiness TBD.

  Chyren Sovereign Intelligence: A Technical Specification for Orchestrated AI · 2026 · DOI

• 9.1 Metrics Not Yet Verified The v3.0 paper described a multi-source validation protocol requiring 30+ independent source confirmations before elevating a pattern to high-confidence status, and a divergence detection mechanism comparing curated vs. live graph patterns. These mechanisms were part of an earlier architectural phase (the Python-based wisdom graph pipeline processing curated books). They are not present in the current Rust autonomous loop. The v4.0 system does not implement multi- source validation; each node enters the graph after a single scraping cycle passes the four honesty layers. 9.2 Semantic Deduplication Not Implemented The current deduplication is label-based. Two nodes with identical content but different labels (e.g., two responses to differently-phrased questions about the same topic) are treated as distinct knowledge. True semantic deduplication would require embedding comparison, which is not present in the current stack. The graph continues to accumulate thematic paraphrases at a slower rate than before the label guard, but does accumulate them. 9.3 Elegant Confabulation Risk As documented in Section 6, Gemma 4 E2B confabulates with domain-expert fluency. For any deployment involving regulatory facts, institutional structures, or numerical claims, the system must operate in pure retrieval mode—summarizing provided documents rather than generating from parametric memory. The current autonomous loop does not enforce this boundary; gap investigation prompts can elicit confabulation on topics where the graph provides insufficient context. 9.4 Race Condition on Concurrent Write the same If --query mode and --autonomous mode are run simultaneously against animus_memory.json, the last writer wins and the other session's nodes are lost. This was observed directly on June 12, 2026, when four query-mode probes were overwritten by the autonomous loop's next save cycle. A file-lock mechanism is not implemented. The operational workaround is strict sequential use: stop the autonomous loop before running queries. 9.5 Single Hardware Profile All performance measurements are from a single hardware configuration (Dell Precision 7610, i7- 11800H, RTX 3050 4GB, Windows 11). Generalizability to other configurations is not characterized.

  ANIMUS v4.0: Honest Reconstruction of an Autonomous Knowledge System — Engine Migration, Phantom Memory Audit, and Epistemic Honesty Engineering on Consumer GPU · 2026 · DOI

• A frontier not yet settled within the program. Naming it protects the core from overclaim. OPEN Everyday “Whether one general formula for identity capacity exists across domains” is open — saying so is a feature, not a weakness. » Open means not yet closed, not secretly failed.

  Glossary of the Identity-Persistence Stack · 2026 · DOI

• Future work may develop machine-readable coordinate profiles for node, data, state, authority, evidence, result, and preservation references. Such profiles should remain non-executable. They may support audit, review, feasibility analysis, public-sector documentation, or implementation planning, but they should not become automatic authority, proof, settlement, execution, or legal-effect mechanisms. Future work may also develop domain-specific coordinate templates. Public administration, healthcare, education, finance, legal preparation, software development, media, telecommunications, semiconductor design, quantum computing, robotics, infrastructure, and regional energy systems may all require different coordinate profiles. Domain-specific templates should preserve the universal non-substitution boundaries while allowing local vocabulary. Further research may connect the Coordinate Matrix to Essential Terra Core module descriptions, Terra Topology case study templates, public data preservation packages, heat-demand reference packages, and provider-independent infrastructure standards. Each development should preserve the distinction between reference grammar, implementation layer, and external process. The long-term goal is not to create a central platform. It is to enable responsible AI and AGI-era infrastructure to remain legible across providers, institutions, devices, local nodes, public records, industrial systems, and preservation contexts while preserving human responsibility and structural boundaries. Machine-readable futures Machine-readable coordinate profiles may become useful for infrastructure audits, document packages, node registries, and preservation manifests. Such profiles should encode structural positions without triggering automated effect. Their design should treat non-execution as a first-class requirement. Potential fields may include object identity, node role, data origin, state source, authority condition, evidence reference, result relation, preservation rule, and external-effect boundary. These fields would support review, not replace it. Standards-oriented futures Standards-oriented work may define how institutions refer to coordinate positions across systems. The standard should not mandate one platform or provider. Its value would be to preserve boundary language across heterogeneous infrastructure. 20 Education and training materials may also be developed to teach layer absorption, axis contamination, false finality, and provider capture. The Coordinate Matrix is most useful when practitioners can identify these risks before implementation pressure hides them.

  Paper 28 — AXION Coordinate Matrix: Seven-Axis Reference Alignment for Hardware-Agnostic AI Infrastructure · 2026 · DOI

• References and Conceptual Background 2 1. Introduction: Why Coordinate Alignment Matters AI infrastructure is often described through the vocabulary of devices, clouds, models, data centers, networks, and applications. That vocabulary is useful for procurement and engineering, but it is not sufficient for AI and AGI-era infrastructure. Generated outputs can appear authoritative across domains, data can move across providers, and local hardware can become entangled with public records, industrial operations, evidence contexts, and preservation duties. The structural problem is not simply how to compute. It is how to keep reference positions distinguishable while allowing heterogeneous infrastructure elements to correspond. Coordinate alignment matters because AI infrastructure contains many different objects that can be mistaken for one another. A data node can be mistaken for an authority layer. A processing result can be mistaken for evidence confirmation. A local record can be mistaken for a public decision. A cloud resource can be mistaken for institutional memory. A hardware endpoint can be mistaken for device control. Without a coordinate grammar, the infrastructure becomes legible only as a pile of tools or an execution pipeline. Both descriptions are inadequate for civilizational AI infrastructure. The AXION Coordinate Matrix addresses this problem by giving each infrastructure element a position across seven bounded reference axes: node, data, state, authority, evidence, result, and preservation. The axes do not perform the functions they classify. They make those functions referable. A node coordinate does not control the device. A data coordinate does not alter ownership. A state coordinate does not create truth. An authority coordinate does not grant legal authority. An evidence coordinate does not certify proof. A result coordinate does not approve effect. A preservation coordinate does not become an archive with sovereign meaning. Each axis is a reference position. This paper therefore develops the Coordinate Matrix as a public structural grammar for hardware- agnostic AI infrastructure. The emphasis is on non-executable alignment. The framework is compatible with many backend types, local nodes, public institutions, private facilities, preservation layers, and external processes, but it does not merge those processes into one universal system. It continues the AGI Structural Alignment Series by moving from portfolio-level non-substitution into the specific coordinate discipline required for AI data infrastructure. Infrastructure legibility rather than infrastructure control The central proposition of coordinate alignment is that infrastructure must first become legible before it becomes actionable. Legibility does not mean visibility to a provider or administrator alone. It means that each object can be understood according to its node position, data condition, state posture, authority boundary, evidence relation, result status, and preservation context. This is a structural requirement rather than a monitoring requirement. Control-centered architectures often treat legibility as a by-product of management. AXION reverses the order. It treats legibility as a boundary-preserving condition. An infrastructure object that cannot be placed across the seven axes should not be treated as ready for institutional or public effect, no matter how easily it can be computed, transmitted, or displayed. 3 Why a coordinate grammar is necessary after portfolio architecture Portfolio architecture explains how many layers coexist without absorption. Coordinate alignment explains how a particular infrastructure object becomes referable inside that portfolio. The two structures are complementary. Portfolio architecture prevents the whole system from collapsing into one layer; Coordinate Matrix prevents an individual object from being interpreted through one axis only. This is the reason Paper 28 follows AXION rather than preceding it. AXION names the exchange architecture. The Coordinate Matrix supplies the internal grammar that allows the exchange architecture to remain hardware-agnostic, provider-independent, and non-executive.

  Paper 28 — AXION Coordinate Matrix: Seven-Axis Reference Alignment for Hardware-Agnostic AI Infrastructure · 2026 · DOI

• In this paper, we reported on work in progress exploring the potential of large language models and agent-based approaches to support computational reproducibility in the social sciences. Using a controlled synthetic benchmark of R-based studies with systematically injected failures, we evaluated two automated repair workflows, a prompt-based approach and an agent-based approach, under validation conditions where the ground-truth is known. Our findings offer clear answers to the research questions posed at the outset of this study. Regarding RQ1 (extent of automated repair) and RQ3 (impact of error complexity), our results suggest that prompt-based workflows can successfully repair and reproduce a non-trivial share of corrupted analyses for which reproduction of computational results would have been impossible without prior repair, particularly in cases involving execution-level and moderately complex errors. However, their effectiveness is strongly conditioned on the availability and structure of contextual information, and performance degrades substantially when errors require reconstructing missing analytical logic (Category C). Concerning RQ4 (influence of contextual information), we found that the effectiveness of prompt-based repair is strongly conditioned on the availability and structure of context. Providing the full text of the publication and supplementary scripts significantly improved success rates for complex test cases, whereas minimal context was often insufficient for logicheavy repairs. This confirms that while LLMs can reduce manual debugging effort, they remain highly sensitive to prompt design and context selection. Finally, addressing RQ2 (comparison of workflows), our analysis shows that agent-based workflows consistently outperform prompt-based LLM approaches across all categories of test cases, with particularly strong gains for complex, multi-step failures. By allowing models to interact directly with the execution environment—inspecting files, iteratively modifying code, and rerunning analyses—agentic systems appear better suited to handling the practical realities of computational reproducibility. This suggests that the added autonomy and environmental access provided by agents is a key factor in achieving higher reproduction success rates, rather than model capability alone. As ongoing work, this study naturally has limitations, and we have planned several next steps to address them. First, we consider expanding the benchmark to include additional studies, a broader range of failure modes, and more diverse analytical patterns common in computational social science. Second, we plan to evaluate additional models and agent architectures to better understand how general these findings are across different AI systems. Third, in the current agent-based setup, the ground-truth outputs are mounted within the same container used for repair.

  Automating Computational Reproducibility in Social Science: Comparing Prompt-Based and Agent-Based Approaches · 2026 · DOI

• In the preceding pages, we have seen that simulations are a powerful and versatile tool throughout all major steps of statistical workflows. In the foreseeable future, we expect the utility and use of simulations to increase even further. In particular, amortized inference, which relies heavily on both model simulations and continuous progress in deep learning, is likely to become a viable, widespread alternative to established inference approaches. Going beyond the techniques available to date, one could imagine simulation-based training of entire world models that are able to rapidly and semi-automatically execute full statistical workflows on real data, essentially playing the role of a statistics expert assisting the user in their analysis. One of the key challenges for such machine-assisted statistics is how to combine the general expertise of the machine with the subject matter knowledge of the user. No matter how many simulations the machine has been trained on, it would still likely miss key subject matter knowledge about the specific real data being analyzed, just as a statistics expert wouldn’t know all the intricate details of the data. This also highlights a more general question pertinent to all inference machines trained on simulated data: how to bridge the statistical gap between simulated worlds and the real world. Put differently, we need to learn how to formally integrate information from simulated data, real data, and subject matter knowledge into inferences that are both fast and trustworthy. Much remains to be simulated.

  Simulations in statistical workflows · 2026 · DOI

• 17.1 Prototype Expansion Problem A finite Prototype Space increases safety but reduces expressive freedom. Novel user intents may be compressed into conservative circuits, causing loss of expressiveness. Future work should define safe mechanisms for propos‑ ing, verifying, logging, and admitting new circuit prototypes. 17.2 Extraction Uncertainty The extraction from natural‑language intent to Semantic IR may itself be uncertain. When extraction cannot be performed deterministically or with sufficient confidence, the compiler must return Clarify or fail closed. 17.3 Proof‑Carrying Synthesis Future work may attach machine‑checkable proofs or signed attestations to synthesized runtime artifacts. This would connect Phase‑2 to Phase‑3 Foundation topics such as Digital Deed, provenance, ROM Identity, and cryp‑ tographic addressability. 11 17.4 Runtime Calibration ReplayLog data may be used to calibrate distance weights, admissibility thresholds, and circuit selection rules. Calibration must not override hard policy constraints or fail‑closed boundaries.

  AIKernel Phase-2 Theory: Semantic Compilation Architecture · 2026 · DOI

• This paper has three main limitations. First, it is a methods/governance preprint, not a benchmark or comparative study. Its primary contribution is conceptual and procedural: a model of how artifact authority can be represented explicitly in transcript-sufficient research systems. Second, the case is unusually explicit. The Reflexive Laboratory is not a typical research archive. It includes a coordination registry, normalized vocabularies, update packets, snapshots, and transcript-visible governance rules. Those properties are precisely what make the distinction between source presence and canonicality observable here. Third, the related-work positioning is intentionally compact in this version. The manuscript is now externally legible as a preprint, but a later release should add fuller bibliographic integration and comparative treatment across other registry or workflow systems.

  Control and Reliability for Research Continuation: Observability, Failure Propagation, and Validation in Transcript-Sufficient Systems · 2026 · DOI

• org/exfor/tools • Python package testing via GitLab CI • Python package build and deployment via GitLab CI • Integration with other NEA GitLab projects Get started with GitLab • Create your own fork / clone • Set up your own workflows • Feel free to open issues / merge requests • Let us know what doesn’t work • Let us know what would be useful for you! • Let me know if your git.

  Deep dive into NEA Exfor workflows · 2026 · DOI

• • Trust model: Shadow Archive trusts the Git hosting platform. A malicious host could theoretically alter timestamps. Mitigations include using multiple hosts or periodically anchoring root hashes to a blockchain. • Selective disclosure: Currently, verification requires revealing the entire repository. Future work could support Merkle proofs for individual files without exposing the full tree. • Formal verification: A formal security proof under standard cryptographic assumptions would strengthen confidence in the combined system. • AiLock Tier 2 boundary: The command-name configurability defense is a discovery-layer control, not a cryptographic guarantee. Developers requiring stronger Tier 2 protection may combine renaming with OS-level execution policy or interactive passphrase requirements. Engineering hardening (future work). The current implementation demonstrates the core security properties as a research prototype. Production deployment requires additional work: • Password rotation: No ailock rekey command currently exists; a production implementa- tion should provide atomic re-wrapping of all per-file keys. 9 • Multi-system manifest sync: The manifest lives in a local cache directory not committed to the private repository; explicit sync or an encrypted in-repo manifest is needed for multi-machine use. • Multi-user key management: The dual key wrapping mechanism supports per-user pass- words, but onboarding a second user is not yet a first-class command. • Partial-failure recovery: Lock operations spanning many files are not currently atomic; production use requires transactional semantics or a recovery journal.

  Protecting Intellectual Priority in the Age of AI: Runnable Encrypted Code and Public Provenance for Private Repositories · 2026 · DOI

• HiRSE Seminar | Metadata for Research Software | M.Gruenpeter | 25/03/2026 | CC-BY 4.0 | #25/46 Objective: #RSMD_cheklist To ensure the collection, curation, and maintenance of research software metadata, the following general requirements are recommended for end users, including researchers, software engineers, curators, and institution staff.

  Making Research Software Visible, Citable, and Preserved: A Metadata Deep Dive for RSEs · 2026 · DOI

• This paper and the DistributedWorkflows.jl package intro- duce a user-friendly interface for managing distributed, task-based workflows. It enables users to generate, visualize, compile, and launch workflows represented as Petri nets through simple meth- ods. The package provides a fully documented public API, allow- ing users to locally test applications before deploying them on ex- pensive clusters, ensuring cost efficiency. With binaries available for multiple Linux distributions, the tool is currently best suited for long-running processes. Upcoming updates aim to enhance the package with specialized transition types for reduced boilerplate code, additional workflow examples to guide users, and convenience functions that streamline Fig. 6. Part of a Petri net modeling different repair synthesis pathways of DNA as in [26]. the process of creating and managing workflows. Furthermore, im- provements to the user interface are planned, making the tool even more intuitive. These features are expected to expand the package’s utility, enabling broader adoption and more efficient handling of distributed workflows across various applications. 8. ACKNOWLEDGEMENTS We extend our gratitude to Fraunhofer ITWM for funding the Ph.D. studies of the first author. As well as, enabling her to work on and develop DistributedWorkflows.jl. We are also deeply grate- ful to Prof. Dr. Anne Frühbis-Krüger and Prof. Dr. Claus Fieker for their invaluable insights into the needs of domain scientists, which guided the design of this package with a strong focus on user experience. Our sincere thanks goes to Dr. Tiberiu Rotaru for his expertise in GPI-Space, particularly during the initial stages of integrating our Julia prototype with C++.

  DistributedWorkflows.jl - A Julia interface to a task-based workflow management system. · 2026 · DOI

• Like any software, the DistributedWorkflows.jl package has certain limitations. It is an interface package and hence relies on tools outside of Julia. It has somewhat of a setup process before be- ing able to parallelize applications. It is currently, recommended for long-running processes due to I/O overhead and requires a shared filesystem. Official support for the workflow manager is limited to Ubuntu 20 and 22 LTS, with compatibility restricted to Linux dis- tributions, excluding macOS and Windows. Additionally, the pack- age relies on Spack for binary installation, which may introduce additional setup complexity for some systems.

  DistributedWorkflows.jl - A Julia interface to a task-based workflow management system. · 2026 · DOI

• Looking ahead, two directions stand out. First, the automation of the shared database lifecycle remains a key priority. While defaults curated by experienced administrators and validated against production data have proven highly reliable, scaling to thousands of tools will require semi-automated updates. Incremental approaches, such as bracketing input sizes for tools with predictable scaling or selectively benchmarking high-impact applications, could provide efficient pathways toward automation without overreliance on machine learning. Second, there is potential for closer integration with live system telemetry. The meta-scheduling function at Galaxy Australia already queries backend load in real time, but similar approaches could extend to memory or I/O characteristics. Combining static defaults with live data would yield more adaptive scheduling while maintaining interpretability. Third, the shared TPV configuration database, combined with historical job metrics, could serve as training data for ML models that infer baseline configurations for new tools lacking curated entries. This is particularly appealing for the cold-start scenario where a new tool is added and no prior execution data exists. Such a hybrid approach could complement the rule-based system by providing informed starting points while retaining TPV’s interpretability and administrator override capabilities.

  Right-Sizing Compute Resource Allocations for Bioinformatics Tools with Total Perspective Vortex · 2026 · DOI

• At the same time we see that it is insufficient to only pass FDOs as JSON objects, as they also have references to other data such as images, which should not need to be re-downloaded.

  Incrementally building FAIR Digital Objects with Specimen Data Refinery workflows · 2022 · DOI

• The scope of nanopublications is limited to the assertional data type and, as the name suggests, nanopublications should remain small in size (limited to single assertions as individual triples or small RDF graphs).

  The Comparative Anatomy of Nanopublications and FAIR Digital Objects · 2022 · DOI

• The claim is limited to the author’s documented proposal and does not assert legal ownership of a general academic word, worldwide priority without qualification, or acceptance by an academic community.

  Ouroboration: Recursive Provenance Loss and False Corroboration in AI-Mediated Information Systems · 2026 · DOI

• While it is well known that engineers frequently copy and paste meta-models, the motivations behind this practice and its potential drawbacks remain uncertain.

  Analysis of EMF Meta-Model Duplication in Open Source Repositories · 2026 · DOI

Most-cited papers in Scientific Computing and Data Management

• The Galaxy platform for accessible, reproducible, and collaborative data analyses: 2024 update · Nucleic Acids Research · 2024 · 851 citations
• BindingDB in 2024: a FAIR knowledgebase of protein-small molecule binding data · Nucleic Acids Research · 2024 · 168 citations
• Threats of a replication crisis in empirical computer science · Communications of the ACM · 2020 · 133 citations
• KRYSTAL: Knowledge graph-based framework for tactical attack discovery in audit data · Computers & Security · 2022 · 57 citations
• Provenance Erasure Rate: A Compression-Survival Metric for Attribution Loss in AI-Composed Search Outputs · Zenodo (CERN European Organization for Nuclear Research) · 2026 · 55 citations
• Assessing fidelity in XAI post-hoc techniques: A comparative study with ground truth explanations datasets · Artificial Intelligence · 2024 · 39 citations
• Share the code, not just the data: A case study of the reproducibility of articles published in the Journal of Memory and Language under the open data policy · Journal of Memory and Language · 2022 · 38 citations
• Utilizing <scp>OpenAI</scp>'s <scp>GPT</scp>‐4 for written feedback · TESOL Journal · 2023 · 26 citations
• Softcite dataset: A dataset of software mentions in biomedical and economic research publications · Journal of the Association for Information Science and Technology · 2021 · 26 citations
• Erasure Skew: A Measurement Program for the Power-Conditioning of Provenance Loss in Retrieval and Composition Systems · Zenodo (CERN European Organization for Nuclear Research) · 2026 · 14 citations

Most recent work

• Provenance Erasure Rate: A Compression-Survival Metric for Attribution Loss in AI-Composed Search Outputs · Zenodo (CERN European Organization for Nuclear Research) · 2026
• Erasure Skew: A Measurement Program for the Power-Conditioning of Provenance Loss in Retrieval and Composition Systems · Zenodo (CERN European Organization for Nuclear Research) · 2026
• Control and Reliability for Research Continuation: Observability, Failure Propagation, and Validation in Transcript-Sufficient Systems · Open MIND · 2026
• Constitutive Provenance: A Sentence-Level Countermeasure Against Composition-Layer Erasure · Zenodo (CERN European Organization for Nuclear Research) · 2026
• Provenance After AI: Semantic Provenance and the Provenance Erasure Rate as Extension of C2PA, Data Provenance Initiative, and EU AI Act Frameworks (v1.1) · Zenodo (CERN European Organization for Nuclear Research) · 2026
• Formal Foundations of Semantic Physics (EA-SEI-FF-01, v0.2 — Post-Assembly Review) · Zenodo (CERN European Organization for Nuclear Research) · 2026
• Provenance Is What Authorship Must Endure: AI-Mediated Writing, the Authorship-Slop Distinction, and the Missing Third Dimension of Provenance · Zenodo (CERN European Organization for Nuclear Research) · 2026
• Integrity Lock Certificate: The Refutation Triad — Mutual Anchoring of the AI_Bleeding Refutation Dossier (EA-LOCK-AIBLEEDING-01 v1.0) · Zenodo (CERN European Organization for Nuclear Research) · 2026
• ADR-001: Genesis ODE — Core vs Experimental Scope · Zenodo (CERN European Organization for Nuclear Research) · 2026
• From FAIR to CURE: guidelines for computational models of biological systems · npj Systems Biology and Applications · 2026

Related topics in Decision Sciences