Creating new drugs is hard. Pharmaceutical companies spend years and millions of dollars to produce new types of molecules in labs, few of which prove effective and make it into medicines and therapies. VantAI is at a forefront of pioneering a new approach, which uses ML and custom algorithms to discover potential molecules in-silico.
Part an AI startup and part a research lab, VantAI combines power of ML and large amounts of pharmaceutical data with their expertise in biology for computational design of targeted protein degraders, a core building block in treating diseases.
But how do you operate a partly remote team of brilliant scientists and engineers, together with pipelines processing terabytes of data on thousands of GPUs efficiently? Here, Deepnote helps VantAI design experiments, share, discuss and document results and enables research explore massive datasets and complex algorithms quickly, so the team can focus on their research for cutting-edge drug discovery.
Experiment design start with a researcher preparing a preliminary outline of the experiment as a Deepnote project, using a prebuilt Docker environment to bring in VantAI's code and dependencies. Datasets, usually hosted in S3 or GCS buckets are connected through Deepnote integrations. During a meeting, the rest of the team can jump in to the project and collaboratively tweak the parameters of the experiment, explore data or feature sets and contribute ideas. VantAI is using large GPU pipelines running on Kubernetes and latest ML stack based on PyTorch Lightning, so it's not quite common to bring several researchers and engineers together to fine tune the experiment.
VantAI keeps a weekly team-wide check-ins where team's results are shown and discussed. There is a lot of code, datasets analysis, experiment results and other important knowledge. To discuss all of this efficiently, all reports are made in Deepnote notebooks and shared ahead of meetings, allowing members to familiarize themselves with latest work and comment on it, significantly reducing meeting times and enabling VantAI's team to work asynchronously.
Deepnote enables researchers to quickly run VantAI's in-house developed algorithms and connect datasets without wasting time on environment setup. Each tool or package is distributed with a link to a Deepnote project. When a researcher wants to use it, they open the project, which is templated with a Docker image from VantAI's private repository, containing the package, all dependencies. It also includes documentation and how-to examples. With everything in one place, researchers can focus on data analysis instead of installing dependencies and scavenging for documentation. The same goes for data exploration. Deepnote projects are integrated with S3 or GCS bucket storage (along with BigQuery, Drive and many other storage solutions), providing instant access in an environment equipped with all tools needed to explore the particular dataset.
Enabling companies to solve important problems
Deepnote enabled VantAI move more efficiently and replaced legacy reporting tools with one single source of truth with a handy python runtime built-in for impromptu experimentation and exploration. Most importantly, VantAI is now able to remove manual toil for researchers and streamline knowledge sharing across the company, and focus on solving hard problems. Unblocking companies like VantAI and enabling them to fully focus on solving some of the world's most important problems are exactly the reasons why we built Deepnote.
We're looking forward to VantAI's successes and encourage anyone interested in building cutting edge ML pipelines, bioinformatics or related research to reach out to firstname.lastname@example.org.