targetrecon 0.1.3

Drug target intelligence aggregator — fetch, collate, and visualize public data for any protein target in one command.

TargetRecon

Drug target intelligence aggregator — fetch, collate, and visualize public data for any protein target in one command.

Aggregate UniProt · PDB · AlphaFold · ChEMBL · BindingDB into a single interactive report — in seconds.

---

What is TargetRecon?

TargetRecon is a Python CLI and web app that pulls data from 5 public databases and compiles it into a single, richly formatted report for any protein drug target. No API keys. No account. No manual copy-pasting.

Think of it as gget for drug discovery — or TargetDB reimagined for the AlphaFold era.

---

Five Data Sources, One Report

Source	Data
UniProt	Function, subcellular location, GO terms, diseases, keywords
RCSB PDB	Experimental structures filtered by resolution
AlphaFold DB	Predicted structure with pLDDT confidence coloring
ChEMBL	Bioactivity data (IC50, Ki, Kd, EC50) sorted by pChEMBL descending
BindingDB	Binding affinity measurements converted to pChEMBL, sorted by potency

Intelligent ID Resolution

Accepts gene names, UniProt accessions, or ChEMBL target IDs:

targetrecon EGFR          # Gene name
targetrecon P00533        # UniProt accession
targetrecon CHEMBL203     # ChEMBL target ID

---

Bioactivity Data

What is pChEMBL?

pChEMBL is a unified potency scale — the negative log₁₀ of the molar affinity. Higher = more potent.

pChEMBL = -log₁₀(affinity_M)

pChEMBL	Affinity	Interpretation
9	1 nM	Very potent
7	100 nM	Potent
6	1 µM	Moderate
< 5	> 10 µM	Weak

ChEMBL natively reports pChEMBL. BindingDB reports raw affinities in nM, which TargetRecon converts using the same formula:

pChEMBL = -log₁₀(affinity_nM × 10⁻⁹)

This makes ChEMBL and BindingDB values directly comparable on the same scale.

Full pipeline — what happens when you run a query

1. Resolve query → UniProt ID + ChEMBL target ID
         │
         ▼
2. Fetch in parallel (async):
   ├── ChEMBL API  → top-N bioactivity records, sorted by pChEMBL desc (server-side)
   ├── BindingDB API → all records for this UniProt ID (one bulk request)
   ├── RCSB PDB    → experimental structures
   ├── AlphaFold   → predicted structure
   └── STRING DB   → protein interactions
         │
         ▼
3. BindingDB: sort by pChEMBL descending (client-side), apply cap
         │
         ▼
4. Merge ChEMBL + BindingDB records into one list
         │
         ▼
5. Apply min_pchembl filter (if set) across the merged list
         │
         ▼
6. Deduplicate by canonical SMILES → Ligand Summary
   (same molecule from both sources → one entry, best pChEMBL kept)
         │
         ▼
7. Sort Ligand Summary by best pChEMBL descending
         │
         ▼
8. Output: TargetReport (bioactivities + ligand_summary + structures + ...)

Fetching strategy per source

ChEMBL — server-side sort + pagination:

• Sends order_by=-pchembl_value to the API
• Only the top-N most potent records are fetched — no wasted API calls
• Records with no pChEMBL value are excluded at the API level

BindingDB — one bulk request + client-side sort:

• Fetches all records for the UniProt ID in a single HTTP call (no pagination)
• Sorts by pChEMBL descending locally, then applies the cap
• Records with no valid affinity value are discarded

Cap behavior

The default is 1000 per DB:

Setting	ChEMBL	BindingDB	Total
Default (1000)	top 1000 most potent	top 1000 most potent	up to 2000
--max-bioactivities 500	top 500 most potent	top 500 most potent	up to 1000
--max-bioactivities all	all records	all records	all available

Because sorting happens before the cap, you always get the most potent compounds — never a random subset.

Interface	No-limit syntax
CLI	--max-bioactivities all
Web UI	Drag the Max bioactivities per DB slider to All
Python API	max_bioactivities=None

Ligand deduplication

After merging ChEMBL and BindingDB, all records are grouped by canonical SMILES (via RDKit). If the same molecule appears in both databases:

• It becomes one entry in the Ligand Summary
• The best pChEMBL across all assays is kept
• sources lists both databases (e.g. ["ChEMBL", "BindingDB"])
• num_assays counts the total assay measurements across both sources

The final Ligand Summary is sorted by best pChEMBL descending — the most potent unique compound is always first.

---

CLI

pip install targetrecon

targetrecon / targetrecon run — Single target

targetrecon EGFR
targetrecon P00533 -f html -f json -f sdf -o ./reports/
targetrecon BRAF --min-pchembl 7.0 --max-resolution 2.5
targetrecon CDK2 --max-bioactivities 5000         # up to 5000 per source
targetrecon CDK2 --max-bioactivities all          # no limit
targetrecon CDK2 --no-bindingdb                   # ChEMBL only
targetrecon CDK2 --no-chembl                      # BindingDB only

Option	Default	Description
-f, --format [json|html|sdf]	html json sdf	Output formats (repeat for multiple)
-o, --output PATH	.	Output directory
--max-resolution FLOAT	4.0	Max PDB resolution in Å
--max-bioactivities INT|all	1000	Max records per DB (ChEMBL + BindingDB separately); all = no limit
--min-pchembl FLOAT	—	Minimum pChEMBL value filter
--top-ligands INT	20	Number of top ligands for SDF export
--use-chembl / --no-chembl	on	Include ChEMBL bioactivity data
--use-bindingdb / --no-bindingdb	on	Include BindingDB bioactivity data
-q, --quiet	off	Suppress progress messages

targetrecon batch — Multiple targets

# Pass targets directly
targetrecon batch EGFR BRAF CDK2 ABL1

# From a file (one target per line, # = comment)
targetrecon batch -i targets.txt

# With filters and format selection
targetrecon batch -i targets.txt -f html -f sdf --min-pchembl 6.0 --skip-errors

# ChEMBL only for all targets
targetrecon batch EGFR BRAF --no-bindingdb

# Unlimited bioactivities
targetrecon batch -i targets.txt --max-bioactivities all

Option	Default	Description
-i, --input PATH	—	Text file, one target per line
-o, --output PATH	./batch_reports	Output directory
-f, --format [json|html|sdf]	html json sdf	Output formats (repeat for multiple)
--max-resolution FLOAT	4.0	Max PDB resolution in Å
--max-bioactivities INT|all	1000	Max records per DB (ChEMBL + BindingDB separately); all = no limit
--min-pchembl FLOAT	—	Minimum pChEMBL value filter
--top-ligands INT	20	Ligands per SDF file
--use-chembl / --no-chembl	on	Include ChEMBL bioactivity data
--use-bindingdb / --no-bindingdb	on	Include BindingDB bioactivity data
--skip-errors	off	Continue if a single target fails
-q, --quiet	off	Suppress progress messages

After completion, a summary table is printed showing structures / bioactivities / ligands per target.

targetrecon serve — Launch web interface

targetrecon serve                  # http://localhost:5000
targetrecon serve --port 8080
targetrecon serve --host 0.0.0.0   # expose on all interfaces

Option	Default	Description
--port INT	5000	Port to listen on
--host TEXT	0.0.0.0	Host to bind
--debug	off	Enable Flask debug mode

---

Web UI

targetrecon serve
# Open http://localhost:5000

• Dark-themed interface with animated molecular backdrop
• Search by gene name, UniProt accession, or ChEMBL target ID
• Molecule sketcher (Ketcher) — draw a structure to find matching targets
• Sidebar controls: max PDB resolution, min pChEMBL, ChEMBL/BindingDB toggles, max bioactivities slider (100–5000, or drag to All for no limit)

Report tabs

Tab	Contents
Overview	UniProt summary, GO terms, diseases, protein stats
3D Viewer	AlphaFold pLDDT coloring + PDB experimental structures (3Dmol.js)
Bioactivity	pChEMBL distribution histogram, method breakdown chart
Ligands	Sortable table ranked by potency — SMILES, ChEMBL links, activity type, source
PDB	All experimental structures with resolution, method, ligand count
Interactions	STRING protein–protein interaction network (Cytoscape.js)

Export from the UI

Every report page has one-click download buttons:

• JSON — full machine-readable report
• HTML — self-contained interactive report (works fully offline)
• SDF — top ligands with 3D conformers, ready for docking

---

AI Agent

An AI chat panel is available on every report page. Click the AI button (bottom-right corner) to open it.

Providers & models

Provider	Models
Anthropic	claude-opus-4-6, claude-sonnet-4-6, claude-haiku-4-5
OpenAI	gpt-4o, gpt-4o-mini
Groq	llama-3.3-70b, mixtral

• Bring your own API key — keys are never stored, forgotten after each browser session
• Context-aware: the agent already knows the target you're looking at
• Tools: search targets, fetch bioactivities, query PDB structures, protein interactions, compare targets
• Streaming responses with stop button
• Resizable and minimizable panel

Example questions

What are the best scaffolds for covalent inhibition?
Which PDB structures are most suitable for docking?
Compare the selectivity profile of this target vs CDK4.
Summarize the druggability of this target.

---

Python API

import targetrecon

# Single target — works in scripts and Jupyter
report = targetrecon.recon("EGFR")
print(report.uniprot.protein_name)      # "Epidermal growth factor receptor"
print(report.num_pdb_structures)         # 50
print(report.num_bioactivities)          # up to 2000 (1000 ChEMBL + 1000 BindingDB)
print(report.best_ligand.best_pchembl)   # e.g. 10.52

# With options
report = targetrecon.recon(
    "BRAF",
    max_bioactivities=5000,   # up to 5000 per source
    min_pchembl=7.0,
    max_pdb_resolution=2.5,
)

# No limit — fetch all available records
report = targetrecon.recon("BRAF", max_bioactivities=None)

# Async (for use with asyncio.run or inside async functions)
import asyncio
report = asyncio.run(targetrecon.recon_async("BRAF"))

# Async with all options
report = asyncio.run(targetrecon.recon_async(
    "CDK2",
    use_chembl=True,
    use_bindingdb=False,      # ChEMBL only
    max_bioactivities=2000,
    # max_bioactivities=None  # no limit
))

Accessing the data

# UniProt info
report.uniprot.protein_name           # e.g. "Epidermal growth factor receptor"
report.uniprot.gene_name              # e.g. "EGFR"
report.uniprot.organism               # e.g. "Homo sapiens"
report.uniprot.function_description   # functional annotation text
report.uniprot.subcellular_locations  # list[str]
report.uniprot.disease_associations   # list[str]
report.uniprot.keywords               # list[str]
report.uniprot.go_terms               # list[GoTerm] — each has .go_id, .term, .category
report.uniprot.sequence_length        # int

# PDB structures
for pdb in report.pdb_structures[:5]:
    print(pdb.pdb_id, pdb.resolution, pdb.method)
    for lig in pdb.ligands:           # list[PDBLigand] — each has .ligand_id, .smiles, .name
        print(lig.ligand_id, lig.name)

# AlphaFold
report.alphafold.pdb_url        # URL to AlphaFold PDB structure
report.alphafold.model_url      # URL to AlphaFold CIF model

# Bioactivity records (sorted by pChEMBL descending)
for b in report.bioactivities[:10]:
    print(b.source, b.activity_type, b.value, b.pchembl_value, b.smiles)

# Ligand summary (deduplicated by canonical SMILES, sorted by best pChEMBL)
for lig in report.ligand_summary[:10]:
    print(lig.name, lig.chembl_id, lig.best_pchembl, lig.best_activity_type, lig.num_assays)
    print(lig.sources)                # e.g. ["ChEMBL", "BindingDB"]

report.best_ligand               # most potent unique ligand overall

Export

from targetrecon.core import save_html, save_json, save_sdf

save_html(report, "EGFR_report.html")
save_json(report, "EGFR_report.json")

# SDF with filters
save_sdf(report, "EGFR_ligands.sdf",
         top_n=50,              # limit to top 50
         min_pchembl=7.0,       # only pChEMBL ≥ 7
         activity_type="IC50")  # only IC50 records

Batch (async, concurrent)

import asyncio, targetrecon

async def run_batch(targets):
    reports = await asyncio.gather(*[
        targetrecon.recon_async(t) for t in targets
    ])
    return reports

reports = asyncio.run(run_batch(["EGFR", "BRAF", "CDK2"]))
for r in reports:
    print(r.uniprot.gene_name, r.num_bioactivities, r.num_unique_ligands)

---

Comparison

Feature	TargetRecon	TargetDB (2020)	gget	Open Targets
AlphaFold integration	✅	❌	✅	✅ (web)
ChEMBL bioactivity	✅	✅	❌	Partial
BindingDB binding constants	✅	❌	❌	❌
Interactive HTML report	✅	❌	❌	Web only
3D structure viewer	✅	❌	❌	Web only
Molecule sketcher → targets	✅	❌	❌	❌
Docking-ready SDF export	✅	❌	❌	❌
AI agent chat	✅	❌	❌	❌
Batch CLI processing	✅	❌	✅	N/A
pip install + single command	✅	Partial	✅	N/A

---

Installation

pip install targetrecon

Quick start:

targetrecon EGFR

Produces EGFR_report.html (interactive, self-contained), EGFR_report.json, and EGFR_top_ligands.sdf — ready for docking.

Development:

git clone https://github.com/nagarh/targetrecon.git
cd targetrecon
pip install -e ".[dev]"

---

Architecture

src/targetrecon/
├── cli.py           # Click CLI — run, batch, serve
├── webapp.py        # Flask web app — UI, report pages, AI agent routes
├── core.py          # Orchestration, aggregation, export (HTML/JSON/SDF)
├── models.py        # Pydantic data models
├── resolver.py      # Gene → UniProt → ChEMBL ID resolution
├── report.py        # Jinja2 HTML report generator (standalone)
├── agent_chat.py    # AI agent — tool definitions, streaming, multi-provider
└── clients/
    ├── uniprot.py   # UniProt REST API
    ├── pdb_client.py# RCSB PDB REST + Search API
    ├── alphafold.py # AlphaFold Database API
    ├── chembl.py    # ChEMBL REST API
    └── bindingdb.py # BindingDB REST API

---

Acknowledgments

Data from: UniProt · RCSB PDB · AlphaFold DB · ChEMBL · BindingDB

Visualization: 3Dmol.js · Chart.js · Cytoscape.js

Sketcher: Ketcher

Inspired by TargetDB and gget.

---

License

MIT License — see LICENSE for details.