--- title: Safety Assessment description: >- Evaluate drug candidate safety with multi-endpoint toxicity prediction and adverse event analysis. tags: - workflows - safety-assessment --- ## Safety Assessment Safety failures are the leading cause of drug candidate attrition in clinical development. This workflow combines computational toxicity prediction with real-world adverse event data to build a comprehensive safety profile for drug candidates early in the discovery process, enabling informed go/no-go decisions before committing to costly preclinical studies. ## Overview A thorough safety assessment integrates multiple lines of evidence: predicted acute and chronic toxicity from molecular structure, organ-specific risk scores, metabolic liability alerts, structural promiscuity flags, and historical adverse event data from marketed drugs with similar structures or mechanisms. ADMET-AI provides toxicity and safety-related ADMET endpoints from SMILES input, and `safety_classify` gives a SAFE, CAUTION or DANGEROUS verdict from antitarget screening, structural alerts and drug-drug interaction potential. ProTox3 and ADMETLab3 are retired; no tool on the platform predicts an acute-toxicity class (LD50) or organ-specific toxicity scores. FAERS (FDA Adverse Event Reporting System) data contextualizes predictions against real-world post-marketing surveillance from structurally or mechanistically related drugs. ## Tool Chain 1. [ADMET-AI](/docs/admet-ai) -- Toxicity and safety-related ADMET endpoints 2. `safety_classify` -- Safety verdict from antitarget screening, structural alerts and DDI potential 3. FAERS Query -- FDA adverse event database search (via chat or Safety Passport) ## Prerequisites - SMILES strings for candidate compounds - Knowledge of the target and mechanism of action - Understanding of the intended patient population and route of administration ## Step-by-Step Guide ### Step 1: Predict toxicity endpoints with ADMET-AI Submit your candidate SMILES to ADMET-AI. Review the toxicity endpoints it covers: hERG, AMES, DILI, ClinTox, Tox21 and CYP inhibition. Acute-toxicity class (LD50) and organ-specific toxicity scores are not available from any tool on the platform, so treat them as open questions for experimental work. ### Step 2: Get a safety verdict Run `safety_classify` and review the verdict and the structural alerts behind it. Compounds with reactive functional groups (Michael acceptors, acyl halides, epoxides) or known toxic substructures require careful risk-benefit evaluation. Not all alerts are disqualifying -- aspirin triggers several structural alerts but remains one of the most widely used drugs. ### Step 3: Profile metabolism and safety endpoints Focus on the safety-relevant ADMET-AI endpoints: hERG channel inhibition (cardiac QT prolongation risk), CYP450 inhibition panel (drug-drug interaction potential), AMES mutagenicity and skin sensitization. Compare them with the `safety_classify` verdict: an endpoint flagged by both carries more weight than one flagged by either alone. ### Step 4: Query FAERS for class-level signals Search the FAERS database for adverse events reported against marketed drugs in the same pharmacological class or with similar structural scaffolds. This contextualizes your predictions with real-world evidence. Look for disproportionality signals (PRR greater than 2, case count above 3) in organ systems flagged by the computational predictions. For example, if ADMET-AI flags DILI risk, check whether related drugs in FAERS show elevated liver injury reports. ### Step 5: Generate the Safety Passport Use the Safety Passport to consolidate all findings into a structured report. The passport aggregates ADMET-AI safety endpoints, FAERS signals, and FDA warning data into a single document with organ-system risk scores and evidence summaries. This serves as the primary safety dossier for decision-making. ### Step 6: Make risk-informed decisions Review the consolidated safety profile in the context of your therapeutic indication, patient population, and competitive landscape. An oncology candidate may tolerate higher toxicity risk than a preventive medication. Document the risk assessment, mitigation strategies (dose limitations, monitoring plans, formulation changes), and the rationale for advancing or terminating each candidate. ## Expected Outputs - ADMET-AI safety endpoint predictions (hERG, AMES, DILI, CYP450) - `safety_classify` verdicts (SAFE, CAUTION or DANGEROUS) - FAERS disproportionality analysis for related drug classes - Safety Passport with consolidated organ-system risk scores - Exportable safety dossier in PDF format ## Tips and Best Practices - Run safety assessment in parallel with ADMET optimization, not as a sequential gate. Early safety signals should inform structural modifications in real time. - Use consensus predictions: flag an endpoint as high-risk only when at least two independent tools agree. Single-tool alerts warrant monitoring but not necessarily termination. - Compare your candidate's toxicity profile against approved drugs in the same indication. Many successful drugs carry known risks that are managed through labeling and monitoring. - For candidates with borderline hepatotoxicity predictions, consider requesting DILI (drug-induced liver injury) risk classification from literature sources via BioBERT. - Document all safety findings and decision rationale. Regulatory agencies expect a clear safety narrative in IND submissions. - Revisit safety assessment after each round of structural optimization to verify that modifications addressing ADMET liabilities have not introduced new toxicity signals. ## Run it from a prompt The same goal as a copy-and-adapt recipe you can paste into a chat: - [Head-to-head safety comparison](/docs/workflows/safety-comparison) — compare two or three drugs on off-targets, adverse-event signals, and pharmacogenomics in one prompt. - [Compound safety profile](/docs/workflows/safety-profile) — off-target, drug–drug interaction, and toxicology prompts for one compound.