OpenBioML - Democratizing Scientific Research

About OpenBioML

Inspired by EleutherAI, OpenBioML aims to become an open, collaborative research laboratory at the intersection of machine learning and biology. From discussing the latest developments to teaming up for cutting edge projects and reproducing closed-source research. We seek to maximize the positive impact of artificial intelligence in life sciences.

We have access to large-scale computational resources and would like to utilize them to accelerate research. If this is something that interests you consider proposing a collaborative research project or getting involved through our discord!

DNA Diffusion

Understanding the code of life: generative models of regulatory DNA sequences based on diffusion models.

The goal of this project is to investigate the application and adaptation of recent diffusion models to genomics data. Diffusion models are powerful models that have been used for image generation (e.g. stable diffusion, DALL-E), music generation (recent version of the magenta project) with outstanding results. A particular model formulation called "guided" diffusion allows to bias the generative process toward a particular direction if during training a text or continuous/discrete labels are provided. This allows the creation of "AI artists" that, based on a text prompt, can create beautiful and complex images.

A collaborative project initiated by Luca Pinello's lab.

Read project proposal

CORE TEAM

Our core team, community members, and generous resource provider, Stability AI are responsible for making all this work. It is thanks to everyone's contributions that OpenBioML is what it is today.

Niccolò Zanichelli

Undergraduate computer science at the University of Parma.

Keaun Amani

Synthetic biology enjoyer and founder of neurosnap.ai, a platform for making machine learning accessible to biologists.

Michael Pieler

DL engineer with a biotechnology background interested in Bio-ML research with a focus on multi-modal setups and large bio models.

Our Mission

Most research groups lack the compute necessary to take on cutting edge research projects as well as to rapidly iterate on novel ideas. The recent success of AlphaFold2 is a testament to this fact: in order to train the final model DeepMind's team utilized 128 TPUv3s for several weeks, but this does not take into account all the resources previously employed to come up with the final architecture. Few if any academic labs have access to such large-scale resources, yet it is clear that talent and innovative ideas can be found all over the world.

It is disproportion between resources available and talent that motivates us: by creating a decentralized research laboratory with access to computational & storage resources previously available only to the most well-funded industrial research labs, we seek to support the broader community and incentivize the release of the most advanced predictive models current technology can offer.

It has been made repeatedly clear that machine learning models often display surprising emergent capabilities that are often times difficult to foresee. AlphaFold2 for example was found capable, of predicting quaternary structures. This capability was discovered by researchers experimenting with the model soon after the release of its checkpoints. We all stand to gain from improved biotechnology, and if machine learning is to become increasingly central to computational biology, we need to ensure these capabilities are discovered and exploited to the fullest.

Become a Member

Democratizing scientific research