How the tools work.

Molecule Lab

Complete analysis parses a SMILES graph using RDKit and preserves the submitted text. RDKit Cleanup applies functional-group normalization and reionization. Fragment selection, additional neutralization and canonical tautomer selection are explicit choices. Every operation records its input and output canonical SMILES.

The descriptor report includes molecular weight, exact mass, Wildman-Crippen logP, TPSA, hydrogen-bond counts, rotatable bonds, fraction sp³ and ring counts. TPSA uses RDKit's default nitrogen/oxygen contribution model. The Morgan fingerprint uses radius 2, 2,048 bits and chirality. Bemis-Murcko scaffold extraction is reported separately; acyclic molecules have no ring-and-linker scaffold.

PAINS and BRENK are the catalogues distributed with the reported RDKit version. Matches identify motifs for review. They do not establish experimental interference, toxicity or inactivity. No biological predictor or applicability-domain model is included.

Local inspection uses RDKit.js in the browser and a smaller, explicitly named alert catalogue. It performs canonicalization rather than the complete standardization workflow. Its results can differ from the server's version and processing choices.

Identity evidence, batch analysis and reports

Molecule Lab records non-isomeric and isomeric canonical SMILES, InChI/InChIKey, component identities, formula, mass and potential stereochemistry. These check a molecular graph; they do not verify an isolated compound experimentally. Multiple components may represent salts, solvates or mixtures, and are flagged for review.

QED uses RDKit default mean weights. Lipinski and Ghose use RDKit descriptors; Veber uses its TPSA branch. The Egan-style rectangular thresholds are cLogP ≤ 5.88 and TPSA ≤ 131.6 Ų. Each criterion and measured value is exported. These implementations are not guaranteed identical to other calculators. No oral efficacy or safety verdict is inferred.

Natural-product likeness uses the RDKit Contrib NP_Score public model with its SHA-256 checksum. Model fragment coverage is reported separately from the score: it is not a calibrated probability or a validated applicability-domain assessment. Biosynthetic family and novelty are not assigned. CNS MPO is not calculated without sourced logD and pKa values.

Batch input supports SMILES, CSV and SDF with up to 500 records, processed in chunks of ten. Failed or uncomputed records are retained. Repeated standardized InChIKeys are flagged without deleting records. The evidence ZIP contains methods, CSV/JSON records, processed structures, SVG figures, a PDF text summary and checksums. Imported docking or external ADMET files remain explicitly user-supplied attachments. Checksums establish file integrity, not scientific certification.

• QED: Bickerton et al.
• Lipinski et al.
• Veber et al.
• Egan et al.
• Ghose et al.
• Natural-product likeness: Ertl et al.

Docking Lab: scope and recorded evidence

Each catalogue entry identifies a target-specific setup and its recorded reference evidence. Use the protocol and software versions in the exported experiment record when reporting a result. Reference recovery alone does not establish prospective ranking accuracy, binding affinity, selectivity or biological activity.

The table lists recorded reference benchmarks. Each workspace checks its installed protocol before accepting new experiments.

ADMET Lab: prediction models and score interpretation

ADMET Lab uses ADMET-AI 2.0.1, an open-source prediction engine built on Chemprop v2 directed-message-passing neural-network ensembles (Swanson et al., 2024). Models were trained and evaluated on benchmark datasets distributed through the Therapeutics Data Commons (TDC). The interface queries a hosted ADMET-AI prediction service; model weights and benchmark datasets are documented in the upstream repository.

The 12-endpoint panel covers absorption and distribution, CYP-mediated metabolism, and safety screening. Classification endpoints return a 0-1 model score; they are not calibrated clinical probabilities. Regression endpoints return a predicted physical value in the units reported by the upstream model. Dataset size and AUROC/AUPRC benchmark metrics are attached to each endpoint card; they describe evaluation-set performance, not per-molecule confidence.

Derived readable labels (e.g., "Low inhibition signal", "Favorable intestinal absorption") are generated from fixed score thresholds applied uniformly. They are a convenience summary for compound prioritization, not a clinical verdict. Thresholds are documented in the source implementation.

For unusual chemical scaffolds, including complex natural products, macrocycles, and organometallic compounds, predictions may fall outside the training distribution. Use predictions to plan follow-up work and confirm relevant properties experimentally. They are not a basis for clinical or regulatory safety claims.

Endpoint panel and dataset sources

• ADMET-AI repository (Swanson et al., 2024)
• Therapeutics Data Commons
• HIA: Hou et al., 2007
• AqSolDB: Sorkun et al., 2019
• P-gp: Broccatelli et al., 2011
• BBB: Martins et al., 2012
• CYP panel: Veith et al., 2009

Projects, measurements and saved records

Workspace routes share inspection, job status, visualization and export components. Target identity and evidence remain specific to each record. Changing molecular input marks previous results as outdated; visualization controls do not start another calculation.

Notebooks save locally in your browser. Signed-in users can save cloud snapshots, archive completed experiments and review them in the Research Hub. Experimental measurements record the structure, assay conditions and source supplied by the researcher; they are separate from computational predictions and are not independently verified by Asidsal.

Reports preserve the evidence available in the saved snapshot. Save a new snapshot to include later changes. See data handling for storage and retention details.

Preserve the input, processing choices, protocol identifier, software versions, reference evidence and exported checksums. Checksums establish file integrity; they do not establish scientific validity. Legacy URLs may redirect to the catalogue and should not be interpreted as available workflows.

Pose recovery, ranking and availability

Best-sampled RMSD describes the closest returned pose to a reference. Top-ranked RMSD describes the pose placed first by the scoring function. Success on one measure does not imply success on the other.

The reference check uses the recorded best-sampled value when available and the top-ranked value for older records. Each protocol defines its acceptance criterion. Check which metric was used before interpreting a benchmark pass.

A recorded benchmark describes the reference experiment. Current availability also depends on the installed target files and service checks. The workspace shows whether it can accept a new run.

Unavailable entries explain their current status in the catalogue.

Reporting and citation

Suggested platform citation: Asidsal Services Ltd. Asidsal Chem: evidence-aware molecular discovery and computational chemistry workflow platform [software]. Version 2026. https://asidsalservices.com/chem (accessed, with the date you used it). Add the job or run identifier and the protocol hash from your report, and documentation version 2026-10-01. Every report carries this reference in its list, with the numbered citations for the structure and the software used.

Target Discovery results also rest on ChEMBL (EMBL-EBI, CC BY-SA 3.0) and UniProt (CC BY 4.0). Cite ChEMBL and UniProt as listed in the report, and keep the share-alike terms when you reuse the tables.

Report the tool name, access date, software versions, processing choices and protocol identifier from your export. Retain the complete experiment package with your research records. Cite the underlying methods and target structure in addition to identifying Asidsal as the interface.

• RDKit documentation and methods
• AutoDock Vina documentation and citation
• Protein-Ligand Interaction Profiler (PLIP)
• RCSB PDB: 3QGT (PfDHFR · Pyrimethamine)
• RCSB PDB: 1IEP (c-Abl · Imatinib)
• RCSB PDB: 7BQY (SARS-CoV-2 Mpro · Quercetin (non-covalent))
• RCSB PDB: 4EY5 (hAChE · Huperzine A (crystal))
• RCSB PDB: 3ERT (ERα · 4-Hydroxytamoxifen (OHT))
• RCSB PDB: 1A9U (p38α MAPK · SB203580 (crystal))
• RCSB PDB: 1MUI (HIV-1 Protease · Lopinavir (ABT-378))
• RCSB PDB: 6COX (COX-2 · SC-558 (crystal, 26 HA))
• RCSB PDB: 8VK6 (ScCYP51 · Oteseconazole (VT-1161))
• RCSB PDB: 6N2W (5-Lipoxygenase · NDGA / nordihydroguaiaretic acid (PDB 30Z))
• RCSB PDB: 2Y9X (Tyrosinase · Tropolone (CCD 0TR, 9 heavy atoms))
• RCSB PDB: 2YAZ (LmDUTPase · dUMP (CCD UMP, 20 HA))
• RCSB PDB: 3LN1 (COX-2 (celecoxib) · Celecoxib (crystal, 26 HA))
• RCSB PDB: 3OG7 (B-Raf V600E · Vemurafenib (PLX4032, crystal, 33 HA))
• RCSB PDB: 4TZK (InhA · Triclosan-class inhibitor (CCD 641, 23 HA))
• RCSB PDB: 1X70 (DPP-4 · β-amino acid inhibitor (CCD 715, 28 HA))
• RCSB PDB: 1US0 (Aldose reductase · IDD594 (CCD LDT, 24 HA))
• RCSB PDB: 3POZ (EGFR kinase · TAK-285 (CCD 03P, 38 HA))
• RCSB PDB: 2HU4 (Neuraminidase N1 · Oseltamivir carboxylate (CCD G39, 20 HA))
• RCSB PDB: 1S3V (Human DHFR · Trimethoxyphenyl quinazoline antifolate (CCD TQD))
• RCSB PDB: 4BDS (Human BChE · Tacrine)
• RCSB PDB: 6XG5 (E. coli DHFR · Trimethoprim)
• RCSB PDB: 2X9N (T. brucei PTR1 · N2-cyclopropyl triazine-2,4,6-triamine (CCD AX3))
• RCSB PDB: 7JRN (SARS-CoV-2 PLpro · GRL-0617 (CCD TTT))
• RCSB PDB: 4KW5 (M. tuberculosis DprE1 · Benzothiazole carbamate (CCD 1W6))
• RCSB PDB: 2GIR (HCV NS5B · Thiophene carboxylate (CCD NN3))
• RCSB PDB: 2V5Z (Human MAO-B · Safinamide)
• RCSB PDB: 2Z5X (Human MAO-A · Harmine)
• RCSB PDB: 2PRG (PPAR-gamma · Rosiglitazone)
• RCSB PDB: 2AM9 (Androgen receptor · Testosterone)
• RCSB PDB: 2XJX (Hsp90-alpha · Onalespib (CCD XJX))
• RCSB PDB: 5DS3 (PARP1 · Olaparib (CCD 09L))
• RCSB PDB: 1OYT (Thrombin · Fluorobenzyl pyrrolopyrrole inhibitor (CCD FSN))
• RCSB PDB: 2W26 (Factor Xa · Rivaroxaban)
• RCSB PDB: 5AKE (Soluble epoxide hydrolase · Dibenzothiophene sulfone urea (CCD 6N0))
• RCSB PDB: 2QBS (PTP1B · Thiophene dicarboxylate (CCD 024))
• RCSB PDB: 5L2M (ALDH1A1 · Trimethylfuranocoumarin inhibitor (CCD 6ZY))
• RCSB PDB: 2BEL (11beta-HSD1 · Carbenoxolone)
• RCSB PDB: 2OBJ (PIM1 kinase · Hydroxyphenyl pyridinecarbonitrile (CCD VRV))
• RCSB PDB: 3HS4 (Carbonic anhydrase II · Acetazolamide)
• RCSB PDB: 456C (MMP-13 · Pyrimidine dicarboxamide (CCD CBP))
• RCSB PDB: 5EDU (HDAC6 · Trichostatin A)
• RCSB PDB: 1YQY (Anthrax lethal factor · Hydroxamate inhibitor (CCD 915))
• RCSB PDB: 1UDT (PDE5A · Sildenafil)
• RCSB PDB: 4ASD (VEGFR2 · Sorafenib (CCD BAX))
• RCSB PDB: 4JPS (PI3K alpha · Alpelisib (CCD 1LT))
• RCSB PDB: 2XP2 (ALK · Crizotinib (CCD VGH))
• RCSB PDB: 6GUE (CDK2 · AZD5438 (CCD FB8))
• RCSB PDB: 4RMG (Sirtuin 2 · SirReal2 (CCD 3TE))
• RCSB PDB: 5EHR (SHP2 · SHP099 (CCD 5OD))
• RCSB PDB: 4N1T (MTH1 · TH287 (CCD 2GD))
• RCSB PDB: 1ZUA (AKR1B10 · Tolrestat (CCD TOL))
• RCSB PDB: 1C8K (Glycogen phosphorylase · Flavopiridol (CCD CPB))
• RCSB PDB: 1LHO (SHBG · 5-alpha-androstane-3-beta,17-beta-diol (CCD AOM))
• RCSB PDB: 1LI4 (SAH hydrolase · Neplanocin A analogue (CCD NOC))
• RCSB PDB: 1VSO (Kainate receptor GluK1 · (S)-ATPO (CCD AT1))
• RCSB PDB: 2B8T (Thymidine kinase · Thymidine (CCD THM))
• RCSB PDB: 3BQD (Glucocorticoid receptor · Deacylcortivazol (CCD DAY))
• RCSB PDB: 3BUR (Steroid 5-beta-reductase · Testosterone (CCD TES))
• RCSB PDB: 3D0E (AKT2 · GSK690693 (CCD G93))
• RCSB PDB: 3F9M (Glucokinase · Glucokinase activator (CCD MRK))
• RCSB PDB: 3LAN (HIV-1 reverse transcriptase · Pyrimidinedione NNRTI (CCD KBT))
• RCSB PDB: 4G2F (EPHA3 · Compound 7 (CCD C07))
• RCSB PDB: 4MP7 (PDK2 · Isoindoline benzenesulfonamide (CCD PFT))
• RCSB PDB: 6M8D (IPMK · Diosmetin (CCD J8D))
• RCSB PDB: 9FJX (Cereblon · S-Lenalidomide (CCD LVY))
• RCSB PDB: 1EQG (COX-1 · Ibuprofen (CCD IBP))
• RCSB PDB: 6UWP (BACE1 · Aminothiazine inhibitor (CCD QKA))
• RCSB PDB: 1DB4 (sPLA2-IIA · Indole phosphonate (CCD 8IN))
• RCSB PDB: 1DF7 (MtDHFR · Methotrexate (CCD MTX))
• RCSB PDB: 3L4Y (MGAM · Sulfonium pseudo-sugar (CCD NR4))
• RCSB PDB: 1G2A (PDF · Actinonin (CCD BB2))
• RCSB PDB: 3LQ8 (MET · Foretinib (CCD 88Z))
• RCSB PDB: 4XMB (Keap1 · Naphthalene bis-sulfonamide (CCD 41P))
• RCSB PDB: 3LPP (Sucrase-isomaltase · Sulfonium sulfate, C12H24O12S2 (CCD KTL))
• ADMET-AI: Swanson et al. (2024)
• Chemprop: directed-message-passing neural networks for molecular property prediction
• Therapeutics Data Commons (TDC)