How OmicVision 10x their data workflow for more precise proteomics

Proteomics is the study of all the proteins in cells and tissue, what’s present, where they are, and how they change, to reveal the mechanisms behind health and disease. Single-cell proteomics goes further by measuring proteins in individual cells, capturing differences that bulk measurements miss. OmicVision specializes in single-cell spatial proteomics, mapping proteins in their native tissue context to enable more precise, effective therapies.

Before Deepnote, OmicVision’s data work was slowed by infrastructure overhead and fragmented tooling:

• Security risk & admin overhead
  Manual AWS S3 API keys were issued to every engineer, increasing security exposure and adding hours of routine key rotation and permissions work.

• Setup delays & wasted cycles
  Sifting through an overwhelming array of instance options on AWS meant each new analysis stalled while engineers guessed at the right CPU/GPU mix.

• GPU quotas & stalled timelines
  Requests for additional GPU capacity regularly sat in AWS queues, blocking model training and pushing critical timelines back.

• Productivity loss & rework
  Setting up deep‑learning packages and libraries from scratch consumed hours per project, delaying insight.

Plug‑and‑play setup

OmicVision picked Deepnote because it is a fully hosted notebook platform with native connectors for MongoDB and AWS. Existing Jupyter notebooks opened unchanged, with all required deep-learning libraries pre-installed and AI assistance ready in-cell.

Serverless GPUs

Deepnote’s GPUs are serverless by design. Launch on demand, run the job, and Deepnote’s auto-shutdown scales the machine back to zero the moment it goes idle. No quota requests, no idle burn.

Unified workspace

Deepnote has become OmicVision’s primary hub for data exploration and workflow testing. Engineers can swap between CPU and GPU machines as needed.

Low-code automation

The team now uses scheduler to run recurring jobs and publishes notebooks as internal Streamlit apps, while keeping access behind Deepnote’s workspace permissions.

Compute spend slashed by 91%

Compute efficiency increased with serverless GPUs and auto‑shutdown, turning always‑on GPU time into on‑demand billing. Idle machines power down automatically and restart in seconds when a user reconnects.

Security risk eliminated

Workspace‑level secret management replaced individually stored AWS keys. Credentials now live in an encrypted vault behind SSO, removing credential sprawl and audit headaches.

Project velocity accelerated

Moving from idea to execution now takes minutes instead of hours, so researchers spend their time extracting insights rather than configuring machines. Each developer now kicks off ≈ 1 additional project every week, without waiting for AWS quota approvals.

AI pair-programming enabled

Team meetings have turned into live-coding sessions where new ideas are tested on the spot. Deepnote’s built-in AI assistant acts as a pair programmer, generating and refining code in-cell as discussion unfolds. Instead of parking ideas for later, scientists prompt the model, watch runnable code materialize, and iterate on the spot.

Looking ahead, OmicVision plans to scale their data science capabilities significantly with Deepnote at the core of their strategy. Deepnote's ability to simplify infrastructure complexity will enable faster onboarding of new team members and scale their analytical capacity without proportional increases in DevOps overhead.

OmicVision is particularly excited about expanding their portfolio of on-demand data applications. The hosting capabilities and cost-effective compute management provided by Deepnote make it practical to deploy specialized analytical tools that previously would have been prohibitively expensive to run continuously.