domains / research / drugbank-xml-analysis / drugbank-database

drugbank-database

Skill card

combined

The two copies cover the same DrugBank XML analysis task but with materially different scope and structure. The aipoch copy is a single, self-contained SKILL.md with a concrete end-to-end Python example (lxml XPath extraction, pandas DataFrames, networkx DDI graph) and explicit dependencies. The FreedomIntelligence copy is a broader, modular skill that references separate files (data-access.md, drug-queries.md, interactions.md, targets-pathways.md, chemical-analysis.md) and adds capabilities the aipoch copy lacks, such as ADMET prediction, chemical similarity/fingerprints, CYP450/transporter analysis, and community detection. Neither is a strict reproduction of the other; crowning one would hide the other's unique steps and constraints.

Repositories

Other copies

Combined skill

---
name: drugbank-database
description: Programmatic access to DrugBank drug and target data; use when you need to download, parse, and analyze DrugBank XML for properties, interactions, pathways, pharmacology, chemical similarity, and ADMET.
license: MIT
---

## When to Use
- Extract structured drug properties (identifiers, synonyms, ATC codes, SMILES, InChI) from DrugBank XML.
- Build and analyze drug-drug interaction (DDI) networks, including severity classification, community detection, and risk scoring.
- Map drugs to targets (proteins/genes), enzymes (CYP450, Phase II), and transporters for mechanism-of-action, repurposing, or off-target analysis.
- Connect drugs to biological pathways (SMPDB) and pharmacology annotations.
- Perform structure-based analysis: molecular similarity, fingerprints, substructure searches, ADMET prediction, and chemical space exploration.
- Generate tabular datasets (CSV/Parquet) for notebooks, dashboards, or ML pipelines.

## Dependencies
- drugbank-downloader (version varies by environment)
- lxml>=4.9
- pandas>=2.0
- networkx>=3.0
- rdkit>=2022.09 (required for structure/chemistry workflows)

## Data Access and Authentication
1. Obtain a DrugBank account (free academic or paid license) and store credentials securely via environment variables or a config file.
2. Use drugbank-downloader to fetch the desired release; comply with DrugBank terms.
3. Place the XML file at a known path (e.g., ./drugbank.xml) or update the path in your code.
4. For repeated work, cache the parsed data to avoid re-parsing the large XML document.

## XML Parsing Approach
- DrugBank is distributed as a large XML document; use lxml.etree for robust XPath-based extraction.
- The XML uses the namespace http://www.drugbank.ca; always include the namespace mapping in XPath queries (e.g., NS = {"db": "http://www.drugbank.ca"}).
- Core extraction patterns:
  - Primary DrugBank ID: db:drugbank-id[@primary='true']
  - Drug name: db:name
  - Calculated properties (e.g., SMILES): db:calculated-properties/db:property[db:kind='SMILES']/db:value
  - Drug interactions: db:drug-interactions/db:drug-interaction with fields db:drugbank-id, db:name, db:description

## Core Extraction and Data Modeling
1. Parse the XML with lxml.etree.parse and get the root element.
2. Iterate over //db:drug elements to extract drug records (drugbank_id, name, type, smiles, and any additional fields such as CAS, InChI, molecular formula, pharmacology, and external identifiers).
3. Build a pandas DataFrame for the drug table; drop rows missing a primary drugbank_id.
4. Extract drug-drug interactions by iterating over each drug's db:drug-interactions/db:drug-interaction children; record source_id, source_name, target_id, target_name, and description.
5. Build a DDI graph with networkx.from_pandas_edgelist (nodes are drugs, edges are interactions, description as edge attribute).
6. For drug-target analysis, extract target records (protein/UniProt ID, action such as inhibitor/agonist/antagonist, enzyme/transporter type) and build a bipartite drug-target graph.
7. For pathway analysis, map drugs to SMPDB pathway identifiers and build a drug-pathway association table.

## DDI Network Analysis
- Compute degree centrality to identify drugs with the most interactions.
- Classify interactions by severity and mechanism where description text allows.
- Perform community detection on the DDI graph to identify interaction clusters.
- Calculate interaction risk scores for polypharmacy regimens by aggregating severity and mechanism attributes.
- Create interaction matrices for specific drug pairs or cohorts.

## Chemical Properties and Similarity
- Extract chemical structures (SMILES, InChI, molecular formula) from calculated-properties or experimental-properties sections.
- Use rdkit to calculate physicochemical properties (MW, logP, PSA, H-bond donors/acceptors).
- Apply Lipinski's Rule of Five and Veber's rules for drug-likeness assessment.
- Generate molecular fingerprints (Morgan, MACCS, topological) and compute Tanimoto similarity between molecules.
- Perform substructure searches with SMARTS patterns.
- Find structurally similar drugs for repurposing and create similarity matrices for clustering.
- Predict oral absorption and BBB permeability using rdkit descriptors or external ADMET models.
- Analyze chemical space with PCA and clustering.

## Target and Pathway Analysis
- Extract drug targets with actions (inhibitor, agonist, antagonist, substrate, etc.).
- Identify metabolic enzymes (CYP450 isoforms, Phase II enzymes) and transporters (uptake, efflux) for ADME studies.
- Map drugs to biological pathways (SMPDB) and extract Gene Ontology (GO) terms for targets.
- Cross-reference with UniProt for protein sequence and function data.
- Find drugs targeting specific proteins and identify drugs with shared targets for repurposing.
- Analyze polypharmacology and off-target effects.

## Typical Workflows
### Drug Discovery Workflow
1. Download and access the latest DrugBank data using the data-access steps above.
2. Build a searchable drug database by extracting core and extended drug fields into a pandas DataFrame.
3. Use chemical-analysis steps to find similar compounds and assess drug-likeness.
4. Use targets-pathways steps to identify shared targets and map to pathways.
5. Use DDI analysis steps to assess interaction risk for candidate combinations.

### Pharmacovigilance Workflow
1. Extract all interactions for a cohort of drugs.
2. Build a bidirectional interaction network and classify by severity.
3. Perform community detection and calculate risk scores.
4. Export interaction matrices and network visualizations for clinical review.

## Performance Notes
- DrugBank XML is large; parse once and cache intermediate DataFrames or pickled objects.
- Use lxml's C-accelerated parser for speed.
- For very large DDI graphs, consider sampling or subgraph analysis to keep networkx operations tractable.
- Use rdkit in batch mode when calculating properties for many molecules.

## Conflicts
- aipoch/medical-research-skills presents a single inline Python example and lists rdkit as optional; FreedomIntelligence/OpenClaw-Medical-Skills defers to separate reference files and treats rdkit-based chemical analysis as a core capability. This combined instruction merges both: it includes the inline example patterns from aipoch and the modular reference structure and additional capabilities from FreedomIntelligence.
- FreedomIntelligence/OpenClaw-Medical-Skills includes community detection and interaction risk scoring in DDI analysis; aipoch/medical-research-skills only shows degree-based ranking. This combined instruction includes both degree-based ranking and community detection/risk scoring.