We are at an incredible inflection point in the traditionally distinct fields of information technology and the life sciences. These broad disciplines are converging to advance fields of study like computational biology and systems biology, resulting in new forms of therapeutics and diagnostics that could have a monumentally positive impact on human health.
At DFJ Venture Capital, we are seeing innovations in multi-modal data generation, computational algorithms, and robotic automation, enabling companies to take on the audacious goal of curing illness and disease for everyone forever.
The gene is the atomic level of biology that encodes the information that gives us life. No kidding, it’s important. In just the last several years, there’s been an explosion of genomics data driven by the cost of sequencing falling faster on a per genome cost basis than even Moore’s Law could have predicted.
The first human genome cost $3 billion to sequence; today there are companies that can do a whole genome sequence for less than $1,000. The race of genomics innovation is unprecedented, and there are no indications that it will slow down anytime soon. It’s a big data world in the human genome.
And really, the genome is just the beginning. We have been able to expand the breadth of data from genome sequences to extensive transcriptomic, methylomic, and metabolomic data.
With this fusion of data inputs, we may be able to draw new correlations between genomic variation and human phenotypes, providing new insights into the drivers of human health at the cellular and molecular levels, and thereby increasing our understanding of metabolic diseases, psychiatric diseases, and cancer, to give us a better shot at living forever.
Deep Learning
All of this data is wonderful, but big data by itself is a big headache. We need to have a way to make sense of it. Not only are technologies for generating genomic data improving, but new computational paradigms developed from other application areas are now able to generate actionable insights from raw data inputs, including genome sequences.
Specifically, I’m talking about deep learning, which is a subset of machine learning that has improved the state-of-the-art in computer vision, natural language understanding, speech recognition, and genomics. It works with almost everything!
Deep learning allows computational models that are composed of multiple processing layers (or neurons) to learn representations of data with many levels of abstraction. It has turned out to be very good at discovering intricate structures in high-dimensional data. What that means for non-experts is that your raw genomic data goes in and wonderful insights come out the other end without you having to learn any of the math in between. I’m being a bit idealistic here, but we’ll get there eventually.
At DFJ, we have many application-level investments in deep learning, but Nervana Systems, a San Diego-based company, took on the ambitious task of building a