Q&A With Simon Arkell, Co-Founder and CEO of Synthetica Bio

Simon Arkell
Co-founder and CEO
Synthetica Bio

Simon Arkell, OLY/MBA is co-founder and CEO of Synthetica Bio, discussed the importance democratizing AI for healthcare. Synthetica Bio is building generative AI platforms designed specifically for the biopharma industry and spoke with Medical Device & Technology about this process.

(MDT:) Can you describe the work you’re doing at Synthetica Bio?

Simon Arkell: When generative AI leapt to the forefront and exploded in people’s minds about six to eight months ago with ChatGPT, it seemed like this is a technology that has an incredible potential to be a sea change in the way that problems are tackled. We put together a team that not only had expertise in biotech and pharma (specifically around oncology, genomics, precision medicion, etc), but also with using artificial intelligence.

One of the key people is our CTO who, along with his team, has put together an architecture which is a real time streaming analytics architecture. Orchestration across all of the different services on the backend is now orchestration across large language models and vector databases. We’re able to put a lot of the new technology that’s come to the forefront of the industry in a platform that orchestrates across many different, various specialized large language models and embeddings within the databases and pulls from data that is very industry specific in a very secure environment.

It could be a customer’s own data, which is proprietary and can’t just be cut and pasted into something like ChatGPT. If you do that, you’re giving that data away to the internet for free. That security is very important. To be able to combine proprietary data with publicly available data and utilize the learnings that we get, our platform can then become a specialist generative AI offering that utilizes the best of breed under the hood.

What that means is that we’re not in the business of building a large language model that is the be all and end all of the industry. We want to use other open source LLMs that are specifically trained on a big platform but also very specific ones that are trained with genomics and claims data, or anything that is going to assist with answering a question.

(MDT:) How does the program answer these questions?

Arkell: When a question is asked of the platform, our agent knows which services to pull from to get the response back. It’s not use using one large language model, it’s getting the best of from multiple sources and bringing that together in an interactive experience for the user. What we now have is a great opportunity in democratizing generative AI, which has become usable to the average human.

We’ve got the ability now to create co-pilots, which are specific AI agents that can go and solve specific problems (like writing a clinical trial protocol). There are untold numbers of co-pilots that can be built.

The democratization is obvious when you think about the consumption and usage of these co-pilots as they help people do their daily jobs as a knowledge worker. Creating those co-pilots still requires the data scientists, the python coder, and the software engineer. So, we’ve created a no-code environment called the flight deck, which is the drag-and-drop, easy-to-use builder of those co-pilots. The democratization is one layer down from those co-pilots that have been used by those teams of knowledge workers.

(MDT:) How important is the democratization of AI in healthcare?

Arkell: It’s huge. The whole premise behind democratization is to provide value to the masses so that they can be assisted and more efficient. If you improve the efficiency in the bio-pharma industry by a few percentage points, that could represent hundreds of millions of dollars. The democratization is only possible when you remove the bottleneck, which for the industry right now is that there are extremely smart developers and data scientists out there and every one is still relying on them to build huge machine-learning models that have a data set built upon five different silos of data. The data scientists are spending a lot of time just wrangling the data so that it’s properly structured and then training the machine-learning model using Python or other technical tools that the average person is not an expert in. Weeks or months later, they’re coming out with a machine-learning model that does one thing, like predicting whether a patient is at risk of progressing from stage 2 to stage 3 small cell lung cancer.

If you can get out of that bottleneck, you can give access to those knowledge workers without having to wait weeks or months and spend millions of dollars to build one model that does one thing. This does everything, all the time, with the ability for the knowledge workers to have access to that data and have a relationship with it in an interactive way. That’s what generative AI promises.

(MDT:) Are there any privacy concerns related to using patients’ data?

Arkell: We’re working on the de-identified side of the equation right now, which is for training, tuning, adding value and specificity, and intelligence of this industry to our platform. That’s all with de-identified data. We’re learning the best ways to do the embeddings, how to respond to certain queries, and how to be smart about waiting on certain answers. These are the things that can lead to hallucination on general platforms that are not specific. We’re able to do it a pretty unique way.