Personalised AI oncology role for Cambridge Cancer Genomics

This month’s AI summit in Vancouver saw Cambridge Cancer Genomics (CCG.ai) present five papers on artificial intelligence and oncology, demonstrating that its precision oncology solutions have real benefits for the treatment of cancer.

The NeurIPS 2019 conference (Neural Information Processing Systems) is one of the world’s largest AI summits. This year’s conference attracted 14,000 machine learning and AI professionals, with 1,429 papers selected for publication out of 4,854 submitted papers, an increase from last year’s 1,010 publications.

Having now arrived back in the UK, the company - highly commended in the Cambridge Independent Science & Technology Awards 2019 - is continuing its mission to transform the way cancer patients are treated, and indeed before they even begin their treatment.

With a staff count of 25, CCG.ai has made startling progress since its inception in 2016.

“It’s a small team,” CTO and co-founder Harry Clifford told the Cambridge Independent. “It’s a peer-reviewed process. What we’re working on specifically is the idea that if we can identify mutations more robustly, that helps everyone in the long run.”

CCG.ai’s platform is called OncOS, as in oncology operating system. The emerging field of precision oncology relies on the accurate pinpointing of genetic changes in the molecular make-up of a tumour, in order to provide personalised targeted treatments.

“It’s a precision AI platform for oncologists to be able to access the latest analytics for sequencing data,” says Harry. “We’re producing analytics with these papers, so that might be gene evolution, or how tumours grow over time, so the oncologist can stay ahead of the cancer, which helps because lots of cancer treatments are very reactive.

“We’re currently going through lots of approval processes. Progress is looking good globally. We’re working with people in California, London and Singapore as well as Cambridge.”

Using simple blood draws, CCG.ai shortens the time required to know whether treatment is working, buying the clinician more time to alter treatment and reduce unnecessary side effects. The technology can detect relapse earlier than standard of care, predict response to therapy and decrease ineffective treatment regimens. Over time, CCG.ai aims to better predict the best therapeutic strategy for cancer patients before their treatment even begins.

There are many potential uses for OncoS’ ability to predict tumour evolution and identify cancer-causing mutations.

“The main thing is the idea that cancer is incredibly complex,” says Harry. “We bring analytics but we want to improve testing for cancer in general, so the drug discovery sector is interested.”

To achieve better testing, all biomarkers – single-point cancerous mutations – are identified, but thereafter the picture can change.

“It’s not so simple with immunotherapies in terms of understanding how a patient will respond. All biomarkers feed in to understanding the cancer genome, which leads to better testing.”

CCG.ai is based on Hills Road with fellow life sciences avatars Prowler.ai and BIOS.

“They’re both machine-learning based so it’s a really cool place to be.”

An investment round is on the way.

“We raised money earlier this year,” says Harry. “We’re venture capital-backed, it’s a mix of a few different people, mostly based out of San Francisco.

“We’re looking for a Series A next year, probably Q1.”

CCG.ai’s NeurIPS papers outline approaches to improve the accuracy and efficiency of “variant calling” in tumour-derived genetic data:

- Paper 1 outlines a neural network that accurately calls genetic mutations in cancer and is well suited to real-world data sets

- Paper 2 presents an approach that results in safe, robust, and statistically confident somatic mutation calls for precision oncology treatment choices

- Paper 3 presents a technique that allows for compression of somatic variants without losing the predictive power of the uncompressed original

- Paper 4 represents somatic mutation data in lower dimensional representations to allow for uptake and analysis in precision oncology research

- Paper 5 outlines a novel data-driven ML method to improve prediction of tumour sub-clonal growth.

Paper Summaries: Deep Bayesian Recurrent Neural Networks for Somatic Variant Calling in Cancer