Research Protocol for Compound databases

• Begin by defining the discovery aim, such as target-based screening, ligand expansion, drug repurposing, natural-product mining, toxicity avoidance, or purchasable hit selection, because the aim determines whether PubChem, ChEMBL, BindingDB, DrugBank, ZINC, NPASS, COCONUT, or a specialized database is appropriate^{[1][2][3][4][5][7]}.

• Next, retrieve compounds with full provenance, including database name, version or access date, identifiers, structure strings, target records, assay descriptions, activity values, units, species, cell line, and literature source when available^{[1][2][3][6][10]}.

• Then, standardize chemical structures and annotations by removing duplicates, resolving identifier conflicts, checking stereochemistry, harmonizing activity units, excluding unsupported predictions from “measured activity” fields, and documenting all exclusion rules^[6][9][10].

• After curation, apply computational prioritization using evidence-weighted criteria such as known target activity, structural similarity to active ligands, docking compatibility, ADME properties, chemical diversity, novelty, and purchasability^{[3][4][5][11]}.

• Finally, validate prioritized compounds experimentally using dose-response testing, orthogonal assay formats, positive and negative controls, target-engagement readouts, and cytotoxicity counter-screens before claiming biological activity or mechanism^[2][3][11].

Objective 1: Select fit-for-purpose compound databases.

• Research approach: match the research question to database type.
• Experimental model: computational compound-discovery workflow.
• Experimental groups: general chemical databases, bioactivity databases, drug-repurposing databases, purchasable-compound databases, natural-product databases, and disease- or tradition-specific databases.
• Key techniques: database comparison, metadata extraction, identifier mapping, and evidence grading.
• Detection indices: compound count, structure availability, target annotation, assay evidence, activity units, purchasability, update status, license/accessibility, and citation support.
• Expected results: a justified database set matched to the project goal.
• Interpretation: use multiple complementary databases when no single source contains structure, bioactivity, target, and availability information^{[1][2][3][4][5][6]}.

Objective 2: Curate and standardize the compound library.

• Research approach: clean chemical records before screening.
• Experimental model: downloaded or queried compound sets.
• Experimental groups: raw database records, deduplicated records, standardized structures, filtered screening library, and excluded records.
• Key techniques: SMILES/InChI standardization, salt removal, duplicate removal, stereochemistry checking, molecular descriptor calculation, activity-unit harmonization, and annotation provenance tracking.
• Detection indices: number of retained structures, duplicate rate, missing identifiers, structural conflicts, assay-type conflicts, and records with measured versus predicted activity.
• Expected results: a traceable curated library suitable for cheminformatics and experimental follow-up.
• Interpretation: screening uncurated records risks false prioritization from duplicated, inconsistent, or poorly annotated compounds^[6][9][10].

Objective 3: Prioritize compounds computationally.

• Research approach: rank curated compounds using target-, ligand-, phenotype-, or availability-driven criteria.
• Experimental model: curated library and selected disease target or phenotype.
• Experimental groups: positive-control ligands, known inactive or decoy compounds when available, candidate compounds, and deprioritized compounds.
• Key techniques: similarity search, molecular docking, pharmacophore screening, ADME filtering, target fishing, network pharmacology, and purchasability filtering.
• Detection indices: similarity score, docking score, target evidence, activity potency, selectivity, physicochemical properties, predicted ADME risk, and commercial availability.
• Expected results: a ranked shortlist of compounds with explicit evidence for each ranking decision.
• Interpretation: computational scores should be treated as prioritization evidence, not proof of biological activity^{[3][4][5][11]}.

Objective 4: Experimentally validate prioritized hits.

• Research approach: test whether database-derived hits reproduce the predicted activity in orthogonal assays.
• Experimental model: biochemical assay, cell-based assay, organoid assay, or animal model appropriate to the target.
• Experimental groups: vehicle control, positive-control compound, negative-control compound, each candidate compound, and dose-response series.
• Key techniques: target-binding assay, enzymatic assay, cell viability assay, reporter assay, Western blot, RT-qPCR, ELISA, flow cytometry, and cytotoxicity counter-screen.
• Detection indices: potency, efficacy, selectivity, cytotoxicity, target engagement, pathway modulation, and reproducibility across assay formats.
• Expected results: a subset of computational hits should show measurable target or phenotype activity.
• Interpretation: only compounds with reproducible activity and target engagement should advance beyond database-based prioritization^[2][3][11].

Objective 1

• Produce a database-selection table that explains why each resource was included, what evidence it contributes, and what limitations it has^[1][2][3][6].

Objective 2

• Produce a curated, deduplicated, traceable compound library with standardized structures and harmonized annotations; high duplicate or missing-data rates would indicate that the database requires additional curation before screening^[6][9][10].

Objective 3

• Produce a ranked candidate list with transparent prioritization criteria; compounds supported by multiple independent data types should be prioritized over compounds supported only by a single computational score^{[3][4][5][11]}.

Objective 4

• Confirm whether computationally prioritized compounds show reproducible activity, selectivity, and target engagement; failure in orthogonal validation indicates that database evidence was insufficient for biological inference^[2][3][11].

1: different databases may assign different identifiers, targets, or annotations to the same compound.

Alternative: harmonize structures with standardized identifiers and retain database provenance for every record^[9][10].

2: natural-product and traditional-medicine databases may contain duplicated structures, inaccessible records, sparse annotations, or uneven quality control.

Alternative: use curated open resources and verify key structures and activities against primary literature before experimental testing^[6][7][8][9].

3: bioactivity data may come from different assay formats, units, species, or cell systems.

Alternative: separate biochemical, cell-based, organismal, and predicted records rather than merging them into a single unqualified activity score^[2][10].

4: virtual-screening hits may not be purchasable or experimentally tractable.

Alternative: include purchasability and supplier availability early by using resources designed for commercially available compounds^[5].

5: computational prediction may produce false positives.

Alternative: require orthogonal experimental validation, including dose-response, target engagement, and counter-screening for cytotoxicity or assay interference^[3][11].