New Frontiers: World’s First Community-Driven AI Store for Biology
Tetra Partner Network Spotlight Series
Larry and Sergey. Bill and Paul. Steve and Steve. It’s hard to imagine, but the greatest companies, Google, Microsoft, and Apple, were founded by just two inspired people with a vision to change the world. In the spirit of honoring entrepreneurs who dare to launch businesses with an original mindset, we’re delighted to launch our Spotlight Series on emerging companies by showcasing Superbio.ai and its co-founders, Berke Buyukkucak and Ronjon Nag.
Before we dig in, please tell us about your backgrounds.
Ronjon: My name is Dr. Ronjon Nag, and I am an inventor and Adjunct Professor in Genetics at Stanford University School of Medicine. I’m also Founder of R42 Group, which invests in AI, biotech and science, and wireless communication technologies, and works with pre-seed startup. I am also the Chairman of Superbio.ai, the startup that Berke founded in 2021.
I currently teach artificial intelligence, longevity science and technology, and healthcare venture capital at Stanford. I have deployed AI systems for 30 years, having founded, advised, and sold companies to Motorola, BlackBerry, and Apple. I received the Institution of Engineering and Technology Mountbatten Medal. In 2021, I received both the IEEE-SCV Outstanding Engineer Award and the IEEE-USA Entrepreneur Achievement Award.
Berke: My resume is only slightly shorter than Ronjon’s. I received both Bachelor’s and Master’s degrees from Brown University in biomedical engineering. I was more of a traditional biomedical engineer. With an amazing group of students from Brown and Rhode Island School of Design, we designed award-winning medical devices. I worked as a wet lab scientist and conducted neurosurgeries on rats to understand pathways of sensory shutdown. I spent significant time teaching myself how to code, I worked as a robotics engineer, machine learning engineer, lab automation engineer before I said “Enough! Why are there no software tools to help me do my work?” And that’s how my Superbio.ai journey began.
What is superbio.ai ?
Both: Superbio.ai is a cloud based platform to implement AI for biology with no code. We really want to accelerate machine learning for biologists – both in the implementation and in the running of experiments. By implementing in the cloud we can also optimize computation costs. Anyone can run machine learning even if they are not a programmer and we have biology specific tools specifically for the biologists.
Scientists and researchers can submit models and tools directly to superbio.ai. We will compensate researchers fairly as well.
Scientists and researchers can submit models and tools directly to superbio.ai. We will compensate researchers fairly as well. If you contribute to the community with a model or tool that’s widely used, Superbio.ai will generate passive income for you!
How did you meet and why did you create superbio?
Berke: In the spring of 2020, I accepted a job offer from a leading cancer diagnostics company. It didn’t take me long to realize I still had an itch for learning, and I could not accept that I had graduated from school and that my education career was over. Living in the Bay Area, Stanford University was an obvious choice to continue learning. Considering my experience and curiosity about machine learning, I wanted to dig deeper, so I jumped down the rabbit hole and took a class called “Intro to AI.” Guess who the professor was: Ronjon!
After 8 weeks, and during the last class, Ronjon mentioned that R42 was offering a fellowship via R42 Institute, a highly selective and truly diverse program where fellows are assigned to a project, usually designed by Ronjon. I was rejected when I first applied! I didn’t back down, though, I came up with 3 project ideas and pitched them to Ronjon, one of which was Superbio.ai, after which I joined the fellowship as a mentor (youngest mentor to this day) leading my own project.
Making AI easy to use is a decades-long project for Ronjon, and I brought the biomedical vertical to it.
Making AI easy to use is a decades-long project for Ronjon, and I brought the biomedical vertical to it. For a long time, Superbio.ai was a side project but we kept working on it. We’d get together every week and discuss how we could move the project forward and who we could talk to. In less than a year, we grew confident that we could raise funds to build out our team and did so by the end of July 2021.
Your community-driven approach to solving the reproducibility crisis is quite unique. How does it work?
Both: We’ve talked with companies and researchers who spent months trying to reproduce academic research projects. Arguably the best models are being published by academia and they are available on Github; but good luck finding the right packages, setting up the right environment, and finding the same computational resources to achieve similar results. Considering scientific progress depends on understanding and building on top of each other’s work, we can somewhat understand why the life sciences industry is slow to adopt AI. Research is stuck with archaic AI consulting models that leave the actual scientist out of the equation.
We’re creating an open-source platform where we transform user-submitted pre-trained AI models (in life sciences applications) into no-code apps available for inference and retraining.
We’re creating an open-source platform where we transform user-submitted pre-trained AI models (in life sciences applications) into no-code apps available for inference and retraining. Essentially, Superbio.ai acts as the middleman who makes sure the models and data is usable and reproducible. Users are also provided an optimized computing environment, whether they work in a lab with a limited budget or in a pharma company with unlimited computing power, they’ll have access to the highest and most affordable cloud standards in biomedical AI. Results will be the same no matter what.
Finally, Superbio.ai is also an environment-friendly solution. Training deep learning models from scratch requires high energy consumption and incurs a large environmental cost. We allow the life sciences sector to reduce its carbon footprint while accelerating its progress: a win-win situation if we know one!
In what research areas does superbio focus? What types of modalities or datasets, and how many?
Both: We’re interested in every life sciences application and dataset: DNA sequencing, single-cell transcriptomics, drug-target interactions, medical images, and many more. The biomedical community has come a long way in making data openly available for research, but data is siloed in many databases. That’s why not only are we aggregating open datasets in our platform, but we are also creating direct access to biomedical databases so it’s super easy to both find and analyze data.
So far, we’ve aggregated over 250 datasets from public health repositories, drug discovery databases, and transcriptomics datasets, and created direct access to the National Center for Biotechnology Information (NCBI) Gene Expression Omnibus (GEO) data repository.
So far, we’ve aggregated over 250 datasets from public health repositories, drug discovery databases, and transcriptomics datasets, and created direct access to the National Center for Biotechnology Information (NCBI) Gene Expression Omnibus (GEO) data repository. Data accessibility is an extremely important part of our offering and in the long run, we rely on the community to share their data so that it’s available and accessible to everyone.
How are academic and commercial customers adopting your platform and using it?
Both: Superbio.ai is free for academia, forever. And we’re lucky to have a long waiting line of academics who signed up for early access during our pre-launch phase.
For industry, we’re currently onboarding a few pharma companies for beta-testing and working with a few others to solve their AI problems. Soon we’ll allow companies to sign up and start using Superbio.ai as they like.
Why is machine learning so important, and yet so difficult to achieve, in drug development?
Both: Where do we start? Data availability, the rising cost of cloud computing, lack of reproducibility… In simple terms, there are not enough AI experts experienced in life sciences applications. One of the most important think pieces on this issue was written by Dr. Jesse Johnson in his newsletter Scaling Biotech. In the article, The Great Hidden Problem Plaguing Tech Biotechs, Dr. Johnson introduces the concept of “shared mental models” in the context of exploring the gap of understanding between wet lab teams and data science teams in biotech companies. Essentially, interdisciplinary teams are required to commit extra time and effort to communicate with each other and that is a more difficult challenge than it sounds. Right now, the industry expects biologists who spent years excelling in biology to excel in data science and vice versa. That’s an unrealistic and unimaginable burden.
Even if in every company or lab, teams arrive at a shared mental model which allows them to work efficiently with each other, there is no standard practice for collecting and sharing data in the world. Today, you can find the same information provided in different ways even in one biomedical database. If you can’t validate your predictions over many datasets with ease, that’s a big obstacle. But how we collect experimental life sciences data differs from one lab to another in the whole world!
Superbio.ai is spearheading the revolution in life sciences by materializing what a shared mental model between wet lab and data teams could look like.
Superbio.ai is spearheading the revolution in life sciences by materializing what a shared mental model between wet lab and data teams could look like. A no-code platform where you can find data, training tools, and pre-trained models in an optimized cloud environment which can be used by anyone in the field no matter what their experiences are. Our goal is to unite everyone and standardize how we conduct AI-driven research.
Is it possible that deep learning models will eventually function autonomously, without human-written code?
Both: Most certainly. We’re already introducing AI into the most delicate parts of our society like the judicial system. However, a model is only as clean as the data it’s trained on; biased data or algorithms generate biased models. And biased models can hurt innocent people. Therefore, even though AI science will move towards autonomy, to fight the bias we need humans in the loop for model due diligence.
At Superbio.ai, every model and data that are submitted will go through both automated and manual quality checks before they’re made available to the public.
At Superbio.ai, every model and data that are submitted will go through both automated and manual quality checks before they’re made available to the public. And if we miss anything, we'll rely on the community to call the models out: by allowing them to rate models and the best models will float and be more visible while the worst sink. We’re lucky to have experts on AI ethics on our advisory board who help us develop Superbio.ai within an ethical framework.
What’s your vision for how deep learning models will ultimately contribute to life sciences research?
Berke: Ronjon always makes this half-joke to biologists: “You’re telling me you’ve been using trial-and-error all this time?!” Deep learning will accelerate science and research by finding patterns in data. The pattern matching and extracting abilities will fuel innovation in precision medicine. It will also become a support tool for patient care.
Superbio.ai will be the locomotive of progress as the only scientist-friendly AI platform in the field.
What can we expect to see from superbio in the upcoming months?
As founders often say, “We’re just starting!”
Both: We’re more than thrilled for the coming months: beta launch, new features every month, more data, more models, community engagements, and research partnerships. As founders often say, “We’re just starting!”
If you wish to contribute to the world’s first AI store for biology with your model or tool, submit it here!