Since the human genome was successfully mapped in 2003, researchers have been making use of technology to organise the growing mountain of genomics data into a form that will eventually benefit actual patients.

When he unveiled the Precision Medicine Initiative in 2015, US president Barack Obama recognised the potential that technology and gene mapping can bring to patients.

© iStock Photo: SolStock

His state of the union address laid out a vision for the groundbreaking initiative: “Doctors have always recognised that every patient is unique, and doctors have always tried to tailor their treatments as best they can to individuals. You can match a blood transfusion to a blood type - that was an important discovery. What if matching a cancer cure to our genetic code was just as easy, just as standard?"

Now innovative startups across the world are taking on this exact challenge, bringing cutting edge artificial intelligence and machine learning techniques to the genomics space and building tools that give medical professionals the means to deliver personalised medicine through more effective diagnostics and drug discovery.


The USA - whether it is thanks to the Obama administration or not - appears to be ahead of the curve when it comes to bioinformatics, the name given to the practice of applying computer technology to the management of biological information. The website AngelList alone shows 152 startups that identify themselves as working in the bioinformatics space, with just 10 residing in the UK and a further 17 across Europe. 

Read next: UK AI startups to watch: the hottest machine learning startups in the UK

Without a precision medicine initiative of its own, European startups have had to go it alone when it comes to collating this biological information. The CEO of leading UK bioinformatics startup BenevolentBio - one half of UK AI startup BenevolentAI - Jackie Hunter told Techworld that since being founded in 2013 a big task for them has been "to find standardised ways to collate, correlate and annotate that data".

The incumbent model

Despite being funded to the tune of £72 million to date, BeneovolentAI is competing in a world where a big pharmaceutical company like Pfizer can spend nearly $8 billion a year on research and development (R&D). But startups have the edge when it comes to agility and technological expertise.

The existing drug development process can take at least fifteen years, costs upwards of tens of billions of dollars and is high risk, according to Hunter. "Most people in the pharmaceutical industry work on things that never make it to market or even to man," she told the recent artificial intelligence in bioscience symposium at the Royal Society in London.

Read next: Everything you need to know about deep learning and neural networks

Improving drug efficacy is Hunter's mission. The idea is to expedite the traditional research process by ingesting huge amounts of data and literature using proprietary algorithms and NVIDIA’s DGX-1 deep learning supercomputer.

BenevolentAI then builds tools on top to match the workflow of drug discovery scientists so that they can easily mine and interrogate the data, ideally reducing the time to market for important drugs.

Fail fast mentality

Current CEO of BenevolentTech - the technology arm of BenevolentAI - Jerome Pesenti comes from a software background, previously working on IBM's Watson project, and he is bringing a few of those techniques to the drug discovery process. 

Pesenti speaks often about building a feedback loop through having both the AI researchers and bioscientists in one place. Controlling the process from end-to-end helps them move faster. “We can go from the conception of the AI to putting it in the hands of scientists, to getting the drugs, to collecting new data through actual trials in animals and in humans and leveraging new technology to interpret that data.”

Read next: Google’s DeepMind promises openness as it begins public consultation over healthcare plans

One key advantage of failing fast in drug development is being able to repurpose a compound for a different target disease, rather than giving up.

"I come from the software world and we talk about fail fast and that's the key there," Pesenti said. "You want to kill a lot of ideas really quickly to get the best one, whereas the dynamic in traditional drug discovery is you're stuck to your compound and you invest a lot and you're rewarded if this one is going to market."

Sophia Genetics

Sophia Genetics

Another company making a lot of noise in the data driven medicine space is Sophia Genetics from Switzerland. It already works with a half-dozen UK hospitals to pool data and bring AI-driven insights to cancer diagnostics, claiming to already diagnose hundreds of patients a day.

Sophia Genetics' algorithms look for genetic variations, or mutations, that are unique to a patient compared to a reference genome. They annotate these variants as thoroughly as possible to see if a medicine has already been characterised as effective for the case, provide suggested actions for the clinician and automatically classify disorders, so the system gets smarter.

Read next: Hacking the human: the startups 3D printing living cells, editing genes and growing meat in laboratories

Sophia has its AI firmly focused on cancer. CEO and cofounder Dr. Jurgi Camblong told Techworld: "Cancer is mutations of our DNA. It tends to propagate in an uncontrolled way, so by using diagnostic sequencing with algorithms we can go deep into the data and eliminate the noise to find the signal. We have a better idea behind the real reason cancer develops and can adopt the right treatment."

Francis Crick Institute

Just this month in London the new Francis Crick Institute opened in Kings Cross. With a budget of £100 million a year and more than 1,250 scientists it is the biggest biomedical research facility in Europe.

One project proposed for the new space comes from David Jones, head of the bioinformatics group at UCL. Jones is embarking on a two year secondment where he will build on his previous success applying machine learning to genome research to not just label the genes but to "relate genes to disease."

He admits this is a difficult challenge "but that isn't a reason not to do it" and the goal is to reach a point where they can "infer which genes may well be involved with particular diseases".

Jones will use a training set from Mendelian (inherited) diseases, "where we know that specific genes when they are disrupted in particular ways create the symptoms of inherited disorders". If he can successfully generalise this approach the system could be used to perform predictive diagnosis for a whole range of diseases.


This all sounds very positive for anyone that doesn't have a vested interest in the big pharmaceutical industry.

However, the lack of due diligence performed across the tech industry recently with the disaster that is still unfolding around Silicon Valley biotech startup Theranos should give us all reason to pause when it comes to combining tech buzzwords with real, medical use cases.

The stakes are really high, both in terms of the money on the line and the breeding of false hope. There is also the risk of gene editing as the next logical step beyond gene mapping, and all of the ethical concerns that come with that particular use case.

As BenevolentTech's Jerome Pesenti said: “It's easy to talk about AI and it's easy to pretend that you're doing something. I think AI is real. The big labs are doing big interesting things, we're doing interesting things, but the key will be in the proof point.”

What we do know is that machines are better equipped to cut through the vast data sets the mapped genome has created than scientists are. The potential to then apply this technology and insight to diagnostics and drug discovery, and actually get it into doctor's hands, is the next unknown.

Find your next job with techworld jobs