Part AI startup and part research lab, VantAI combines the power of ML and pharmaceutical data with their expertise in biology, in order to computationally design targeted protein degraders, a breakthrough novel class of therapeutics to tackle previously undruggable disease targets such as in cancer.
Their work presents a lot of hard scientific problems, many without a known solution. VantAI's R&D needs to iterate quickly, and collaboration and environment setup are often a source of friction in other interdisciplinary teams.
Data exploration, model prototyping, documentation and sharing of experiment results
PyTorch, Huggingface, DGL, DeepChem, PyTorch Geometric, RDKit, Kubernetes
Researchers prepare prototypes and experiments as Deepnote projects. Projects contain a prebuilt environment with VantAI’s code and dependencies, ready to be used with zero setup. Data, hosted in GCS buckets or in BigQuery, are seamlessly connected via Deepnote integrations to the project, instead of having to be downloaded to each person’s computer, reducing access time for everyone.
Deepnote enables VantAI to distribute their in-house algorithms and connect their datasets without spending time on environment setup. Each algorithm or dataset comes with a link to a Deepnote project based on a VantAI’s private Docker image, containing all dependencies and tools for data exploration. It means the environment is set up or updated only once, instead of per each user. Projects also serve as interactive documentation with examples, improving accessibility so even non-technical stakeholders can explore the data and bring ideas to the table.
Deepnote projects are fully collaborative, so team members from diverse disciplines can edit code or add comments, in real time. Projects have a shared program state - variables and program objects are visible in the same way to everyone, so referencing or explicitly printing the state of a variable is instantaneous. Collaboration now happens directly in code like a live code review, without a need to infer the program state in your head. This level of interactivity makes brainstorming sessions end even ordinary meetings incredibly productive.
... and clever input widgets bridge the gap between a no-code GUI app for non-coders and a full fledged programming environment for our engineers. More and more, the VantAI team is able to move discussions around abstract ideas and complex data out of presentation tools (eg. PowerPoint) and into direct implementations in code.
We're working in uncharted territory and our work is highly R&D focused. Machine learning is a very empirical discipline so iteration speed is everything - working in Deepnote is like code-review and rapid prototyping at the same time, saving valuable time in the iteration cycles. But as opposed to code review via Github, you have direct access to the runtime and program state which makes understanding complex models much easier and leads to much more spontaneous creative ideas.
Deepnote helped VantAI to radically increase iteration speed, brought interactivity and frictionless collaboration into team sessions and opened access to code and datasets for non-technical scientists and stakeholders. Unlocking this extra creativity and potential to solve some of the world's most important problems is exactly the reason why we built Deepnote.
We look forward to VantAI's successes and encourage anyone interested in building large-scale ML pipelines or cutting-edge biotechnology research to reach out to firstname.lastname@example.org.