fbpx

4 Companies Using Deep Learning for Drug Discovery

Table of contents

In a recent article we talked about how “deep learning” is a way to analyze “big data” in order to find obscure relationships in massive data sets that help you learn things that would take years and years of research to discover. Take drug discovery as an example. As investors, we like to fast forward to the FDA approval process when in fact, many years of research are needed to identify a drug compound that might be effective in combating a disease prior to seeking FDA approval. In an attempt to dramatically shorten the drug discovery life cycle, some software startups are developing platforms that enable you to discover new drug compounds and predict their likely success, solely by applying proprietary “deep learning” algorithms to ginormous data-sets. Here are 4 startup companies using “deep learning” for virtual drug discovery.

TwoXAR

Click for company website

Palo Alto-based TwoXAR (pronounced “two-zar”) was founded in 2014 by two men, both named Andrew Radin. While how these guys met is a story in itself, this Palo Alto Company raised $3.4 million just a few months ago in a seed round led by renowned tech investor Andreessen Horowitz. TwoXAR uses their DUMA™ Drug Discovery platform to evaluate large public and proprietary datasets to identify and rank high probability drug-disease matches in minutes rather than years. Here’s how they do it:
Two_XAR_DUMAWith under 10 employees, TwoXAR has managed to test their technology on more than 20 diseases so far. The Company is also a part of the elite Stanford-backed “StartX Med Program” and is working with academic researchers at the University of Chicago and Michigan State University to develop their platform along with some unnamed bio-pharmaceutical organizations.

Atomwise

Click for company website

Founded in 2012, San Francisco based Atomwise was spun out of high-profile tech incubator Y Combinator and has taken in $6.3 million in seed funding from such notable technology investors as Draper Fisher Jurvetson and Khosla Ventures. The Atomwise software platform currently runs on an IBM Blue Gene/Q supercomputer, and all that computing power allows them to do things like evaluate 8.2 million compounds to find potential cures for multiple sclerosis in a matter of days. Last year they announced some progress made on finding a cure for ebola. In addition to working on confidential projects with Merck and Autodesk, Atomwise is also conducting ongoing research efforts with both academic and corporate clients.

Insilico Medicine

Click for company website

Founded in 2014, Baltimore based Insilico Medicine has taken in a seed funding round for an undisclosed amount from several small VCs in Hong Kong. Based out of Johns Hopkins University, the Company is developing 4 product offerings as seen below:

Insilico_Products

Insilico refers to themselves as a “contract research organization”, and they have published +50 peer-reviewed publications in 2014-2015. The Company is closely partnered with Nvidia, and uses their GPUs (graphics processing units) to run their algorithms on. In addition to a large number of academic collaborations, they are also working with Novartis, Champions Oncology, and multiple other undisclosed partners. Their goal is to become “the largest drug discovery company in aging and age-related diseases“. Maybe they should also partner up with Human Longevity or Calico.

Berg Health

Click for company website

Founded in 2006, Berg Health is backed by Silicon Valley real estate billionaire Carl Berg. Similar to the other startups we’ve covered here, Berg Health wants to use their “Berg Interrogative Biology™ discovery platform” to cut the time, and expense, of drug development “in half.” The company has developed what might be the first cancer drug identified by artificial intelligence, BPM 31510, that is already in early-stage trials. Here’s a list of their partners:

Berg_Partners

With +200 employees and an early start, Berg Health seems like they are ahead of their competitors for now with an actual drug candidate and an IP portfolio of +450 patents pending or issued.

The power of “deep learning” for drug discovery shows incredible promise. The most fascinating thing about this application of deep learning is that the builders of these algorithms don’t actually need to have any medical expertise at all. Just by teaching the computer how to learn, they are able to uncover extremely obscure relationships that help us learn more about how to effectively treat diseases, even ones that don’t exist yet. The technology could also have limitless applications since most these deep learning solutions are industry agnostic!

For retail investors, the closest thing to a publicly traded play on deep learning is IBM (NYSE:IBM) at the moment. Unfortunately, you won’t get any pure-play exposure to deep learning by buying shares in IBM. Retail investors will need to wait for exposure opportunities via corporate mergers/acquisitions or IPOs in the coming years.

Since “deep learning” technology is in the very early stages of development, we’re probably going to see quite a few more funding rounds before any of these companies look to raise capital through an IPO.

Share

Leave a Reply

Your email address will not be published.

  1. I believe, it is a bit misleading oversimplification that the specialists developing “deep learning” algorithms may not have any medical degree or expertise. At least, a very close collaboration with pharma is needed as the problem here is to develop appropriate “readouts” of information, which is crucial for modeling. Without knowing and understanding what to analyze and how it can be interpreted, any machine learning in itself is blind.

    1. Thank you for the comment AndriiB.

      You’re right but the point being made is that the guys building those algorithms can be industry agnostic. If you told an deep learning engineer that you had 50 input variables and then 5 output variables that you wanted to optimize based on tweaking certain inputs, the deep learning algorithm wouldn’t care what problem you were trying to solve. Nor would the engineer. If you can perfect a deep learning algorithm to “learn” then it should be able to “learn” across all industries. Setting up the experiments and interpreting the results certainly need to be done by subject matter experts in each industry.

  2. Hello, the whole thing is going perfectly here
    and ofcourse every one is sharing facts, that’s actually fine, keep up writing.