ChemAudit

Process CSV, TSV, TXT, or SDF files of molecules through ChemAudit's batch pipeline with Redis-backed progress tracking, on-demand analytics (scaffold, chemical space, clustering, MMP, taxonomy, R-group), and nine export formats. Use when user says "validate this file", "batch validation", "process these molecules", "upload CSV of SMILES", "export results", "scaffold analysis", "cluster compounds", "Butina clustering", or "PDF report for these molecules". Deployment-profile-gated file size and molecule-count limits — query /config first.

Look up molecules across PubChem, ChEMBL, COCONUT, Wikidata, and SureChEMBL patent databases, resolve chemical identifiers (CAS, ChEMBL ID, PubChem CID, ChEBI, UNII, DrugBank, names), and run cross-database consistency checks. Use when user says "look up in PubChem", "ChEMBL bioactivity", "natural product search", "COCONUT lookup", "Wikidata lookup", "resolve CAS number", "resolve ChEMBL ID", "cross-database check", "find this compound", "patent search", "SureChEMBL", or "identifier resolution". Supports SMILES, InChI, InChIKey, CAS, ChEMBL ID, PubChem CID, ChEBI, UNII, DrugBank ID, Wikipedia URL, and compound names (OPSIN + PubChem fallback).

Audit dataset health with a Fourches-style 5-component score, detect contradictory bioactivity labels across duplicates, compare two dataset versions (added / removed / modified / unchanged), and generate curation reports. Use when user says "audit this dataset", "dataset health score", "contradictory labels", "duplicate activity conflict", "dataset diff", "compare dataset versions", "curation report", "data quality check", or "clean this dataset for ML". Accepts CSV and SDF with optional activity columns.

Diagnose SMILES parse errors with position and fix suggestions, compare InChI strings layer-by-layer, check format round-trip lossiness (SMILES→InChI→SMILES and SMILES→MOL→SMILES), compare three standardization pipelines side-by-side, and pre-validate SDF/CSV files for structural integrity. Use when user says "why does this SMILES fail", "diagnose this structure", "InChI diff", "layer comparison", "round-trip check", "compare InChI strings", "file pre-validation", "SDF integrity", "M END missing", or "fix this SMILES error".

Validate and score chemical structures using ChemAudit's 16 deep validation checks, 1,500+ structural alerts, and multi-rule drug-likeness scoring. Use when user says "validate this molecule", "check this SMILES", "is this compound valid", "ML-readiness score", "drug-likeness", "deep validation", "quality score", "PAINS check", or asks about stereochemistry, valence, tautomers, or structural issues. Supports SMILES, InChI, MOL blocks, IUPAC names (OPSIN + PubChem), and database identifiers (ChEMBL ID, CAS, PubChem CID, InChIKey) via the /resolve endpoint.

Run molecules through ChemAudit's configurable 10-step QSAR-Ready curation pipeline for ML-ready SMILES with full per-step provenance and InChIKey-change detection. Use when user says "QSAR ready", "prepare for QSAR", "curate for ML", "standardize dataset for modeling", "strip salts and desalt", "remove duplicates by InChIKey", "canonicalize tautomers for ML", "prepare SMILES for machine learning", "QSAR-2D curation", or "QSAR-3D curation". Supports single molecules and batch CSV/SDF files with pasted SMILES fallback.

Standardize chemical structures using ChemAudit's ChEMBL-compatible pipeline (Checker → Standardizer → GetParent → optional Tautomer) with full per-stage provenance, charge/bond/ring-change tracking, and stereochemistry comparison. Use when user says "standardize this molecule", "normalize structure", "strip salts", "neutralize charges", "canonical SMILES", "ChEMBL standardization", "parent molecule", "tautomer canonicalization", or "show me what changed after standardization". Shows every transformation applied with before/after InChIKey and atom-level diffs.

Filter generative-chemistry outputs through ChemAudit's 6-stage multi-stage funnel with a composite 0-1 scorer and REINVENT 4-compatible scoring endpoint. Use when user says "filter molecules", "drug-like filter", "structure filter", "filter generative output", "REINVENT scoring", "REINVENT Component", "lead-like filtering", "fragment filter", "funnel these SMILES", "novelty filter", or "score generative batch". Four built-in presets (drug_like, lead_like, fragment_like, permissive). Sync for ≤1000 SMILES, async for larger.