A comprehensive, web-based cheminformatics platform — a dozen validated engines plus RegIntel, a citation-grounded regulatory assistant — designed to accelerate early-stage pharmaceutical development and streamline regulatory compliance.
MolWard unifies predictive degradation profiling, ICH Q1E long-term and accelerated (Arrhenius) stability analysis, and a dual-methodology ICH M7 hazard assessment into a single automated workflow, alongside a growing set of deterministic regulatory calculators and CMC method tools — ICH Q3 impurity limits, dissolution f2 equivalence, RP-HPLC chromatogram prediction and column selectivity, FDA HED/MRSD dose scaling, and PDE/ADE/OEL cleaning-validation limits. It empowers pharmaceutical scientists to identify chemical liabilities, project shelf-life, predict mutagenic risks, scout analytical methods, and set defensible exposure limits — before ever stepping into the laboratory.
The modern pharmaceutical landscape faces unprecedented regulatory scrutiny over drug-substance stability and mutagenic impurities. Following recent global safety crises involving the "Cohort of Concern" (specifically N-nitrosamines), agencies such as the FDA and EMA now mandate highly stringent safety thresholds.
Traditionally, identifying these risks required resource-intensive experimental forced-degradation studies, long-term stability programs, and costly in vitro Ames testing — empirical methods that frequently create substantial bottlenecks in early drug formulation. MolWard was engineered to eliminate these bottlenecks by providing rapid, highly accurate computational foresight.
MolWard operates on robust cheminformatics tools, deterministic chemistry rules, regulatory statistics, and modern machine learning.
MolWard is built on rigorous, transparent science. Our statistical QSAR model has been strictly validated against unseen test sets, achieving 77.71% accuracy and an exceptional ROC-AUC of 0.869 — ensuring a high degree of confidence and a low rate of false positives. Every prediction is further supported by PCA applicability-domain verification and SHAP feature interpretability. Every Q1E interpretation is guard-railed to the analyzer's computed numbers and a code-controlled list of real ICH citations.
Every MolWard tool applies published, peer-reviewed science and official regulatory equations to your inputs. Where a value is defined by an ICH, FDA, EMA or USP guideline, we compute the exact formula from the guideline — nothing is invented, approximated for convenience, or generated by a language model. Where a prediction is model-based, it runs on transparent chemistry rules or a validated statistical model with a stated applicability domain, and every result is framed as a screening aid for a qualified reviewer.
Regulatory thresholds and limits are the literal equations from the source guideline — reproducible and auditable.
No generative AI sets a number, limit, or condition. Any AI-drafted text is guard-railed to the computed values and an allow-list of real citations.
Extrapolation caps, standardized conditions and physical limits are enforced, so results never drift into scientifically implausible territory.
A deterministic library of expert reaction rules maps hydrolytic, oxidative, photolytic and thermal pathways from the structure. Each theoretical degradant is characterised with rule-based EI-MS fragmentation, Woodward–Fieser UV estimation and calculated physicochemical descriptors — every proposed species traces back to a documented degradation chemistry rule, not a guess.
Grounded in: established organic degradation mechanisms and ICH Q1A(R2) stress-testing principles. Deterministic — the same structure always yields the same pathways.
Long-term stability data are fit by least-squares linear regression; the shelf life is the time at which the one-sided 95% confidence bound of the mean line meets the specification limit — exactly as ICH Q1E prescribes. Multi-batch data are pooled only when ANCOVA supports it (Appendix A), and extrapolation is capped to the guideline limit.
Grounded in: ICH Q1E, Q1A(R2), Q1D, Q6A. AI-drafted interpretation is restricted to the computed numbers and an allow-listed citation set.
A dual methodology per ICH M7(R2): an expert rule-based system for established structural alerts, run in parallel with a statistical QSAR (Random Forest) trained on 6,700+ curated Ames records. Predictions carry a PCA applicability-domain check and SHAP interpretability. Nitrosamine acceptable-intake limits follow the FDA CPCA scheme.
Grounded in: ICH M7(R2), FDA CPCA guidance, Ashby–Tennant & Benigni–Bossa alerts, Hansen/ECVAM datasets.
Every threshold is the exact figure from the ICH Q3 series, selected by the "whichever is lower" rule and scaled to the maximum daily dose you enter. Nothing is interpolated — the tool reproduces the guideline tables.
Grounded in: ICH Q3A(R2), Q3B(R2), Q3C(R9), Q3D(R2), Q3E / USP <1663>.
The similarity factor f2 (and difference factor f1) is computed on reference and test dissolution profiles exactly per the FDA/EMA model-independent method, applying the "one measurement above 85%" and coefficient-of-variation eligibility rules automatically. A supplementary Weibull/first-order kinetic fit characterises release shape for scale-up and post-approval change dossiers.
Grounded in: FDA SUPAC-IR / Guidance on dissolution testing, EMA guideline on the investigation of bioequivalence (f2).
The Human Equivalent Dose is the animal dose normalised by body surface area using the FDA-assigned Km factors; the Maximum Recommended Starting Dose divides the HED of the most sensitive species by a safety factor. These are the FDA (2005) equations, applied verbatim.
Grounded in: FDA (2005) "Estimating the Maximum Safe Starting Dose", Km factors for 13 species.
The Permitted / Acceptable Daily Exposure is derived from a NOAEL through the full ICH Q3C Appendix 3 modifying-factor chain (F1–F5), then extended to an occupational limit and cleaning-validation carryover. Each factor is shown transparently so the derivation can be checked line by line.
Grounded in: ICH Q3C(R9) Appendix 3, EMA shared-facility guideline (2014), ISPE Risk-MaPP / EMA HBEL.
Retention is modelled with established Linear Solvent Strength (LSS) theory: a molecule's pH-corrected lipophilicity (log D) sets its retention factor, which is modulated by the organic strength of the mobile phase and the chemistry of the column. Peak resolution follows the standard plate-count equation, and the recommended wavelength is chosen so every analyte remains detectable. This is an explicitly semi-quantitative scouting aid — it predicts elution order, resolution ranking and wavelength on a standardized column grid, not bench-ready absolute retention times.
Grounded in: Linear Solvent Strength theory (Snyder–Dolan), USP <621> system-suitability principles. Descriptors (log P, pKa, UV) are computed with RDKit; the retention model runs deterministically in the browser.
Each of 817 real, published reversed-phase columns is characterised on the Hydrophobic-Subtraction Model's five orthogonal interaction terms — hydrophobicity, steric resistance, and hydrogen-bond acidity/basicity/ionic interaction. The Factor of Similarity (Fs) distance between two columns' term sets ranks equivalent, swap-ready backups (Fs ≤ 3) and orthogonal columns for maximum selectivity change (Fs ≥ 10) to pull apart co-eluting peaks.
Grounded in: Snyder–Dolan Hydrophobic-Subtraction Model, USP <621> column-equivalence principles.
Classical experimental designs — full and fractional factorial, Plackett-Burman, Box-Behnken and central-composite — are generated in coded units, then a response-surface model is fitted by ordinary least squares with ANOVA, a formal lack-of-fit test and a Pareto of standardized effects. The ICH Q8(R2) design space is mapped as the multivariate region where every response simultaneously meets its acceptance criterion, and a Monte-Carlo simulation propagates model and residual uncertainty to trace the edge of failure. Every coefficient, p-value and model equation is shown so the analysis can be checked line by line.
Grounded in: ICH Q8(R2) Pharmaceutical Development, classical response-surface methodology (Box–Hunter–Hunter / Montgomery), least-squares regression and Monte-Carlo risk analysis.
The compendial calculations an analyst runs every day, applied exactly as written and computed locally in the browser. System suitability follows USP <621> — including the distinction analysts most often get wrong, the USP tailing factor measured at 5% peak height versus the EP symmetry factor measured at 10%. Buffer recipes are solved from Henderson-Hasselbalch so both salts are weighed directly rather than titrated by eye; LOD/LOQ follow the two ICH Q2 approaches; the USP <905> acceptance value applies the full Stage-1/Stage-2 conditional logic; and outliers are screened with the Grubbs and Dixon tests against their published critical values.
Grounded in: USP General Chapters <621> and <905>, Ph.Eur. symmetry factor, ICH Q2(R2), Henderson-Hasselbalch, and the Grubbs/Dixon outlier tests. Every result is a direct application of the published formula — nothing is fabricated.
A retrieval-augmented assistant that answers regulatory questions only from a curated corpus of source documents — ICH guidelines, the U.S. Code of Federal Regulations (21 CFR), FDA guidances, warning letters and annual reports. Every answer returns the verbatim source passage and a link; the model is numerically and citation guard-railed, and it refuses to answer beyond the corpus rather than guess. No hallucinated citations, ever — true to the same "no black box" principle as every other MolWard tool.
Grounded in: the retrieved source passages themselves — ICH, 21 CFR, and FDA guidances / warning letters / annual reports — enforced by a numeric guard and an allow-listed citation set; every answer is labelled a draft for expert review.
Anticipate chromatographic peaks and source USP/EP reference standards during SIM development.
Project shelf-life and investigate out-of-trend pulls with ICH Q1E statistics and an audit-ready report.
Generate rapid, defensible dual-methodology risk assessments for regulatory submissions.
Identify synthesis-route risks, particularly concerning nitrosating agents and amine precursors.
Scout chromatography columns and mobile phases, and compare dissolution profiles by f2 for scale-up and biowaivers.
Experience regulatory-grade in silico assessment, free during our Beta.