My company, Xpressomics, was born a couple of years ago when I was conducting post-doc work on hypoxia research and ran into problems when performing gene expression analysis on my microarray data. I felt that the analytics process involved too many steps, required help from bioinformaticians and was too time-consuming. As a biologist I needed a simple solution with up-to-date annotations and requiring no programming skills. From that came an idea to develop a gene expression analytics solution, which by today has evolved into a gene expression search engine.
Taking an idea from an academic setting and turning it into a business poses challenges, and below I have highlighted some key lessons. While others may have different experiences, these were the ones that, to my mind, are worth sharing.
A first, key aspect for transferring scientific know-how into a business is an intellectually diverse team. At a very early stage I managed to link up with people skilled in software engineering, cloud infrastructure and business development. I felt that complementary mindsets help to sustain progress and creativity. I would recommend the same for any academic start-up.
Secondly, there’s the vision. You should have a gut feeling of the problem awaiting a solution, and how you are going to do it. In our case, we noted that the process of collecting samples, processing microarrays and analyzing the data was very slow. This would normally take a couple of months. What if somebody had already performed an experiment answering my question and I’m not aware of it?
Our vision was that we felt the data analysis process could be significantly shortened. Considering the accelerating growth of genetic information, we reckoned that an optimal solution would enable individual researchers to tackle big data problems on their own while requiring little computer science skills and on-site hardware. After all, it’s the person who designed the experiment who has the most insight into the problem. In our vision, an easy-to-use application should be able to turn differential expression analysis of microarrays or RNA-seq into something as easy as performing a t-test or ANOVA in a typical data analysis package. Such an aim is fully compliant with the advances of cloud computing, as it is now possible to deliver results from high-performance computation to every laptop running a web browser.
Looking back, it’s interesting to see how the product has evolved over time. Initially the idea was to provide a highly customizable tool for life scientists to analyze their data via a visual programming interface. Yet, after testing it a little while we understood that the product would have to be made simpler to reach a wider audience of researchers. It was a key lesson: the end user perspective of the product is radically different from the developer’s. Next, we understood that performing differential expression analysis was not going to cut it alone. Similar desktop solutions already existed and we had to up the ante, and it was not certain that providing the service solely in the cloud was a strong selling point.
Instead, we took a more general approach and identified the interpretation of data as a major bottleneck. Comparing new data to previously published experiments is probably the most pervasive pattern in the scientific methodology. We started with a pilot study indexing differential expression profiles from around 20,000 microarrays in a multi-arm toxicogenomic study (the Japanese Toxicogenomic Project). Today, we provide a gene expression search engine that allows the querying of genes for differential expression in public data sets. We have specialists curating experimental factors in the meta data followed by differential expression analysis starting from the raw data. Over a thousand experiments have been analyzed, producing more than 25,000 transcriptome-wide gene expression profiles. Experiments are sorted by relevance in response to the query, so that the user can easily identify factors that have most effect on the genes of interest. We expect that the query engine will facilitate new discoveries and provide better overview of gene function by highlighting conditions that affect its expression most. For the sake of simplicity, you can perform the search just by entering one or more gene symbols as the query. And the power of the search engine is growing rapidly as we index new profiles each month.
Here is the third takeaway: get feedback early. The pivots we’ve made have been our way of responding to the comments we have received. This poses a question: do you embrace customer feedback and pivot to new products, or stick with the vision and carry on? It’s difficult to know, but I recommend keeping an open mind and follow one’s gut feeling – this is about as scientific as that process can be made.
And fourth, it is important to remain agile. With only a handful of people we have not had the luxury of spending too much time and resources on development and commercialization. Actually this has been a good thing as it has kept our venture lean and focused. And it will serve us well as we develop our solutions for the future.