Based in London, Desktop Genetics is startup that combines the two buzz words in science: CRISPR and machine learning. We met them at the Pioneers Festival in Vienna and we spoke with their Chief Executive Officer and Technical Lead, Riley Doyle.
WUNDERDING: What are you doing at Desktop Genetics?
Riley Doyle: We’re a biotech startup based in London that works on making CRISPR genome editing safer and more effective with machine learning. We do all of the informatics that need to go into designing an effective CRISPR experiment – from sequencing the genome of the target cell, analysing it, designing the optimal guide RNAs to go in, and getting those manufactured and delivered to the end user who is a scientist or, one day, a doctor somewhere, and then analysing the results to make sure that there are no off-target effects and that we have introduced the intended change with no other unwanted side effects.
WUNDERDING: There is this CRISPR hype. Has CRISPR really revolutionized genetic engineering?
Riley Doyle: It’s a really exciting time right now, because after 40 years of work in genetic engineering and gene therapy, we finally saw the first two gene therapies approved in Europe last year. We have the first cell therapies likely heading to approval this year [NB: Novartis’s therapy approved last]. The first CRISPR clinical trials were going to start this year – though they have been pushed back to 2018. It’s a really dynamic time in the industry. We’ve got these sci-fi therapies coming onto the market that can really make a difference in a whole range of otherwise really untreatable conditions.
WUNDERDING: What’s the reason for this kind of CRISPR hype at the moment?
Riley Doyle: I think CRISPR has been very popular since it came on the market in 2012-2013. The reason for that is that it works okay most of the time – even if you don’t do anything terribly sophisticated. But that’s not good enough for high-impact medical applications. You have to be much more precise in your genome editing as there is a serious risk of introducing unwanted edits to the patient’s cells and, for example, causing leukemia – that was a big problem back in 1998 with the earlier type of gene therapy. So, we partner with the experts in a particular disease or application area and help them design their CRISPR vectors. And then we also handle the sourcing and manufacturing of those vectors so they get the final vector ready to use. And this is quite important because different manufacturers have different levels of expertise in the different chemistries that can be used to manufacture a guide RNA. It’s important that we work with them and vet their capabilities and their quality and things like that. We’ve already assembled that network of partners, so we’re able to supply high quality and consistent reagents to our users and our partners – which is one of the more subtle and less widely reported aspects of CRISPR. There’s the whole supply chain that needs to come into play now, so we do that in addition to the design. You can think of it in the sense that if the end user is the person that has the particular problem, we then help them by providing a one-stop solution to get all of the different pieces that they’re going to need in order to have a successful experiment.
WUNDERDING: Could you please tell us more about the origins of Desktop Genetics?
Riley Doyle: We were founded in 2012, where Victor, Eddie and I met during grad school at Cambridge. We won the business plan competition that year and then we went on to found Desktop Genetics that August. We’ve been based in London ever since. We also started up another office in Kendall Square in Boston which is the heart of the CRISPR industry at the moment.
WUNDERDING: I am a scientist or a company and I have heard about CRISPR. I would like to run my own experiments using CRISPR since it is supposed to be cheap and very easy to use. Why should I contact Desktop Genetics before I start my work?
Riley Doyle: One of the things that would happen if you wanted to be an expert on CRISPR is that you would have to go back to 2012 and read about 2 to 6 scientific papers a week to stay on top of the field. We have the largest collection of experimentally validated guide RNAs in the world that we are able to feed into our machine learning pipeline. We are also able to keep track of all of the rules that have been generated from all of the publications, while staying on top of our own work and our work with customers. Thanks to that, we’re always up to date with the latest best practices in CRISPR – so it’s a good partnership for people who’ve heard about this very powerful technique and who want to use it to solve some problems. They don’t have a time machine. They can’t just stop everything and start reading papers for years and years to try and catch up, which is why there’s a good synergy between us. I think that it’s going to be really important as more and more laboratories adopt this technique and work in the field of CRISPR.
WUNDERDING: Could you please give us an example of what you are doing?
Riley Doyle: We worked with some researchers at a government lab. They had a cell line, a model of cancer, and they wanted to introduce some targeted mutations to see if those mutations confer resistance to chemotherapy. We helped them design that experiment, source the molecules and ingredients they needed to do CRISPR, and then execute that and verify that it was done properly. We then worked with a pharmaceutical company who wanted to go and do a new drug discovery program to find out which genes were essential for cancer cells to survive. They needed to design tens of thousands of guide RNAs and as we already have this AI – the DeskGen platform – that can do this, we were able to go and design this vast array of CRISPR experiments for them in moments. And the other thing we’re able to do, is to take their prior experimental data and incorporate that into the design process to improve the performance by a factor of two. As a result, they were able to get a much more cost effective and efficient experiment. As a third example, we had a company which had been working on a cell therapy. Regulators came to them and asked them to show that they didn’t introduce any off-target effects on their cells and so we worked with them to do very high coverage sequencing of those edited cells and compared them to the unedited sample to show that, yes in fact, they were likely safe to introduce into patients. And thus, they were able to begin the clinical trial.
WUNDERDING: CRISPR is a very versatile tool and there are so many possible use cases. Nevertheless, do you think there are challenges when, for example, using CRISPR in agriculture to breed plants with extraordinary properties?
Riley Doyle: One of the biggest challenges actually, is in crop science, where we know a lot less about plant genomes. They are quite complicated, and historically plant research has been underfunded relative to, say, human medicine. We know a tremendous amount about the mouse genome and the genome of these model human cell lines. In fact, we’ve shown that prior information improves the performance of your CRISPR experiment and your ability to effectively predict the guide RNA performance. And what’s interesting, is that crop sciences then don’t have that information, so there’s a lot more ground to cover and to catch up on. Another example would be in industrial applications of synthetic biology, where you have to modify a lot of the standard techniques to work with those particular genomes. For example, the way you deliver the guide RNA to the cell has to be changed. The other thing is that we have a number of design rules for the guide RNAs, and these are encoded into a machine learning model. There are about 4906 predictors of guide RNA performance that we look at and that we have demonstrated are necessary to show how well they are going to work. And those parameters are different for the different species. We’re not entirely sure why this, is but that’s what the data seems to suggest at this time and so therefore, what works well for editing a human cell line is not necessarily going to apply to wheat or barley or dogs or horses. You actually have to take this dataset and re-run your AI over it to develop a new model for that particular genome.
WUNDERDING: Another problem is what you call the “annotation of the genome”. What does it mean?
Riley Doyle: The annotation of the genome is another big problem. You can design a guide RNA but where you design that guide RNA is based on something called „genome annotations“. It is sort of the map. And if your map is wrong – which it actually is in a lot of cases – then you’re going to be cutting somewhere completely different than where you thought. So even if your guide RNA was the perfect guide, it’s not going to do what you want it to, because you were trying to cut this chain but actually that chain lives somewhere else. We’ve seen this happen in a number of cases. One thing you have to remember is that every person here at this conference has 45 million variations in their genome relative to the person sitting next to them. And those variations will change the effect of the guide RNA and how well it works, and which guide RNA you should choose to do a particular type of experiment. And that same principle is even more true when looking at other organisms, other cell lines and things like that. So there’s a tremendous amount of data that needs to be curated and managed in order to effectively do CRISPR at an industrial scale. I think one of the bits and pieces that we have to remind people of, is that they come to us with a very novel organism and then that organism’s genome has barely been sequenced – they don’t even know where some of the genes are in it. But if you don’t have that bioinformatics knowledge, you actually can’t really do the most effective CRISPR experiment. So we help them understand what’s available, what’s known, and we then work with them to take, for example, data their lab generated using earlier techniques and improve those models so that they can go and do a better experiment.
WUNDERDING: Thanks a lot for the interview.
You could watch the whole interview as a video:
And if you would like to read more about CRISPR, you could read our interview with Rachel Haurwitz from Cariobou Biosciences.