Automated SSbD Workflows Using FAIR Electronic Laboratory Notebooks:Application to Polymer Informatics

Jeliazkova, Nina; Kochev, Nikolay; Radeva, Lyudvika; ILIEV, Luchesar; Jeliazkov, Vedrin; CAMPANA, GIAMPAOLO

doi:10.5281/zenodo.17647359

Automated SSbD Workflows Using FAIR Electronic Laboratory Notebooks:Application to Polymer Informatics

Jeliazkova, Nina; Kochev, Nikolay; Radeva, Lyudvika; ILIEV, Luchesar; Jeliazkov, Vedrin; CAMPANA, GIAMPAOLO

Publisher: Zenodo
DOI: 10.5281/zenodo.17647359
Repository: zenodo.org
Language: en

Abstract

Safe-and-Sustainable-by-Design (SSbD) is a voluntary European framework guiding chemical and material development towards sustainability, but practical implementations face gaps in data, AI validation, interoperable workflows, and collaborative tools 1. We present a digital framework2 that addresses these gaps by combining a FAIR-compliant Electronic Lab Notebook (eLabFTW3) with a graph-based representation of SSbD workflows. Materials, processes, lab results, and decisions are captured as structured, linked resources, enabling transparency and traceability across multidisciplinary teams. The ELN can be populated manually by researchers or automatically via scripts that retrieve data from public or internal databases (e.g., AMBIT, CompTox, REACH) and create structured entries. Python workflows access the ELN API to extract metadata, experimental outcomes, and decisions to generate SSbD reports, including hazard tables and selection rationales. Predictive model outputs - from VEGA4, EPA tools5, and other QSAR models - are processed through conformal prediction6, a model-agnostic framework that provides prediction sets or intervals that are guaranteed to contain the true outcome with a user-specified confidence level (e.g., 90%). Rather than relying on a single model or a fixed threshold, conformal prediction enables integration of multiple predictive models by aggregating their prediction sets, for example by intersection or union, or by weighting based on the interval width. To demonstrate the scalability of the approach, we developed a cheminformatics pipeline for PFAS-alternative polymer screening. Using SLN7 and SMIRKS8 notation, 7.7 million polymer candidates were generated and filtered based on experimental feasibility and lab-scale constraints. Simulated chemical transformations yielded 7.5 million final structures. Surface tension was used as a screening property, estimated using an empirical relation with cohesive energy density (CED). CED was computed using two QSPR models: Fedors’ method (via Ambit-GCM9) and Zhao’s method (via CDK10), with 118 SMARTS patterns enabling automated group-based property estimation. This approach links automated candidate generation, predictive evaluation, and structured documentation into a unified SSbD workflow. It supports transparent decision-making, reproducibility, and co-creation, and is adaptable across material domains. Link to tools and example implementations are available at https://sbd.adma.ai. Acknowledgements: This work has received funding from the European Union’s Horizon Europe research and innovation programme under grant agreement No. 101092164 (ZeroF) and No. 101130073 (PHAntastic). Also funded by the Swiss State Secretariat for Education, Research and Innovation (SERI). References 1. Leopold, A. et al. Recommendations and Research Gaps to Improve the Function of the Safe and Sustainable by Design (SSbD) concept: SETAC – EC Consultation series 2023-24. https://zenodo.org/records/14228553 (2024) doi:10.5281/zenodo.14228553. 2. Jeliazkova, N., Kochev, N., Iliev, L., Jeliazkov, V. & Campana, G. Towards Addressing Co-Creation Gaps: Automating Safe-and-Sustainable-by-Design Workflows with Electronic Lab Notebooks. in 21st Global Conference on Sustainable Manufacturing (GCSM 2025) accepted (2025). 3. CARPi, N., Minges, A. & Piel, M. eLabFTW: An open source laboratory notebook for research labs. J. Open Source Softw. 2, 146 (2017). 4. VEGA-QSAR. Virtual models for Evaluating the properties of chemicals within a Global Architecture. http://www.vega-qsar.eu/. 5. Lowe, C. N. & Williams, A. J. Enabling High-Throughput Searches for Multiple Chemical Data Using the U.S.-EPA CompTox Chemicals Dashboard. J. Chem. Inf. Model. 61, 565–570 (2021). 6. Angelopoulos, A. N. & Bates, S. Conformal Prediction: A Gentle Introduction. Found. Trends® Mach. Learn. 16, 494–591 (2023). 7. Kochev, N., Jeliazkova, N. & Tancheva, G. Ambit‐SLN: an Open Source Software Library for Processing of Chemical Objects via SLN Linear Notation. Mol. Inform. 40, 2100027 (2021). 8. Kochev, N., Avramova, S. & Jeliazkova, N. Ambit-SMIRKS: a software module for reaction representation, reaction search and structure transformation. J. Cheminform. 10, 42 (2018). 9. Kochev, N., Paskaleva, V., Pukalov, O. & Jeliazkova, N. Ambit-GCM: An Open-source Software Tool for Group Contribution Modelling. Mol. Inform. (2019) doi:10.1002/minf.201800138. 10. Willighagen, E. L. et al. The Chemistry Development Kit (CDK) v2.0: atom typing, depiction, molecular formulas, and substructure searching. J. Cheminform. 9, 33 (2017).

Full text

Automated SSbD Workflows Using FAIR Electronic Laboratory Notebooks: Application to Polymer Informatics NINA JELIAZKOVA IDEACONSULT LTD SSBD25 CONFERENCE, ZURICH, SWITZERLAND, 12 NOVEMBER 2025 Climbing the Peaks: Trials, Competition, and Success Matterhorn 1860: Edward Whymper first sees the mountain while sketching the Zermatt region. 1861–1863: Multiple attempts via the Lion (Italian) ridge; storms, injuries, and a 26-hour tent halt force retreats. 14 July 1865: Whymper’s party reaches the summit via the Hörnli ridge; four climbers die on descent. 1865: Carrel’s Italian team climbs the Lion ridge successfully a few weeks later. Eiger North Face 11 August 1858: First ascent of the mountain (west flank) by Charles Barrington with guides 1930s (1934–36): Repeated tragic attempts; the face earned the nickname “Murder Wall” 24 July 1938: First successful ascent of the north face (Nordwand) by the German–Austrian party Climbing the SSbD Mountain: Design Paths (or Directed Acyclic Graph -DAG) Nodes Edges: input/outputs Chemical / Materials Processes Product Example DAG https://sbd.adma.ai Exploring the (SSbD) design space - Alternative synthesis routes Nodes Edges: input/outputs Chemical / Materials Processes Product Re-design (selection node) https://sbd.adma.ai Exploring the (SSbD) design space - Alternative synthesis routes Nodes Chemical / Materials Processes Product Re-design (selection node) Molecular Design space of (ORMOCER) coatings There is also design space for additives, coating processes and textiles… https://sbd.adma.ai Molecular and process design space Each route is a sequence of connected decisions and experiments -a graph. This structure lets us trace evidence and automate documentation. Exploring new alternatives (re-design) means adding new nodes and links. Nodes Chemical / Materials Processes Products Re-design (selection node) green/red colors denote GO/No GO decisions + timestamps https://sbd.adma.ai The DAG implementation backbone – Electronic Lab Notebook Electronic Lab Notebook (eLabFTW) configured to represent essential SSbD resources with structured information -Chemical/Material -Processes -Product application(s) -(re)design nodes while serving as a structured repository for ◦laboratory experiments (with templates) ◦hazard assessments (with templates) ◦computational models ◦environmental sustainability evaluations , etc… Demo at https://elab.sbd.adma.ai/ Resources and experiments are categorized by type and by status. Each resource is assigned a dedicated page with metadata, linked experiments and resources. Resources and experiments can be linked, forming the graph https://sbd.adma.ai ELN Experiments - Lab experiments –physicochemical characterization, bioassays, etc. - Computational models -Hazard assessments -Overall SSbD assessment Important: Resources and Experiments can be linked ! Assessment information can be added either manually or automatically See e.g. DEHP links https://elab.sbd.adma.ai/database.php?mode=vie w&id=1 The ELN is a native environment to store experimental data, and the graph structure ensures linking this information to design decisions. Similarly, computational models are represented as eLabFTW experiments and linked to e.g. hazard or performance assessments. https://sbd.adma.ai ELN is programmatically accessible, information retrieval can be automated Automation is enabled via the eLabFTW API*. External tools can be launched from within the workflow: -QSAR models for property prediction - Performance prediction models -Toxicological databases Combines expert judgment and algorithmic evaluation. Supports consistent, reproducible SSbD decision-making Automated extraction of: -Metadata (materials, processes) -Hazard and performance results -Decision points and justifications Output: -Automatically generated SSbD documentation: - Tabulated hazard summaries -Visual graphs and traceability views (next slide) *API (Application Programming Interface)