Skip to content
DraconiticResearch
Search Sources API
186,631 studies31 sourcesUpdated 59 min ago
PreprintCC BYChemRxiv

From Computation-Ready to Claim-Ready: An Evidence Framework for MOF Machine Learning

  • Natalia Lydia Papadantonaki1,2
  • Shayan Abaei3,4
  • Joshua Emptoz1,2,5
  • Mehrdad Asgari1,2,5
5 affiliations
  1. 1Lucy Cavendish College
  2. 2University of Cambridge
  3. 3School of Chemical Engineering, Faculty of Engineering
  4. 4University of Tehran
  5. 5Department of Chemical Engineering and Biotechnology
Open at sourcePDFDOICite

Abstract

Scientific data are routinely transformed so they can be stored, joined and used by machine-learning models. These steps enable scale, but they can also remove the context that tells us what a prediction means and how far it can be trusted. Metal–organic frameworks (MOFs) make this problem especially clear: one nominal material may appear as an experimental crystal, an activated simulation structure, a charge-assigned model or a row of descriptors, while information about protonation, counterions, coordination environment or measurement conditions may disappear along the way. We introduce claim readiness: a framework for asking whether the evidence preserved through a data pipeline is sufficient for the scientific claim made at the end of it. By tracing chemical evidence, identifier joins and evaluation choices, we show that data can be highly usable for machine learning yet incomplete for chemistry-sensitive interpretation, and that a model can still be useful for ranking candidates even when its value-level accuracy decreases under a harder split. The wider lesson is not specific to MOFs: trustworthy scientific machine learning requires a traceable connection between the representation used by the model and the evidence needed to support the resulting claim. Highlights • Claim readiness matches scientific claims to evidence preserved in data pipelines • Model-ready materials data can lose chemistry needed for interpretation • Data joins change which materials enter representation comparisons • Harder data splits reduce accuracy while useful candidate ranking can remain

Related

  1. Application of spectroscopic and multispectral imaging technologies on the assessment of ready-to-eat pineapple quality: A performance evaluation study of machine learning models generated from two commercial data analytics tools
    Evanthia Manthou, Sergio-Llaneza Lago, Evaggelos Dagres +5
    Computers and Electronics in AgriculturePeer-reviewedLicence not stated38 citedGeneral biologyGeneral chemistryHAL
  2. FMS2Lib: A Machine Learning-Ready Tandem Mass Spectral Library for Nontargeted Analysis of Non-conventional PFAS
    Tao Yang, Xiaoyun Wang, Diwen Yang +7
    ChemRxivPreprintCC BYFull textEnvironmental chemistryChemRxiv
  3. Machine-Learning Models Bypass Charge Computation for Modeling 14N Quadrupolar Relaxation in Ionic Liquids
    Federico Civaia, Yanan Li, Florin Teleanu +4
    ChemRxivPreprintCC BYPhysical chemistryChemRxiv
  4. NephroTox AI: an explainable machine learning framework with uncertainty quantification for drug-induced nephrotoxicity prediction
    Jixing Liu, Jiabing Guo, Henry H. Y. Tong +1
    Journal of CheminformaticsPeer-reviewedCC BY-NC-ND 4.0Theoretical & computational chemistryCrossref
  5. An integrated machine learning framework for accurate, real-time and non-destructive vitamin C quantification from Vis/NIR spectral data
    Saleh Yasour Pourali, Amir Hossein Afkari-Sayyah, Ebrahim Taghinezhad +1
    Scientific ReportsPeer-reviewedCC BY-NC-ND 4.0Full textAnalytical chemistryCrossref

Cite

BibTeXRIS
BibTeX
@misc{papadantonaki2026from,
  title = {{From Computation-Ready to Claim-Ready: An Evidence Framework for MOF Machine Learning}},
  author = {Natalia Lydia Papadantonaki and Shayan Abaei and Joshua Emptoz and Mehrdad Asgari},
  howpublished = {ChemRxiv},
  year = {2026},
  month = oct,
  doi = {10.26434/chemrxiv.15009825/v1},
  url = {https://chemrxiv.org/doi/full/10.26434/chemrxiv.15009825/v1},
  note = {CC BY},
}
RIS
TY  - UNPB
TI  - From Computation-Ready to Claim-Ready: An Evidence Framework for MOF Machine Learning
AU  - Natalia Lydia Papadantonaki
AU  - Shayan Abaei
AU  - Joshua Emptoz
AU  - Mehrdad Asgari
PY  - 2026
DA  - 2026/10/04
PB  - ChemRxiv
DO  - 10.26434/chemrxiv.15009825/v1
UR  - https://chemrxiv.org/doi/full/10.26434/chemrxiv.15009825/v1
AB  - Scientific data are routinely transformed so they can be stored, joined and used by machine-learning models. These steps enable scale, but they can also remove the context that tells us what a prediction means and how far it can be trusted. Metal–organic frameworks (MOFs) make this problem especially clear: one nominal material may appear as an experimental crystal, an activated simulation structure, a charge-assigned model or a row of descriptors, while information about protonation, counterions, coordination environment or measurement conditions may disappear along the way. We introduce claim readiness: a framework for asking whether the evidence preserved through a data pipeline is sufficient for the scientific claim made at the end of it. By tracing chemical evidence, identifier joins and evaluation choices, we show that data can be highly usable for machine learning yet incomplete for chemistry-sensitive interpretation, and that a model can still be useful for ranking candidates even when its value-level accuracy decreases under a harder split. The wider lesson is not specific to MOFs: trustworthy scientific machine learning requires a traceable connection between the representation used by the model and the evidence needed to support the resulting claim. Highlights • Claim readiness matches scientific claims to evidence preserved in data pipelines • Model-ready materials data can lose chemistry needed for interpretation • Data joins change which materials enter representation comparisons • Harder data splits reduce accuracy while useful candidate ranking can remain
ER  -