CRYO

(whoever stumbles on this repo, pls note im still working on it as a single person team, if some features donot work pls raise an issue or PR :))

AI-powered biology research platform designed for autonomous scientific discovery. Synthesize literature, analyze proteins, repurpose drugs, interpret genomic variants, run omics pipelines, simulate metabolic drug responses, and generate publication-ready interactive reports — branch and explore research hypotheses like a visual flowchart.

Built on: Hermes Agent with 28+ biology tools | Genome-scale metabolic digital twin engine (Human-GEM, iJO1366, Yeast8, etc.) | Interactive report engine v4 with charts, diagrams, callouts, timelines | Google Gemini 3 Pro Preview | React Flow workspace | PostgreSQL persistence

Screenshots

Quick Start

git clone <repo-url> cryo && cd cryo
cp .env.example .env           # Set GEMINI_API_KEY, JWT_SECRET (required)
bash cryo.sh                   # Builds Docker image + starts all services (db, api, frontend)
open http://localhost:3000     # Access CRYO in browser

Default superuser: creator@cryo.in / creator@shivam0705

Services:

🖥️ Frontend (React 19 + TypeScript): http://localhost:3000
🔌 API (FastAPI + SSE): http://localhost:8000
🗄️ PostgreSQL 17: localhost:5432
📊 Reports serve from /cryo-data/users/{uid}/conversations/{cid}/reports/

Two Interfaces

Chat View (`/`)

Traditional single-thread chat with sidebar conversation history, slash commands, streaming responses, file upload, and report viewer panel.

Workspace View (`/workspace`)

Multi-canvas research workspace built on React Flow:

┌─────────────────────────────────────────────────────────────────┐
│ CRYO Workspace                                    [+ New Node]  │
│                                                                  │
│ ┌──────────────┐         ┌──────────────┐                       │
│ │ 🧬 EGFR       │────────▶│ 🔬 Osimertinib│                      │
│ │ protein info  │         │ drug info     │                      │
│ │              │         │               │   ┌──────────────┐   │
│ │ [messages]   │         │ [messages]    │──▶│ 📋 Report:    │   │
│ │ [/commands]  │         │ [Branch btn]  │   │ EGFR in NSCLC │   │
│ └──────────────┘         └──────────────┘   └──────────────┘   │
│                                                                  │
│ Pan: drag background · Zoom: scroll · Resize: drag node corner  │
└─────────────────────────────────────────────────────────────────┘

Features:

Multiple research nodes — each is an independent chat with its own conversation
Branching — hover any assistant response → click Branch → spawns connected child node with context
Resizable nodes — drag bottom-right corner
Slash commands in every node (all 30 commands available)
File upload — drag-drop or click to attach data files in every node
Collapsible panels — left (workspace list) and right (node list), draggable width
Multiple workspaces — create, switch, rename, delete (max 10 per user, max 50 nodes per workspace)
Persistent — nodes, positions, edges, conversations all saved to PostgreSQL
Messages reload — refresh page → messages load from conversation history
Visual connections — animated cyan arrows between branched nodes
Minimap — overview of all nodes in corner

Slash Commands

Type / in any chat or workspace node. Commands are grouped by category:

Literature

Command	Example	What It Does
`/pubmed <query>`	`/pubmed CRISPR glioblastoma`	Search PubMed literature
`/biorxiv <query>`	`/biorxiv single-cell RNA-seq`	Search bioRxiv preprints

Protein & Structure

Command	Example	What It Does
`/protein <gene>`	`/protein TP53`	Protein/gene lookup (UniProt)
`/structure <id>`	`/structure EGFR`	Protein 3D structures (PDB)

Drug & Variant

Command	Example	What It Does
`/drug <name>`	`/drug temozolomide`	Drug/compound info (ChEMBL)
`/targets <disease>`	`/targets glioblastoma`	Disease-target associations
`/variant <rsid>`	`/variant rs28934578`	Variant clinical significance
`/vep <pos>`	`/vep 17:7675088:C:T`	Variant effect prediction
`/repurpose <disease>`	`/repurpose Huntington disease`	Drug repurposing candidates

Simulation

Command	Example	What It Does
`/digital_twin <drug> [--cell_line <line>]`	`/digital_twin imatinib --cell_line MCF7`	Metabolic drug response simulation
`/simulate <drug> [--cell_line <line>]`	`/simulate metformin --cell_line HeLa`	Alias for `/digital_twin`

Omics Databases

Command	Example	What It Does
`/ppi <gene>`	`/ppi TP53`	Protein-protein interactions (StringDB)
`/kegg <query>`	`/kegg cell cycle`	KEGG pathway search
`/reactome <genes>`	`/reactome BRCA1,BRCA2,ATM`	Reactome pathway enrichment

Analysis Pipelines (requires file upload)

Command	Example	What It Does
`/deseq <file> vs <control>`	`/deseq counts.csv vs control`	Differential expression (PyDESeq2)
`/scrna <file>`	`/scrna data.h5ad`	scRNA-seq clustering (Scanpy)
`/annotate <file>`	`/annotate scrna_processed.h5ad`	Cell type annotation (CellTypist)
`/atac <file>`	`/atac sample.bam`	ATAC-seq peak calling (MACS3)
`/chip <file> vs <input>`	`/chip chip.bam vs input.bam`	ChIP-seq peak calling (MACS3)
`/meta <fastq>`	`/meta sample_R1.fastq.gz`	Metagenomics (Kraken2 + HUMAnN3)
`/ms <file>`	`/ms proteinGroups.txt`	Mass spectrometry proteomics
`/sec <file>`	`/sec sec_data.csv`	SEC chromatography analysis

Research Workflow

Command	Example	What It Does
`/novelty <topic>`	`/novelty CRISPR base editing sickle cell`	Research novelty/saturation check
`/paper <topic>`	`/paper spatial transcriptomics TNBC`	Full manuscript planning pipeline

Output

Command	Example	What It Does
`/compare <A> <B>`	`/compare BRCA1 BRCA2`	Compare genes/proteins/drugs
`/export <topic>`	`/export TP53 variants`	Export to Excel
`/report <topic>`	`/report glioblastoma drug targets`	Generate interactive HTML report
`/chart <topic>`	`/chart cancer mutation frequency`	Generate visualization

File Upload & Collections

Upload Feature (v2 — 2026-05-15)

Every chat (chat mode and workspace nodes) has a drag-drop file upload button with real-time progress tracking.

Workflow:

Click the 📎 icon or drag-and-drop a file onto the input area
Real-time progress bar shows upload status
Auto-classified: suggested command + server path inserted into input
Send to run the analysis or share with agent

Auto-classification Engine:

File Pattern	Detected As	Suggested Command
`counts.csv`, `expression.csv`	RNA-seq counts	`/deseq`
`.h5ad`, `.h5`, `*.hdf5`	scRNA-seq	`/scrna`
`*.bam`	BAM alignment	`/atac` or `/chip`
`.fastq.gz`, `.fq.gz`, `*.fastq`	FASTQ reads	`/meta`
`proteingroups.txt`	MS proteomics	`/ms`
`sec.csv`	SEC chromatography	`/sec`
`.xlsx`, `.xls`	Spreadsheet	`/export`
`.fa`, `.fasta`	FASTA sequence	`/protein`
`.pdf`, `.png`, `*.jpg`	Images/documents	`/analyze_image_vlm`

Specs:

Accepted formats: .csv, .tsv, .txt, .h5ad, .h5, .hdf5, .bam, .fastq, .fastq.gz, .fq, .fq.gz, .xlsx, .xls, .parquet, .json, .fa, .fasta, .pdf, .png, .jpg, .jpeg
Max file size: 2 GB
Storage: /cryo-data/uploads/{user_id}/
Metadata: PostgreSQL uploads table tracks filename, size, data_type, conversation_id, usage_count

Collections (v1 — 2026-05-15)

Organize research artifacts into topic collections. Reference collections in chat to automatically inject context.

Usage:

/collection create Alzheimer_drug_targets
/collection add <topic> → adds to active collection
/collection list → shows all collections + item counts
/collection show <name> → displays collection contents
@collection:Alzheimer_drug_targets → mentions collection in chat (injects metadata)

Schema:

Collections: user_id, name, created_at, description, metadata
Items: collection_id, content_type (paper, gene, drug, pathway), identifier, extracted_data
Auto-populated from: reports, literature searches, drug lookups, gene annotations

Digital Twin v3 (Multi-Backbone Simulation Engine)

CRYO includes a genome-scale metabolic drug simulation engine supporting multiple organism models with real drug target lookup and cell-line personalization.

Supported Models (Backbones)

Backbone	Organism	Reactions	Genes	Cell Line Support
Human-GEM	Homo sapiens	12,931	2,848	✅ 49 CCLE lines
iJO1366	E. coli K-12	1,366	1,337	❌ Generic only
Yeast8	S. cerevisiae	3,953	1,045	❌ Generic only
Plasmodium	P. falciparum	1,267	705	❌ Generic only

Usage Examples

/digital_twin glucose_inhibitor              # Human-GEM: hardcoded target on MAR09034
/digital_twin metformin --cell_line HeLa     # Human-GEM + CCLE GPR scaling (MCF7, HeLa, etc.)
/simulate 5-fluorouracil --cell_line HCT116  # TYMS target lookup, GDSC IC50 validation
/digital_twin imatinib --cell_line MCF7      # ABL1/KIT/PDGFRB from ChEMBL; ~8 biomass (MCF7-specific)
/digital_twin trimethoprim --model ijo1366   # Pathogen: folA gene mapping, iJO1366 backbone
/digital_twin fluconazole --model yeast8     # Yeast: ERG11 target, S. cerevisiae model
/gem stats --model ijo1366                   # Query genome stats: 1,366 reactions, 1,337 genes

Recent Validations (2026-04-30)

Drug	Backbone	Target	Cell Line	Biomass Change	Notes
`trimethoprim`	iJO1366	folA → DHFR	—	-90% ✓	Pathogen digital twin
`sulfamethoxazole`	iJO1366	folP → DHPS2	—	-90% ✓	Antibacterial
`fluconazole`	yeast8	ERG11 → r_0317	—	-90% ✓	Antifungal
`metformin`	Human-GEM	MT-ND1 (Complex I)	MCF7	-18% ✓	Cell-line specific
`imatinib`	Human-GEM	ABL1/KIT/PDGFRA	MCF7	0% (kinase)	Expected (signaling)

Pipeline: Drug → FBA → Report

Drug target resolution
- Human drugs: ChEMBL REST + DGIdb GraphQL with SQLite cache (7-day TTL)
- Pathogen drugs: Custom pathogen_targets_db for E. coli, S. cerevisiae, P. falciparum
- Fallback: Query as reaction ID directly (e.g., MAR09034 for glucose exchange)
Model selection
- --model ijo1366: E. coli iJO1366 (antibacterial context)
- --model yeast8: S. cerevisiae (antifungal context)
- Default (Human-GEM): Cancer drug response, metabolic disease
Cell line personalization (Human-GEM only)
- CCLE expression data (49 cell lines × 19,215 genes)
- Gene expression → reaction GPR constraints (TPM < 1.0 → upper_bound = 0.001)
- Media adaptation: cancer_warburg (generic) or human1_minimal (CCLE-constrained)
FBA simulation
- Baseline flux balance (growth rate as biomass production)
- Perturbed: 90% inhibition on drug-targeted reactions
- Delta computation: (perturbed_biomass - baseline) / baseline × 100%
GDSC2 validation (Human drugs + cell lines)
- Lookup experimental IC50 (μM) from 235,748 drug-cell pairs
- Display in report with trial phase data
HTML report + PNG plot
- Markdown with :::diagram (pathway), :::callout (findings), tables
- Flux bar chart (top 10 reactions, delta values)

`--cell_line` flag

Without --cell_line — uses generic cancer Warburg model (glucose-limited):

/digital_twin metformin
→ cancer_warburg media (glucose-limited)
→ biomass baseline: 62.43  (generic Human1)

With --cell_line — constrains the model to cell line's actual RNA expression:

/digital_twin metformin --cell_line MCF7
→ loads MCF7 RNA-seq from CCLE (19,215 genes, TPM values)
→ reactions where ALL driving genes have TPM < 1.0 → upper_bound = 0.001
→ 1,212 reactions constrained for MCF7
→ biomass baseline: 8.06  (MCF7-specific)

Supported cell lines (49): MCF7, HeLa, A549, HCT116, PC3, LNCaP, U87MG, HepG2, K562, Jurkat, T98G, MDA-MB-231, SK-BR-3, BT474, ZR75-1, PANC1, and 33 more.

Known scientific limitations

Limitation	Root Cause	Impact
TKIs always show 0% biomass change	ABL1/EGFR/PDGFRA are signaling kinases not in Human1	Drug annotation + GPR still work; no direct flux inhibition
Most drugs show 0% biomass change	FBA degrees of freedom — solver routes around single blocks	Hardcoded targets and Complex I drugs show real effects

Best drugs to test

Drug	Target	Effect
`glucose_inhibitor`	MAR09034 (glucose exchange)	Hardcoded — biomass drops >10%
`atp_synthase_inhibitor`	MAR04137 (ATP synthase)	ATP production blocked
`metformin`	MT-ND1 (Complex I)	Mitochondrial ETC target
`5-fluorouracil`	TYMS (thymidylate synthase)	Nucleotide synthesis
`methotrexate`	DHFR (dihydrofolate reductase)	Folate metabolism

Setup (one-time)

# Downloads GDSC2 (~10MB), creates SQLite drug cache
docker exec cryo-api-1 python /app/scripts/setup_digital_twin.py

# Preprocess CCLE data (download OmicsExpressionTPMLogp1HumanProteinCodingGenes.csv
# from depmap.org/portal/download/ → place in /cryo-data/ccle/)
docker exec cryo-api-1 python /app/scripts/preprocess_ccle.py

Report Engine v4

Reports are interactive HTML pages with:

Plotly.js charts (bar, pie, line, scatter) — via :::chart blocks
Mermaid.js diagrams (pathway flowcharts) — via :::diagram blocks
Callout boxes (info/warning/success/danger) — via :::callout blocks
Progress bars (mutation frequencies, trial enrollment) — via :::progress blocks
Timelines (drug approval history) — via :::timeline blocks
Sortable tables — auto-parsed from markdown pipe tables
Sidebar TOC with scroll-spy
Search bar for in-report text search
Dark/light mode toggle
Print button (clean print layout)
Cover page with CRYO branding and report ID

Reports open in a slide-in side panel within CRYO — no new browser tab.

Architecture

┌──────────────────────────────────────────────────────────────────┐
│                    BROWSER (localhost:3000)                        │
│  React 19 · TypeScript · Tailwind 4 · Vite 6 · React Flow 12    │
│                                                                    │
│  ┌─ Chat View ──────────┐  ┌─ Workspace View ──────────────────┐ │
│  │ Sidebar + ChatPage   │  │ React Flow canvas (pan/zoom)       │ │
│  │ ChatInput + SlashMenu│  │ Resizable ChatNodes (mini chat)    │ │
│  │ FileUploadButton     │  │ Branching (context inheritance)    │ │
│  │ ReportPanel (slide)  │  │ File upload per node               │ │
│  │ MessageBubble (md)   │  │ Workspace persistence (PG)         │ │
│  └──────────────────────┘  └────────────────────────────────────┘ │
│                                                                    │
│  Collections: Topic-based metadata org (papers, genes, drugs)    │
│  File Mentions: @file:name syntax in chat                        │
└───────────────────────────┬──────────────────────────────────────┘
                            │ HTTP + SSE
┌───────────────────────────▼──────────────────────────────────────┐
│                    FastAPI Backend (localhost:8000)                 │
│                                                                    │
│  /api/auth/*              JWT auth (signup, login, me)             │
│  /api/chat/*              Conversations, SSE streaming             │
│  /api/workspace/*         CRUD + node positions/edges save         │
│  /api/collections/*       Topic org (create, add, show)            │
│  /api/uploads             File upload, list, delete (2GB)          │
│  /api/reports/*           Serve HTML/Excel/PNG reports             │
│  /api/digital-twin/*      Metabolic simulation REST endpoints      │
│  /api/gem/*               Genome-scale model queries               │
│  /api/health              Health + dependency check                │
│                                                                    │
│  HermesBridge             Slash command dispatch, agent wrapper     │
│  Report Engine v4         Markdown + :::blocks → interactive HTML   │
│  Digital Twin Service v3  Multi-backbone FBA + GPR scaling         │
│  VLM Client               Gemini Vision 2 integration (images)     │
└──────────┬───────────────────────────┬───────────────────────────┘
           │                           │
┌──────────▼──────────┐  ┌────────────▼────────────────────────────┐
│   PostgreSQL 17     │  │   Hermes Agent (per-request)             │
│                     │  │   google-gemini-3-pro-preview · 32K       │
│  users, api_keys    │  │                                          │
│  conversations      │  │  28+ CRYO Tools:                         │
│  messages           │  │   pubmed_search · biorxiv_search         │
│  workspaces         │  │   fetch_citation · uniprot_lookup        │
│  workspace_nodes    │  │   pdb_search · chembl_search · targets   │
│  workspace_edges    │  │   opentargets_search · clinvar_lookup    │
│  uploads            │  │   ensembl_vep · stringdb_ppi             │
│  collections        │  │   kegg_pathway · reactome_enrichment     │
│  papers, genes      │  │   differential_expression (PyDESeq2)     │
│  proteins, drugs    │  │   scrna_analysis · cell_annotation       │
│  variants           │  │   atac_seq · chip_seq (MACS3)            │
│  knowledge_edges    │  │   metagenomics · proteomics_ms · sec_*   │
│                     │  │   novelty_check · manuscript_pipeline    │
│                     │  │   compile_report · generate_excel/chart  │
│                     │  │   verify_claim · analyze_image_vlm        │
│                     │  │   deep_research · digital_twin · /gem     │
└─────────────────────┘  └──────────────────────────────────────────┘
                          
┌─────────────────────────────────────────────────────────────────┐
│  Data & Services (bind-mounted /cryo-data/)                      │
├─────────────────────────────────────────────────────────────────┤
│  users/{uid}/                                                   │
│    uploads/                      ← Uploaded files (2GB max)     │
│    conversations/{cid}/          ← Per-conversation artifacts   │
│      reports/*.html              ← Interactive reports (v4)     │
│      sources/*.json              ← Markdown (editable)          │
│  models/                                                        │
│    human1/human1.xml             ← Human-GEM (12.9k reactions) │
│    ijo1366/ijo1366.json          ← E. coli (1.3k reactions)    │
│    yeast8/yeast8.json            ← S. cerevisiae (3.9k reactions) │
│  ccle/                                                          │
│    ccle_expression_human1.parquet ← 49 cell lines × 19,215 genes │
│  gdsc/                                                          │
│    gdsc2_sensitivity.csv         ← 235,748 drug-cell IC50s     │
│  cache/                                                         │
│    drug_targets.db               ← SQLite (ChEMBL/DGIdb, 7d TTL) │
└─────────────────────────────────────────────────────────────────┘

Optional Services:
┌─────────────────────────────────────────────────────────────────┐
│  VLM OCR Server (localhost:8001) — Image analysis microservice   │
│  Gemini Vision 2 · Tesseract OCR · JSON output                  │
└─────────────────────────────────────────────────────────────────┘

Tools Reference

Core Biology (18 tools)

Tool	Source	Input	Output
`pubmed_search`	NCBI E-utilities	Query string	PMIDs, titles, abstracts, citation count
`biorxiv_search`	bioRxiv API	Query string	Preprints, authors, dates
`fetch_citation`	CrossRef + PubMed	PMID or DOI	APA/MLA/Chicago formatted
`uniprot_lookup`	UniProt REST	Gene name (e.g., TP53)	Protein info: domains, GO, orthologs
`pdb_search`	RCSB PDB	Protein name or ID	3D structures, resolution, ligands
`chembl_search`	ChEMBL REST	Drug/compound name	SMILES, properties, targets, IC50
`opentargets_search`	OpenTargets GraphQL	Disease or gene	Disease-target association scores
`clinvar_lookup`	ClinVar/NCBI	Variant (rsid or HGVS)	Clinical significance, pathogenicity
`ensembl_vep`	Ensembl REST	Genomic position	SIFT/PolyPhen predictions, impact
`compile_report`	Report Engine v4	Markdown content	Interactive HTML with charts/diagrams
`get_last_report`	Disk/DB	Conversation ID	Raw markdown for editing
`generate_excel`	openpyxl	Data + sheet names	Multi-sheet `.xlsx` spreadsheet
`generate_chart`	matplotlib/Plotly	Data + chart type	Standalone PNG or interactive HTML
`verify_claim`	Multi-source	Claim text	Verification status + sources
`analyze_image_vlm`	Gemini Vision	Image file	Image analysis (gels, microscopy, etc.)
`deep_research`	gpt-researcher	Topic	Deep multi-source research report
`multi_agent_research`	open_deep_research	Topic	Multi-perspective research synthesis
`scientific_skill`	133 skill packs	Biopython, DeepChem, ESM, MedChem	Code templates + step-by-step execution

Omics Databases (3 tools)

Tool	Source	What It Does
`stringdb_ppi`	STRING v12 REST	PPI network + functional enrichment for gene list
`kegg_pathway`	KEGG REST API	Pathway search, details, gene members
`reactome_enrichment`	Reactome AnalysisService	Pathway enrichment for gene set, FDR-filtered

Analysis Skills (10 tools)

These tools return a code template + step-by-step instructions executed by Hermes agent. They require uploaded data files.

Tool	Stack	Input	Pipeline
`differential_expression`	PyDESeq2	Count matrix CSV	DESeq2 → volcano plot → DEG table
`scrna_analysis`	Scanpy	H5AD file	QC → norm → UMAP → Leiden → markers
`cell_annotation`	CellTypist	H5AD (normalized)	Cell type labeling + confidence scores
`atac_seq`	MACS3	BAM + metadata	Peak calling (--shift -75) → BED + FRiP
`chip_seq`	MACS3	IP + input BAM	Narrow/broad peaks, summit annotation
`metagenomics`	Kraken2 + HUMAnN3	FASTQ reads	FastQC → Kraken2 → Bracken → profiling
`proteomics_ms`	MaxQuant output	proteinGroups.txt	LFQ norm → PCA → volcano → pathways
`sec_report`	scipy	SEC trace CSV	Savitzky-Golay → peaks → oligomeric state
`novelty_check`	PubMed API	Research topic	Saturation score (1–10) + recent papers
`manuscript_pipeline`	Structured workflow	Topic + context	8-stage planning: abstract → figs → submission

VLM & Image Analysis (New — 2026-05)

Service	Model	What It Does
`analyze_image_vlm`	Gemini Vision 2	Upload gel photos, microscopy images → extract data
`ocr_pipeline`	Gemini Vision 2 + Tesseract	Standalone VLM + OCR microservice (`:8001`)

VLM OCR Server (optional container):

docker run -p 8001:8001 cryo-vlm:latest
# POST /process_image + image file → structured JSON

Genome-Scale Model API (GEM Graph — v1, 2026-04)

Query any supported backbone (Human-GEM, iJO1366, Yeast8, Plasmodium).

Endpoint	What It Returns
`GET /api/gem/backbones`	List all 4 models + load status
`GET /api/gem/stats?backbone=ijo1366`	Reaction/gene/metabolite counts
`GET /api/gem/gene/{gene_id}?backbone=ijo1366`	Gene info + associated reactions
`GET /api/gem/reaction/{rxn_id}?backbone=ijo1366`	Reaction details (substrate, product, GPR)

Slash command:

/gem stats --model ijo1366    # CLI query of above endpoints

Environment Variables

Variable	Description	Default
`GEMINI_API_KEY`	Google AI Studio API key	required
`JWT_SECRET`	JWT signing secret	required
`HERMES_MODEL`	LLM model	`gemini-3-pro-preview`
`HERMES_MAX_ITERATIONS`	Max tool loops	`15`
`NCBI_API_KEY`	NCBI E-utilities key (free, 10 req/s)	optional
`NCBI_EMAIL`	Required when using NCBI key	optional
`CRYO_DATA_DIR`	Persistent data dir	`/cryo-data`
`CRYO_UPLOAD_DIR`	File upload directory	`{CRYO_DATA_DIR}/uploads`
`CRYO_REPORTS_DIR`	Reports directory	`{CRYO_DATA_DIR}/users`
`CRYO_CACHE_DIR`	SQLite cache directory	`{CRYO_DATA_DIR}/cache`
`CRYO_MAX_WORKSPACES_PER_USER`	Max workspaces	`10`
`CRYO_MAX_NODES_PER_WORKSPACE`	Max nodes	`50`
`POSTGRES_*`	Database config	`cryo:5432/cryo`
`MACS3_PATH`	MACS3 binary	`macs3`
`KRAKEN2_PATH`	Kraken2 binary	`kraken2`
`KRAKEN2_DB`	Kraken2 database path	`/cryo-data/kraken2_db`
`HUMANN3_PATH`	HUMAnN3 binary	`humann`
`FASTQC_PATH`	FastQC binary	`fastqc`
`DRUG_LOOKUP_CACHE_TTL_DAYS`	Drug target cache TTL	`7`
`PUBMED_MAX_RESULTS`	Max PubMed results	`20`
`STRINGDB_DEFAULT_CONFIDENCE`	StringDB confidence threshold	`0.4`
`REACTOME_FDR_THRESHOLD`	Reactome enrichment FDR	`0.05`

Data

cryo-data/                              ← bind-mounted from host
  ├── uploads/{user_id}/                ← uploaded data files (all types)
  ├── users/{user_id}/
  │   └── conversations/{conv_id}/
  │       ├── reports/*.html            ← generated interactive reports
  │       └── sources/*.json            ← raw markdown for editing
  ├── models/human1/human1.xml          ← Human-GEM (cached singleton)
  ├── ccle/
  │   └── ccle_expression_human1.parquet  ← 49 cell lines × 19,215 genes (17MB)
  ├── gdsc/
  │   └── gdsc2_sensitivity.csv         ← 235,748 drug-cell IC50 pairs
  └── cache/
      └── drug_targets.db              ← SQLite ChEMBL/DGIdb cache (7-day TTL)

Project Structure

cryo/
├── api/                             # FastAPI backend
│   ├── routers/
│   │   ├── auth.py                  # JWT auth
│   │   ├── chat.py                  # SSE chat + conversations
│   │   ├── workspace.py             # Workspace CRUD + save
│   │   ├── uploads.py               # File upload, list, delete
│   │   └── digital_twin.py          # Simulation REST endpoints
│   ├── models/
│   │   ├── user.py                  # User model
│   │   ├── conversation.py          # Conversation + Message models
│   │   ├── workspace.py             # Workspace + Node + Edge models
│   │   └── upload.py                # Upload tracking model
│   └── services/
│       ├── hermes_bridge.py         # Agent wrapper + slash dispatch (30 commands)
│       ├── report_engine.py         # v4 HTML report engine
│       └── digital_twin/
│           ├── drug_lookup.py       # ChEMBL + DGIdb + SQLite cache
│           ├── ccle_loader.py       # CCLE parquet + GPR scaling
│           ├── gdsc_validator.py    # GDSC2 IC50 lookup
│           ├── personalizer.py      # Media + GPR pipeline
│           ├── media_registry.py    # Cancer Warburg / minimal media
│           ├── perturbation.py      # Drug inhibition logic
│           ├── simulator.py         # FBA baseline + perturbed
│           ├── reporting.py         # Digital twin HTML report + citations
│           ├── service.py           # Orchestration entry point
│           └── data/
│               └── media_registry.json
├── frontend/src/
│   ├── components/
│   │   ├── ChatInput.tsx            # Slash command input + file upload
│   │   ├── ChatNode.tsx             # Workspace node (mini chat + file upload)
│   │   ├── ChatMessage.tsx          # Markdown + bionic reading
│   │   ├── SlashMenu.tsx            # Command dropdown (30 commands, grouped)
│   │   ├── FileUploadButton.tsx     # Shared upload (chat + workspace)
│   │   ├── ReportPanel.tsx          # Slide-in report viewer
│   │   └── Sidebar.tsx              # Chat view sidebar
│   └── pages/
│       ├── ChatPage.tsx             # Traditional chat
│       └── WorkspacePage.tsx        # Multi-canvas workspace
├── hermes-agent/tools/
│   ├── cryo_digital_twin.py         # Digital twin tool schema
│   ├── cryo_literature.py           # PubMed/bioRxiv tools
│   ├── cryo_protein.py              # UniProt/PDB tools
│   ├── cryo_drug.py                 # ChEMBL/OpenTargets tools
│   ├── cryo_variant.py              # ClinVar/VEP tools
│   ├── cryo_reports.py              # Report/chart/export tools
│   ├── cryo_omics_databases.py      # StringDB/KEGG/Reactome tools
│   └── cryo_analysis_skills.py      # DESeq2/Scanpy/MACS3/etc skill tools
├── scripts/
│   ├── setup_digital_twin.py        # One-time data setup (GDSC + cache)
│   ├── preprocess_ccle.py           # CCLE CSV → parquet (49 cell lines)
│   └── verify_human1_exchanges.py   # Inspect Human1 reaction IDs
├── tests/
│   ├── test_digital_twin.py         # 8 integration tests
│   └── test_api.py                  # API endpoint tests
├── cryo-data/                       # Persistent data (bind-mounted)
├── db/schema.sql                    # PostgreSQL schema
├── SOUL.md                          # Agent persona + :::block examples
├── cryo.sh                          # One-command start script
└── docker-compose.yml               # 3 services: db, api, frontend

Docker

bash cryo.sh                      # Rebuild image + start all + tail logs
docker compose up -d              # Start without rebuild
docker compose logs api -f        # Watch tool calls
docker compose restart api        # Reload after .env changes

Tests

docker exec cryo-api-1 python -m pytest tests/test_digital_twin.py -v
# 8/8 passing (~60s)

docker exec cryo-api-1 python -m pytest tests/ -v

License

MIT

Name		Name	Last commit message	Last commit date
Latest commit History 28 Commits
api		api
db		db
docker		docker
frontend		frontend
hermes-agent		hermes-agent
integrations		integrations
media		media
scripts		scripts
tests		tests
vlm_ocr		vlm_ocr
.dockerignore		.dockerignore
.env.example		.env.example
.gitignore		.gitignore
README.md		README.md
SOUL.md		SOUL.md
cryo.sh		cryo.sh
docker-compose.yml		docker-compose.yml
scope.md		scope.md
test_queries.md		test_queries.md

Folders and files

Latest commit

History

Repository files navigation

CRYO

Screenshots

Quick Start

Two Interfaces

Chat View (/)

Workspace View (/workspace)

Slash Commands

Literature

Protein & Structure

Drug & Variant

Simulation

Omics Databases

Analysis Pipelines (requires file upload)

Research Workflow

Output

File Upload & Collections

Upload Feature (v2 — 2026-05-15)

Collections (v1 — 2026-05-15)

Digital Twin v3 (Multi-Backbone Simulation Engine)

Supported Models (Backbones)

Usage Examples

Recent Validations (2026-04-30)

Pipeline: Drug → FBA → Report

--cell_line flag

Known scientific limitations

Best drugs to test

Setup (one-time)

Report Engine v4

Architecture

Tools Reference

Core Biology (18 tools)

Omics Databases (3 tools)

Analysis Skills (10 tools)

VLM & Image Analysis (New — 2026-05)

Genome-Scale Model API (GEM Graph — v1, 2026-04)

Environment Variables

Data

Project Structure

Docker

Tests

License

About

Resources

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Chat View (`/`)

Workspace View (`/workspace`)

`--cell_line` flag

Packages