NEW YORK (GenomeWeb) – Deep Genomics, a Toronto-based bioinformatics startup, is seeking to build a business around a series of products that use machine learning techniques to identify genetic variants that alter cellular biochemistry and lead to the development of diseases like cancer among other genetically motivated ailments.
This week, the company launched the first of several planned products for the genomics market. The so-called Splicing Index (SPIDEX) uses a computational model, which leverages deep learning techniques in particular, to search a database of millions of genetic mutations culled from peer-reviewed resources. This particular iteration of the model is tailored to identify mutations that affect RNA splicing, changing cellular biochemistry in ways that lead to disease.
It's the first of several planned products based on a general modeling approach that was developed by researchers in the laboratory of Brendan Frey, a professor in the University of Toronto's departments of electrical and computer engineering and Deep Genomics' president and CEO, along with collaborators at a number of other institutions. The team published a paper in Science last December that described their technique and, more specifically, its application to identify variant-driven splice alterations involved in neurological disorders like autism as well as cancer.
The particular model described in Science is now being commercialized as SPIDEX. Over the next five years, Deep Genomics plans to develop and launch additional models — about six months apart — using the same deep learning techniques that will be able to identify variants that impact transcription, polyadenylation, RNA stability, translation, and protein stability in various diseases, Frey told GenomeWeb. When the company completes its development efforts, it will offer a "comprehensive" infrastructure for identifying genetic variations that impact and change cellular processes.
One of the main benefits of the system is that it's focused on alterations in cellular processes, which makes it agnostic to disease, Frey said, unlike many existing technologies that are trained to hunt for mutations in the context of particular diseases. "It's a little bit like teaching someone to read by giving them the text of the book and telling them what the gist of the story is," he said. Deep Genomics' approach on the other hand is more akin to providing pictures that accompany each sentence within the book, thus teaching the learner to associate specific bits of text with specific images.
"By providing lots of examples like that, our computer system learns how to read the text of the genome," he explained. "It doesn't know anything about disease, it just knows how to read."
The result is a system than can read the genome, identifying how changes affect genes and proteins in the cell, and it then uses that information to make causal mechanistic predictions that could be relevant to disease, he said. "So instead of just saying that this mutation looks like it's correlated with disease, our system can say this mutation is going to cause a decrease in this particular protein, and the decrease in the protein is going to associate with that disease. It gives us that mechanistic step, and that's really what's crucial about our technique compared to others."
Moreover, the system is able to generate networks of variants that highlight connections between mutations that change cellular biochemistry as well as mutations that have similar mechanisms of action, providing additional information that could be useful to researchers in diagnosing novel genetic diseases, he said. Although the company isn't targeting a specific disease area, Deep Genomics does have interests in studying cancer and neurological diseases like autism and plans to work on projects aimed at identifying biomarkers that could be used in diagnostic tests in these areas, Frey said.
Deep Genomics' products would be most useful to genetic testing companies such as Invitae, Personalis, and Counsyl, as well as molecular diagnostics laboratories in hospitals and clinics, Frey said, and it is targeting these sorts of users initially with its products —so far, the company has secured contracts with a number of unnamed customers in the genetic testing space in both academia and industry. The company also believes that its offerings can be of use to pharmaceutical companies seeking to identify new markers that could serve as drug targets as well as to companies involved in gene editing, and it plans to target these firms in the future.
Its pricing model is still in flux and does vary to some extent from customer to customer, but Deep Genomics currently has two options for customers. It charges clinical laboratories an undisclosed fee per sample for analysis. For firms with more of a research and development focus, the company offers an annual license that gives users access to the database and model.
"We are [also] very open to collaborations and partnerships, and we [offer] other pricing models that have more to do with mutual benefit," Frey said. In one such partnership, Deep Genomics might contribute its analysis capabilities offering a reduced fee per genome analyzed while a partner company might share its proprietary datasets with Deep Genomics — these datasets can be used to train and validate the computational models that underlie its system.
In addition to last year's Science paper, a second paper from Frey's lab describing a new genome analysis technique based on deep learning is about to be published in Nature Biotechnology. Their efforts have yielded insights into how genomes influence life and disease, according to Frey, however, "[we] realized that if we really wanted to transform medicine ... we needed to address the medical community as clients and be fully responsive to [their] needs ... and ultimately responsive to the needs of patients and families that are dealing with the problems of genetics," he told GenomeWeb.
To that end, Deep Genomics officially opened its doors last October and spent the last several months putting all the necessary components in place, including drawing up business and technology development plans, exploring investment opportunities, and signing intellectual property agreements. "We are trying to focus on being careful and doing good science and coming up with high-quality products and information that will benefit our clients," Frey said.
Deep Genomics also is talking to a number of venture capital firms and potential investors about possible financing opportunities, Frey said — the company is currently supported by an undisclosed amount of funding from angel investors.