What is ProtSpace?

ProtSpace is a browser-based tool for exploring protein language model (pLM) embeddings. It maps the embedding space, not a sequence-similarity network, so it reveals relationships that sequence similarity misses. Overlay biological annotations to turn high-dimensional protein data into testable hypotheses. Bundles you explore are never uploaded.

Why Use ProtSpace?

Protein language models (like ProtT5, ESM2, Ankh) create embeddings that capture biological information in hundreds or thousands of dimensions. ProtSpace helps you:

• See patterns: Visualize how proteins cluster based on their embeddings
• Explore relationships: Find proteins with similar properties
• Discover insights: Identify functional groupings and evolutionary clusters
• Share findings: Export figures or the dataset itself for others to explore

Key Features

Feature	Description
No Installation	Runs entirely in your browser at protspace.app
Privacy-First	Exploring a .parquetbundle is fully client-side, the file stays local
Multiple Views	Compare several dimensionality-reduction views of the same embedding space
Rich Annotations	Color by UniProt, InterPro, Taxonomy, or custom expert annotations
3D Structures	View protein structures from the AlphaFold Database
Export Options	Save images (PNG, PDF), data (Parquet), and protein IDs

How It Works

1. Prepare data: Generate a .parquetbundle file using our Google Colab notebook or Python CLI
2. Load file: Drag & drop onto the Explore page
3. Explore: Navigate, filter, and discover patterns in your protein data

Privacy and Security

Exploring a .parquetbundle happens entirely in your browser:

• No uploads: Your bundle is parsed and rendered locally and never leaves your computer
• One exception: Dropping a .fasta sends those sequences to the ProtSpace prep backend, which computes the embeddings a browser cannot, see Is my data uploaded to a server?
• No tracking: We don't collect any usage data
• Open source: Fully transparent codebase on GitHub

Use Cases

• Functional Analysis: Group proteins by predicted function
• Evolutionary Studies: Identify convergent evolution patterns
• Quality Control: Check embedding model outputs for biases
• Education: Teach protein bioinformatics interactively
• Publication: Create figures for papers and presentations

Performance

ProtSpace can handle datasets with 570,000+ proteins (full Swiss-Prot scale) directly in your browser.

Next Steps

• Quick Start - Get started in 5 minutes
• Using Google Colab - Prepare your data
• Using the Explore Page - Learn all the features