Enabling Scientific Research and Analyses Through Automated MLby@ioannistsamardinos
192 reads

Enabling Scientific Research and Analyses Through Automated ML

by JADBioDecember 29th, 2021
Read on Terminal Reader
Read this story w/o Javascript
tldt arrow

Too Long; Didn't Read

JADBio's main activity is developing and commercializing the first Automated Machine Learning (AutoML) tool for molecular biological and biomedical data. Founder Ioannis Tsamardinos: "Our vision is to make Machine Learning a commodity, and through it, discover new knowledge, therapies, and drugs". Automated Machine Learning has the potential to tremendously boost our productivity in analyzing data. Causal Discovery seeks to go a step beyond standard Machine Learning and predictive modeling and identify causal relations and causal effects.

People Mentioned

Mention Thumbnail

Companies Mentioned

Mention Thumbnail
Mention Thumbnail
featured image - Enabling Scientific Research and Analyses Through Automated ML
JADBio HackerNoon profile picture

HackerNoon Reporter: Please tell us briefly about your background.

My name is Ioannis Tsamardinos, and I’m the CEO & co-founder at JADBio. I’m also a professor at the Computer Science Department of the University of Crete. I have over 20 years of service in academia and applied science and over 130 publications in international journals, conferences, and books. My work on AI Planning and temporal reasoning was applied to the NASA Deep Space I software. In my spare time, I’m an enthusiastic dad of two and a spear fisherman.

What's your startup called? And in a sentence or two, what does it do?

Our startup is called JADBio. JADBio's main activity is developing and commercializing the first Automated Machine Learning (AutoML) tool for molecular biological and biomedical data, called Just Add Data Bio, or JADBio for short.

JADBio fully automates the machine learning process allowing life scientists and others to create predictive models without coding. Quite importantly, it performs knowledge discovery by identifying the set of measured quantities that carry all predictive information.

In life sciences, it is important to design cost-effective diagnostics, perform precision medicine, identify drug targets, and understand the underlying molecular mechanisms involved regarding the data. Apart from enabling non-coders to run analyses, it can boost the expert analysts' productivity while guaranteeing the statistical validity of results, enabling the replicability of analyses, and ensuring the data and predictive model provenance.

Our vision is to make Machine Learning a commodity, available to everyone, and through it, discover new knowledge, therapies, and drugs.

What is the origin story?

We started out in 2013 by creating Gnosis Data Analysis, a spinoff company from the University of Crete (Greece) that wanted to develop algorithms and machine learning solutions for biomedical data. Through services, we were bootstrapping the development of our product that we named Just Add Data Bio or JADBio for short.

JADBio as an AutoML platform was born in 2019  by its founders,  Vincenzo Lagani, Giorgos Borboudakis, Pavlos Charonyktakis, and myself.  We developed a business plan, raised funding, found strategic partners, and focused on R&D for JADBio and its commercialization.

What do you love about your team, and why are you the ones to solve this problem?

Our team represents a combination of passion, experience, and expertise. We count 5 PhDs, 2 decades of research, and a roster of smart developers and business people. I love working with smart people and people I can learn from. There is no shortage of strong opinions in our team, so our meetings are never boring 😃 .

Why are we the ones to solve the problem? First of all, why not? Innovation is, by definition, unexpected and doesn’t always come from the “usual suspects”, whoever they may happen to be in someone’s mind. Regarding the specific problem we are solving, we started by building the first AutoML system ever back in 2004! It was called Gene Expression Model Selector or GEMS; the term AutoML was not coined up yet.

The first commercial system appeared 10 years later! GEMS was ahead of its time, incorporating feature selection and performing a correct estimation of performance for small sample datasets, both of which functionalities are still missing from many commercial systems. Unfortunately, it was too early for the market, and I was still too much of an academic at the time to commercialize it. Finally, I believe we possess a unique advantage in bringing machine learning to life sciences: on the one hand, we have a solid foundation in standard AI, what is often referred to as Good-Old Fashioned AI, statistics, machine learning; on the other, we have 20 years of experience in interdisciplinary research with clinicians and biologists. We know how they think, what they need, and how modern AI and Machine Learning can solve their problems.

If you weren’t building your startup, what would you be doing?

I guess I’d be miserable. Once I would stop being miserable, I would try to find the next best thing to do. I think this is the most exciting time for our field. I love research and scientific innovation. I would like to be part of a project that could have a big impact on the world, that has a vision and ambition. These days, much of the research in AI and Machine Learning takes place in big companies or startups, not necessarily academia. So, I would find such a project and try to be part of it.

At the moment, how do you measure success? What are your core metrics?

We measure success with several quantitative metrics. We measure the sign-ups, registrations, number of performed analyses, number of models downloaded, and numerous other aspects of the usage of our tools. Of course, we also measure revenues, partnerships, and other business development metrics. But, we also try to measure the impact and value provided by JADBio to our clients. This is not always known, of course, since we do not know the data the clients analyze, why, or what the results are. So as a proxy, we use the published scientific papers that are based on results obtained using JADBio; so far, there are more than 37 such scientific papers.

What’s most exciting about your traction to date?

I get the most excited when I see a scientific paper published by a client or end-user with the use of JADBio. Especially when the authors of the paper are not machine learning experts but life scientists with no prior experience in data science; people, who a year ago may think that this AI mumbo-jumbo is beyond their reach.

We see clients performing hundreds of analyses in the span of days or a week or so; analyses that used to require several weeks or months, each by experienced analysts to complete. I find that satisfying and a validation of our efforts and research.

Apart from traction with individual users, I get excited with our solution partners. These are third-party software companies with synergistic products. We are making our products interoperable and plan on joined distribution, sales, and marketing. Existing partnerships include Indivumed, QIAGEN, Biolizard, Elucidata, and Ledidi, while more are coming up. The combined offerings of products cover a wider spectrum of user needs. I am very optimistic that when interoperability is achieved, we’ll see a big boost in our joined sales.

What technologies are you currently most excited about, and most worried about? And why?

Of course, Deep Learning is continuously producing super exciting results. It is all over the news and the scientific literature. However, it mostly considers image, video, and text data. I am excited about technologies that are particularly pertinent to molecular and clinical data; Automated Machine Learning and Causal Discovery.

Of course, my opinion is biased because I am involved in research in these fields. Automated Machine Learning has the potential to tremendously boost our productivity in analyzing data. Causal Discovery seeks to go a step beyond standard Machine Learning and predictive modeling and identify causal relations and causal effects. It can answer what-if scenarios, e.g., what I should expect if I perform this treatment or intervention. Questions that are not possible to answer correctly with standard predictive models.

The most exciting technologies are the ones to have the most impact and obviously are also the scariest ones. I am particularly worried about deep fakes and living in a world that we cannot believe our eyes as evidence. I am worried about learning so much about people’s behavior that “they” can manipulate us to their advantage. I am worried that the difference between the institutions (companies, governments) with access to data and AI technology will create substantial societal inequalities. We are fast moving towards a utopia or dystopia; the future will tell.

What drew you to get published on HackerNoon? What do you like most about our platform?

I have known Hackernoon for years and have read tons of content. This platform is a place where you can read exciting and in-depth articles from independent and knowledgeable editors from different fields.

What advice would you give to the 21-year-old version of yourself?

Don’t waste time. Take action… It's never too early. Don’t waste time with small-minded, limited vision people. Dream big. Don’t sell yourself short. Don’t worry; you’ll make it. Worrying takes away from enjoying the trip.

What is something surprising you've learned this year that your contemporaries would benefit from knowing?

Well, in 2021, we’ve seen numerous quality scientific publications in life sciences using autoML tools, from biologists, clinicians, and pharmacologists that knew nothing of AI and Machine Learning before they used it. It was really striking what people can do if they are provided with the right tools. What now seems accessible to a few experts could become a commodity in a few years’ time, available to everyone. I truly believe ML will change our future, and obviously, the tools needed for it to be approachable by all will also see huge growth.