BioDockify Pharma AI

🧬 BioDockify Pharma AI

AI Research Assistant for Pharmaceutical Sciences

BioDockify Pharma AI is a pharmaceutical research platform built on the Agent Zero v2.0 core, with 22+ integrated modules covering all 8 pharmacy departments. It features AutoDock Vina molecular docking with MM-GBSA free energy scoring, 56 biostatistics analysis types (SPSS/jamovi-level), 12 advanced Academic Writer skills (EQUATOR reporting guidelines, PRISMA systematic review pipeline, per-section quality scoring, De-AIGC rewrite, claim verification), 7-level drug interaction database (50+ clinically significant pairs), Perplexity-style citation system with hybrid search (BM25 + FAISS + RRF), Obsidian bidirectional sync, a Knowledge Base with retrieval-grounded RAG chat, an Academic Writer with pharma-specific claim verification and ICH compliance checks, a 36,145-journal recommender, 4 AI sub-agents, and bulletproof backup to PC.

---

Fork Attribution

BioDockify Pharma AI is a pharmaceutical research fork of Agent Zero, an open-source agentic framework created and maintained by Jan Tomasek and the Agent Zero team.

BioDockify Pharma AI is NOT standalone software. It is built on Agent Zero, which is free and open-source. All core framework functionality, architecture, and capabilities belong to Agent Zero.

We thank Jan Tomasek and the Agent Zero team for their dedication to open AI software.

---

Features

BioDockify AI Engine — Local LLM Support

Runs fully offline. No cloud. No API spend. No PHI/compound data egress.

BioDockify supports any OpenAI-compatible local LLM through its AI Engine module. Bonsai-8B is one model option (installed separately), but you can use any model via Ollama, LM Studio, or any OpenAI-compatible server.

22 Consolidated Research Modules

Statistics — 56 Analysis Types (SPSS/jamovi-level)

Academic Writer — Pharma-Specific Enhancements

Merged Modules (Accessible via Parent Dashboards)

MM-GBSA Free Energy Scoring

After AutoDock Vina completes, MM-GBSA free energy scoring runs automatically (CPU-only, no GPU required):

Output: Per-pose MM-GBSA energies, Z-scores, and consensus with Vina (0.4*Vina_Z + 0.6*MMGBSA_Z). Displayed as a table in the Docking results.

4 Specialized Sub-Agents

---

Quick Start — One Command

Students: just run docker compose up -d and open http://localhost

That's it. BioDockify starts with all features ready. Connect a local LLM via Ollama/LM Studio for offline AI, or use any cloud provider.

docker compose up -d
# Visit http://localhost
# Local AI downloads automatically on first start
# Subsequent starts are instant

Prerequisites

• Docker Desktop (Windows/macOS) or Docker Engine (Linux) — required
• 8GB+ RAM recommended (12GB+ for large docking jobs)
• Internet — needed to pull the Docker image (~18 GB). Once pulled, BioDockify works offline. Add a local LLM via Ollama/LM Studio for fully offline AI chat.

AI Model Options

You can mix: local Ollama for main (offline) + cloud for utility (optional). Fully user-driven.

Docker Compose (full template)

services:
  biodockify:
    image: tajo9128/biodockify-pharma-ai:latest
    container_name: biodockify
    ports:
      - "80:80"
    volumes:
      # 1) ALL DATA — change host path to YOUR folder
      - F:/biodockify_pharma_usr:/a0/usr
      # 2) BACKUPS — change host path to YOUR backup folder
      - F:/biodockify-backups:/app/data
      # 3) DOCKER SOCKET (for volume listing)
      - /var/run/docker.sock:/var/run/docker.sock
    environment:
      - TZ=Asia/Kolkata
      # Optional cloud API keys (not required — local AI works without them)
      # - OPENAI_API_KEY=sk-your-key-here
      # - ANTHROPIC_API_KEY=sk-ant-your-key-here
      # - OPENROUTER_API_KEY=sk-or-your-key-here
    extra_hosts:
      - "host.docker.internal:host-gateway"
    restart: unless-stopped

Backup to PC

Option A — In-app (recommended): Open Backup & Recovery panel → click "Save to PC" → .zip downloads to your Downloads folder.

Option B — Script (Windows): Double-click backup-data.bat to save all research data (workspace + agent memory + knowledge base) to your Desktop.

Option C — Automatic: Backups run daily at 3 AM + on every container restart. They appear in ~/biodockify-backups/ on your PC.

Your Data = Your Folders on Your PC

All research data lives in folders YOU create on your PC. Delete the container? Your data stays. Recreate it? Point to the same folders.

Step 1 — Create 2 folders on your PC:

F:\biodockify_pharma_usr     ← ALL your research data lives here
F:\biodockify-backups        ← backups go here

Step 2 — Edit docker-compose.yml, set your paths:

⚠️ Windows users: use forward slashes (F:/folder) NOT backslashes. Copy path from Explorer → replace all \ with /. Example: F:\biodockify_pharma_usr → F:/biodockify_pharma_usr

volumes:
  # YOUR data folder → container path (do not change /a0/usr)
  - F:/biodockify_pharma_usr:/a0/usr

  # YOUR backup folder → container path (do not change /app/data)
  - F:/biodockify-backups:/app/data

Step 3 — Start:

docker compose up -d

Backup & Restore:

• Backup to PC: Backup panel → "Save to PC" → saves to your F:/biodockify-backups folder
• Restore from PC: Backup panel → select backup → "Restore"

Upgrading:

docker compose down
docker compose pull
docker compose up -d        # same folders → data comes back

---

Research Workflow

User Request
     ↓
Agent0 (Main Orchestrator)
     ↓
├─→ Researcher ─→ Hacker (if blocked)
│         ↓
│    Biostatistician (SPSS-level stats) + QSAR (predictions)
│         ↓
│    Pharmacophore (feature detection)
│         ↓
├─→ Molecular Toolkit (Docking: Vina → MM-GBSA)
│         ↓
│    Deep Analysis (3D View, Interactions, Clusters)
│         ↓
├─→ Pipeline Engine (25-stage autonomous workflow)
│    Debate → Experiment → Self-Heal → Verify → Quality Gate → HITL
│         ↓
└─→ Writer (output) + Journal Recommender (36,145 journals) + Mol Optimizer

---

What's New in v7.9.8

ADMET — Separate Module with SwissADME-Grade Analysis

HOMO-LUMO Calculator (NEW)

200+ RDKit Descriptors (NEW)

Interactive Molecular Plots (NEW)

8 interactive Plotly dashboards: Property Distributions, Correlation Heatmap, Chemical Space PCA, Drug-likeness Analysis, 3D Chemical Space, Parallel Coordinates, Scatter Matrix, Summary Table.

QSAR — 20+ ML Models

Models: PLS, Ridge, Lasso, ElasticNet, KNN, DecisionTree, RandomForest, GradientBoosting, ExtraTrees, AdaBoost, Bagging, SVR, KernelRidge, MLP, XGBoost, LightGBM, CatBoost, Voting, Stacking.

Pharmacophore — 8 Outputs

2D plot, 3D HTML, distance CSV, distance heatmap, distribution plot, feature CSV, fingerprint summary, properties CSV.

Literature Search — 10 Databases

database: "all" searches all 10 in parallel: europe_pmc, pubmed, semantic_scholar, biorxiv, arxiv, google_scholar, scopus, wos, elsevier, springer.

Infrastructure

• Docker Hub: linux/amd64 only, provenance disabled (fixes pull failures)
• Volume: single named volume biodockify_pharma_usr + backup bind mount
• Backup: 2 buttons (Save to PC / Restore from PC)
• Version: v7.9.8

---

What's New in v7.5.8

Local AI Engine bundled inside BioDockify — one docker compose up does everything

The biggest architectural change: llama-server is now bundled inside the BioDockify Docker image itself. No separate sidecar container, no extra image pull, no profiles, no scripts for students to run.

Technical details

• Dockerfile.release: multi-stage build — extracts llama-server + all ~30 shared libraries from ghcr.io/ggml-org/llama.cpp:server, copies to /opt/llama-server/, registers with ldconfig
• exe/init_bonsai.sh: verifies bundled model at /opt/llama-server/models/ (no download needed)
• exe/init_and_run_llama.sh: supervisord entrypoint — calls init_bonsai.sh then execs llama-server
• docker-compose.yml: single container, 1 volume + 1 backup bind mount (no sidecar, no init container)
• modules/local_llm/: data-driven model catalog + runtime registry + pharma prompt library + hardware detection
• api/local_llm.py: 7 actions (status, hardware, catalog, runtimes, prompts, readiness, benchmark)
• tests/test_local_llm.py: 27 tests covering schema, runtime contract, compose wiring, startup scripts

Version history (v7.5.2 → v7.5.8)

---

What's New in v7.5.1

Complete Research Pipeline: Literature → Knowledge Base → Academic Writer

• Literature Search: 10 databases with automatic full-text retrieval
• Deep Research: Multi-database comprehensive gathering (50-300 papers)
• Agent Tool Prompts: 7 new tool prompts teach the agent how to call research modules
• Academic Writer KB Integration: "Use KB Sources" UI with category dropdown, 100K char budget
• KB Storage fix: 3 modules used a Flask-bound import — replaced with stdlib-only auto_store

Backup & Recovery — Bulletproof

• All 3 data locations captured: /a0/usr, /a0/.a0proj, /a0/data
• Backups visible on PC: Container mounts ~/biodockify-backups:/a0/usr/backups
• Save to PC / Restore from PC buttons: Real download/upload flow
• Daily auto-backup: Cron job at 3 AM + on-startup backup

---

What's New in v7.9.2

Full pharma department module architecture refactor + advanced writing skills

This release upgrades the Academic Writer, expands all 8 pharmacy departments, and refactors the module architecture for testability.

Academic Writer — Advanced Writing Skills (from PaperForge + Rigorous)

• De-AIGC Rewrite — detects AI-typical patterns, calculates AIGC risk score (0-100), suggests human-sounding rewrites for journal submission
• Section Analyzer (S1-S10) — per-section quality scoring: Abstract, Introduction, Methods, Results, Discussion, Conclusion, References
• Citation Gap Finder — detects pharma claims without citations (IC50, safety, regulatory, clinical trial)
• Terminology Checker — catches inconsistent abbreviations, drug names, units
• Scientific Rigor Review (R1-R7) — 7 dimensions: Originality, Impact, Ethics, Data Availability, Statistical Rigor, Technical Accuracy, Consistency
• Quality Control Layer — deduplicates and curates review feedback
• Executive Summary — 2-step synthesis: strengths, weaknesses, action items, verdict (READY FOR SUBMISSION / NEEDS REVISION)

EQUATOR Reporting Guidelines + AI Disclosure + PRISMA Pipeline

• 6 EQUATOR guidelines: CONSORT (clinical trials, 37 items), STROBE (observational, 22 items), ARRIVE (animal research, 23 items), STARD (diagnostic accuracy, 22 items), TRIPOD (prediction models, 23 items), CHEERS (health economics, 24 items)
• AI Usage Disclosure generator — venue-specific statements for ICMJE (NEJM/Lancet/JAMA/BMJ), Nature Portfolio, Science, IEEE, ACL/EMNLP, FDA/EMA regulatory submissions
• PRISMA systematic review pipeline — 34-item checklist, flow diagram builder, RoB 2 (RCTs), ROBINS-I (non-randomized), GRADE certainty

Perplexity-style Citations

• Source type prefixes: [kb:1], [web:2], [page:3], [pubmed:5]
• Hybrid search: BM25 (keyword) + FAISS (vector) + Reciprocal Rank Fusion
• Grounded KB Chat: real RAG endpoint replacing the broken textarea hack
• Table-aware chunking: docking/ADMET/SAR tables kept intact

Obsidian Integration

• Bidirectional sync: BioDockify ↔ Obsidian vault (manual button click)
• Frontmatter: Obsidian-standard YAML (title, tags, source, category, created)
• In-app panel: right-canvas rail → "BioDockify AI Engine" → Status/Models/ Runtimes/Benchmark tabs

Pharma Departments — All 8 Upgraded

• Pharmaceutics: 6→10 actions + DOE (full factorial/fractional/Taguchi)
• Clinical Pharmacy: DDI database 15→50+ pairs, AUC-vancomycin, Beers Criteria 2023, STOPP/START v2
• Pharma Analysis: ICH Q2(R2) validation, USP <621> chromatography
• Natural Products: 6→12 plants, IC50 4PL, dereplication
• Regulatory: Fixed EMA search (was stub)
• Pharmacology: Unchanged (already strong at 7 actions)
• Medicinal Chemistry: Unchanged (already strong at 10 actions)

Literature Search — Fixed

• Fixed PubMed PMCID extraction (was returning empty → Tier-1 full text broken)
• Added Europe PMC PMCID resolver
• Added openAccessPdf to Semantic Scholar fields
• deep_research PubMed: switched esummary → efetch (restores abstracts)
• Database name normalization (case-sensitive fix)

Architecture

• Proper modules/ directories: clinical (7 files), formulation (4), pharma_analysis (3), natural_products (4)
• modules/pharma_utils/ — shared calculations (f2, Cheng-Prusoff, SST, content uniformity, Beers, DOE, Chou-Talalay, SAR)
• modules/obsidian/ — bidirectional sync engine
• modules/rag/ — citations, hybrid search, table chunker
• modules/writing/ — 12 advanced writing skills
• 41 tests passing

Version history (v7.5.2 → v7.9.2)

---

Docker Hub

Image: tajo9128/biodockify-pharma-ai:latest

https://hub.docker.com/r/tajo9128/biodockify-pharma-ai

---

License

BioDockify Pharma AI is open-source under the MIT License, inherited from the Agent Zero framework by Jan Tomasek. See the original repository for framework license details.

Documentation

• User Guide — 28 chapters covering installation, modules, research workflows
• Local AI Engine Guide — architecture, install, pharma prompts, troubleshooting
• Architecture — system design and component overview
• AGENTS.md — developer reference and conventions
• CHANGELOG — release history

Support

• GitHub Issues
• Docker Hub