Now in Beta — run your first in silico assessment free, no credit card required.
About MolWard

Next-generation in silico toxicology, stability, CMC & degradation profiling

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.

What is MolWard?

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 problem we solve

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.

Our core technologies

MolWard operates on robust cheminformatics tools, deterministic chemistry rules, regulatory statistics, and modern machine learning.

  • Predictive Forced-Degradation Engine. Simulates severe stress (extreme pH, oxidative stress, photolysis, elevated temperature) using a proprietary library of 50+ deterministic reaction rules — generating theoretical degradant structures and calculating EI-MS fragmentation, UV-Vis spectra, and pKa profiles.
  • Stability (ICH Q1E). Fits the mean regression line through long-term stability data, constructs the one-sided 95% lower confidence bound, and reports the exact time it intersects your lower specification limit — your projected shelf-life — while automatically flagging out-of-trend points.
  • Dual-Methodology ICH M7(R2) Assessment. A two-pronged evaluation: an expert rule-based system identifying established toxicophores (Ashby-Tennant alerts, Benigni-Bossa rules), and a statistical QSAR Random Forest classifier — trained on 6,700+ compounds from the Hansen Mutagenicity and ECVAM databases — predicting Ames mutagenicity independently of structural alerts.
  • Dynamic Nitrosamine CPCA Scoring. Executes the Carcinogenic Potency Categorization Approach by computationally parsing the steric environment and alpha-hydrogen count of the pharmacophore, assigning the exact FDA/EMA Acceptable Intake limit (from 18 to 1500 ng/day).
  • Structural Read-Across & Pharmacopeia Integration. Mutagenic predictions are backed by literature references from a known-toxin database, and theoretical degradants are cross-referenced against USP and Ph. Eur. databases — giving analytical chemists direct catalog numbers for reference standards.
  • ICH Q3 Impurity Limits Suite. Applies the published ICH Q3A(R2)/Q3B(R2) reporting, identification and qualification thresholds, Q3C(R9) residual-solvent PDEs, Q3D(R2) elemental-impurity PDEs by route, and the Q3E / USP <1663> analytical evaluation threshold — every limit derived from a single maximum daily dose.
  • Dissolution Equivalence (f2). Computes the FDA/EMA similarity (f2) and difference (f1) factors between reference and test dissolution profiles, with automatic multi-point weighting rules and a Weibull/first-order model fit — for scale-up, post-approval change, and biowaiver justification.
  • RP-HPLC Chromatogram Prediction. Applies Linear Solvent Strength theory to a molecule's pH-corrected log D across a standardized column/mobile-phase grid, ranking conditions by predicted resolution and recommending a detection wavelength for method scouting.
  • Column Selectivity (HSM / Fs). Scores reversed-phase HPLC columns on the published Hydrophobic-Subtraction Model's five orthogonal terms, ranking equivalent (swap-ready) and orthogonal columns by Factor of Similarity (Fs) for method transfer and impurity separation.
  • FDA HED & MRSD Dose Scaling. Normalises an animal dose by body surface area (Km ratio) across 13 species to compute the Human Equivalent Dose, then derives the Maximum Recommended Starting Dose for first-in-human trials per the FDA (2005) guidance.
  • PDE / ADE / OEL & MACO Limits. Derives the Permitted/Acceptable Daily Exposure from a NOAEL through the full ICH Q3C / EMA F1–F5 modifying-factor chain, then extends it to the Occupational Exposure Limit, an OEB containment band, and the cleaning-validation maximum allowable carryover (MACO).

Scientific validation

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.

13
ICH / FDA-aligned engines
77.71%
Cross-validated accuracy
0.869
ROC-AUC discriminatory power
6,700+
Curated empirical Ames records
18–1500
ng/day AI limits assigned (CPCA)
The Science Behind Each Tool

Deterministic by design — no black boxes, no hallucinated conditions

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.

Guideline-exact math

Regulatory thresholds and limits are the literal equations from the source guideline — reproducible and auditable.

No hallucination

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.

Realistic bounds

Extrapolation caps, standardized conditions and physical limits are enforced, so results never drift into scientifically implausible territory.

Degradation

Forced-Degradation Prediction

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.

Stability · ICH Q1E

Shelf-Life Projection

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.

Toxicology · ICH M7

Mutagenicity Assessment

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.

Impurity Limits · ICH Q3

Impurity, Solvent & Elemental Limits

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>.

Dissolution Equivalence · FDA / EMA

f2 Similarity & Model Fitting

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.

f2 = 50 · log{ [1 + (1/n)·Σ(Rt−Tt)²]^-0.5 · 100 }

Grounded in: FDA SUPAC-IR / Guidance on dissolution testing, EMA guideline on the investigation of bioequivalence (f2).

Dose Scaling · FDA 2005

HED & MRSD

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.

HED (mg/kg) = animal dose · (Km_animal ÷ Km_human) MRSD (mg/kg) = HED ÷ safety factor (≥ 10 default)

Grounded in: FDA (2005) "Estimating the Maximum Safe Starting Dose", Km factors for 13 species.

Exposure Limits · ICH Q3C / EMA

PDE / ADE / OEL & MACO

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.

PDE = ADE = (NOAEL · body weight) ÷ (F1·F2·F3·F4·F5) OEL (mg/m³) = ADE ÷ inhalation volume (8-h TWA) MACO (mg) = ADE_prev · MinBatch_next ÷ MaxDose_next

Grounded in: ICH Q3C(R9) Appendix 3, EMA shared-facility guideline (2014), ISPE Risk-MaPP / EMA HBEL.

Chromatogram Predictor · RP-HPLC

Method Scouting

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.

Column Selectivity · HSM / Fs

HPLC Column Comparison

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.

log(k / kref) = η'H − σ'S* + β'A + α'B + κ'C

Grounded in: Snyder–Dolan Hydrophobic-Subtraction Model, USP <621> column-equivalence principles.

QbD DoE & Design Space · ICH Q8(R2)

Design of Experiments & Design Space

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.

Ŷ = b₀ + Σ bᵢxᵢ + Σ bᵢⱼxᵢxⱼ + Σ bᵢᵢxᵢ² Design space = { x : every response within its limits }

Grounded in: ICH Q8(R2) Pharmaceutical Development, classical response-surface methodology (Box–Hunter–Hunter / Montgomery), least-squares regression and Monte-Carlo risk analysis.

QC Calculators · USP / ICH

Everyday Analytical QC

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.

T = W₀.₀₅ / (2f) · N = 16(t_R/W)² · AV = |M − X̄| + k·s LOD = 3.3σ/S · LOQ = 10σ/S · Grubbs G = |x − X̄| / s

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.

RegIntel · ICH · 21 CFR · FDA

Citation-Grounded Regulatory Search

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.

Who Uses MolWard

A daily-use utility for cross-functional teams

Analytical Chemists

Anticipate chromatographic peaks and source USP/EP reference standards during SIM development.

Stability & QA Teams

Project shelf-life and investigate out-of-trend pulls with ICH Q1E statistics and an audit-ready report.

Regulatory Toxicologists

Generate rapid, defensible dual-methodology risk assessments for regulatory submissions.

Process Chemists

Identify synthesis-route risks, particularly concerning nitrosating agents and amine precursors.

CMC & Process Development

Scout chromatography columns and mobile phases, and compare dissolution profiles by f2 for scale-up and biowaivers.

Built on rigorous, transparent science

Experience regulatory-grade in silico assessment, free during our Beta.