Safety and Hazard Assessment in Material Design: The In Silico Revolution Meets SSbD

How QSAR, read-across, AOP models, and the EU Safe & Sustainable-by-Design framework turn safety assessment from a late-stage gate into a design-time decision

Safety and hazard assessment used to arrive late in the materials development cycle — after kilograms of a candidate had been synthesized, characterized, and ready for formulation trials. When a late-stage hazard finding forced reformulation, months of work evaporated. In 2026, in silico toxicology has matured enough to flip that timeline: QSAR, read-across, AOP-based models, and hybrid machine-learning toxicity predictors now run on candidate molecules in seconds, letting designers eliminate hazardous chemistries before they are synthesized. Layered with the EU Safe and Sustainable-by-Design (SSbD) framework, this is the operational toolkit that makes “benign by design” a working principle rather than a slogan. Simreka embeds these capabilities directly into sustainable-formulation workflows.

The In Silico Toolbox

Modern in silico hazard assessment is not one technique but a stack. At the base are QSAR models — quantitative structure-activity relationships that predict toxicological endpoints from molecular descriptors. Above them sit read-across methods, which infer a chemical’s properties from structurally and mechanistically similar substances with experimental data. AOP-based models (Adverse Outcome Pathway) link molecular-initiating events to apical endpoints through biologically meaningful chains. Dose-response and toxicokinetic models estimate internal exposure, not just intrinsic hazard. And toxicogenomics-based approaches add gene-expression signals to the prediction of outcomes. Together, this stack can produce defensible hazard profiles for chemicals where no animal studies have been conducted.

Hybrid ML for Eight Critical Toxicity Endpoints

Recent hybrid ML toxicity models cover eight endpoints that matter across consumer, industrial, and occupational uses: cardiac toxicity, inhalation toxicity, dermal toxicity, oral toxicity, skin irritation, skin sensitization, eye irritation, and respiratory irritation. Models combining classical QSAR descriptors with modern ML (graph neural networks, transformer embeddings) consistently outperform either approach alone, with performance good enough to replace many standard screening-level animal tests. The growing acceptance of these models in regulatory submissions — particularly in the EU under REACH and the US under TSCA — is redefining what a “sufficient safety dossier” looks like for a new substance.

The Regulatory Toolbox: OECD QSAR Toolbox, VEGA, Danish (Q)SAR

Safe and Sustainable-by-Design (SSbD)

The EU’s Safe and Sustainable-by-Design framework, developed by the JRC, structures safety and sustainability assessment across the chemical and material lifecycle. It defines five evaluation steps: hazard assessment of the chemical/material, health and safety aspects across the life cycle, human health and environmental aspects of the production, environmental sustainability of the final application, and socio-economic aspects. In practice, SSbD formalizes what in silico hazard assessment and LCA have been doing separately — bringing them under one decision framework so a candidate is not judged safe if it merely meets one dimension while failing others.

Read-Across as the Pragmatic Workhorse

Despite the sophistication of modern ML toxicity models, read-across remains the everyday workhorse of industrial hazard assessment. When experimental data exists for structurally or mechanistically similar substances in high-quality databases, that data supports extrapolation to the target chemical with regulator-acceptable defensibility. The key is methodological rigor: clearly defined similarity criteria, transparent applicability-domain checks, and uncertainty quantification. Done well, read-across shortens safety-data-collection timelines from years to weeks.

What Industry Actually Deploys

A mature 2026 safety-assessment workflow for a new formulation candidate looks like this: (1) compute QSAR-predicted toxicity for each component across the eight critical endpoints using a validated hybrid ML model; (2) run read-across through the OECD QSAR Toolbox with documented similarity justification; (3) layer on AOP analysis for any endpoints with biologically meaningful pathways; (4) estimate internal exposure via PBPK models; (5) combine hazard and exposure into risk; (6) flag any candidates exceeding acceptable thresholds for reformulation. The whole workflow runs in hours on a candidate shortlist, not weeks on single molecules.

Common Traps

Three traps catch teams new to in silico hazard assessment. First, applicability domain blind spots: QSAR predictions outside the chemical space the model was trained on are unreliable and must be flagged as such. Second, false precision: point predictions without uncertainty bands mislead decision-makers; always report confidence. Third, regulatory mismatch: not every in silico prediction is accepted by every regulator for every endpoint; knowing the acceptance status of your tools is as important as the predictions themselves.

How Simreka Integrates Safety Assessment Into Formulation

The Simreka Regulatory Compliance module runs QSAR-style hazard screens and regulatory-list checks on every candidate component and formulation, surfacing risk before a recipe is finalized. The Simreka AI-Formulator treats toxicity endpoints as multi-objective constraints, so generated candidates that would fail safety screens are excluded from the shortlist rather than discovered later. The Simreka LCA & Impact Assessment module complements hazard assessment with cradle-to-grave impact data, implementing the SSbD multi-dimensional view. The Simreka Recycled & Alternative Materials module surfaces alternative feedstocks when a substance fails hazard screening, so reformulation is same-day instead of multi-month.

Conclusion

Safety and hazard assessment has moved from a late-stage gate to a design-time decision. QSAR, read-across, AOP models, and hybrid ML toxicity predictors now give materials designers the ability to screen out hazardous candidates before they are synthesized. The EU’s Safe and Sustainable-by-Design framework wraps these tools into a multi-dimensional evaluation that aligns with regulatory direction of travel. The organizations that embed in silico safety inside their formulation platforms — as a daily input rather than an occasional audit — will design safer, more sustainable materials faster, and avoid the costly late-stage surprises that have traditionally plagued the industry.

Frequently Asked Questions

Are QSAR predictions accepted by ECHA and the US EPA?

Yes, with caveats. Both agencies accept QSAR for specific endpoints when accompanied by robust applicability-domain analysis, validation statistics, and expert judgment. The OECD QSAR validation principles are the common benchmark.

Can in silico toxicology fully replace animal testing?

For screening-level decisions and many regulatory purposes, increasingly yes. For novel modes of action or complex human-health endpoints, in vitro and targeted in vivo studies remain necessary.

How do I document a read-across justification?

Use the ECHA Read-Across Assessment Framework (RAAF). Document source chemicals, similarity rationale, mechanistic support, applicability domain, and uncertainty analysis. Expect scrutiny.

What does SSbD add beyond existing safety and LCA workflows?

Structure and integration. Individual tools exist separately; SSbD pulls them into a single multi-step evaluation so a chemical is judged across hazard, life-cycle, production, application, and socio-economic dimensions together.

How do I handle UVCB or mixture substances?

Standard QSAR is challenged by UVCBs. Approaches include representative-component assessment, fraction profiling, and mixture-specific AOP reasoning. The OECD QSAR Toolbox has modules for UVCBs; expect higher uncertainty.

What is the right organizational model for in silico safety?

Computational toxicology embedded in the R&D organization (not sitting in regulatory affairs only) so that predictions inform design rather than audit it. Cross-functional teams combining computational, regulatory, and formulation expertise work best.

Bibliographical Sources

1. ScienceDirect. In Silico Toxicology Protocols. https://www.sciencedirect.com/science/article/pii/S0273230018301144
2. ACS Omega. QSAR Classification Modeling Using ML with a Consensus-Based Approach. https://pubs.acs.org/doi/10.1021/acsomega.4c09356
3. Environmental Science & Technology. Predictive Models and Integration Strategies for Chemicals and Materials Impact Assessment. https://pubs.acs.org/doi/10.1021/acs.est.5c04489
4. JRC. A Framework for Assessing In Silico Toxicity Predictions. https://publications.jrc.ec.europa.eu/repository/bitstream/JRC62586/lbna24705enc.pdf
5. PMC. In Silico Approaches in Organ Toxicity Hazard Assessment (Liver). https://pmc.ncbi.nlm.nih.gov/articles/PMC8955833/
6. PMC. In Silico Prediction of Toxicity and Its Applications for Chemicals at Work. https://pmc.ncbi.nlm.nih.gov/articles/PMC7223298/
7. PMC. In Silico Toxicology Data Resources to Support Read-Across and (Q)SAR. https://pmc.ncbi.nlm.nih.gov/articles/PMC6580867/
8. PMC. Implementation of In Silico Toxicology Protocols within a Visual and Interactive Hazard Assessment Platform. https://pmc.ncbi.nlm.nih.gov/articles/PMC8754399/
9. ScienceDirect. The Predictivity of QSARs for Toxicity — Recommendations. https://www.sciencedirect.com/science/article/pii/S2468111324000409
10. QSAR Lab. In Silico Methods in the Safety Assessment of Chemicals. https://www.qsarlab.com/en/in-silico-methods-in-the-safety-assessment-of-chemicals/

Make Every Formulation Safer by Design

Screen for hazards at the candidate stage, not the audit stage. Request a Simreka Demo → and see QSAR and read-across embedded directly in generative formulation workflows.

Tag Cloud