32-33 - LIVER: Hard to process

Liver diseases are a sorely neglected global killer – the cause of one in every 25 deaths or two million fatalities every year.

While viral infections have historically played a dominant role in liver disease, today’s stubbornly high (and rising) mortality rates reflect globally ageing populations that are living with lifestyle-related risk factors, such as obesity and alcohol.

The liver has an incredible capacity to bounce back, but after a lifetime of inflammation from such lifestyle factors, late-stage scarring (cirrhosis) may develop. Sadly, most of us living with cirrhosis are unaware that we have it until it’s too late, when we are faced with liver failure and a significant risk of liver cancer. Primary liver cancer is now one of the top three causes of cancer deaths in 46 countries.

Today, liver transplantation is the only definitive treatment for advanced liver disease. However, donor organs are in critically short supply, and approximately a third of livers that are made available for transplant are of inadequate quality and must be discarded.

In the UK, for example, around 10% of people on the transplant list either die whilst waiting for a liver, or are removed from the list due to a deterioration in their condition. In many developed economies, the waiting list statistics are worse.

Clearly, there is a very real, very urgent need for more effective liver therapies.

Chronic liver disease, however, is a silent killer and studying silent chronic human disease is arguably among one of the great challenges of modern medicine.

Those who dedicate their careers to the pursuit of developing new liver disease treatments are held back by a poor understanding of human liver disease biology as a function of nature, nurture and prolonged lives.

Researchers are over-reliant on short-lived animal models that have largely failed to predict effective therapies in humans.

So, it stands to reason that there has been a call for more human data and better AI algorithms that could accelerate research and improve the accuracy of modelling.

Unfortunately, despite the sky-rocketing popularity of the technology, AI is not a panacea and, in reality, generating greater volumes of data usually comes with a reduction in the quality of the data.

This creates the risk of an overwhelming number of noisy data correlations, which lack clear causation or relevance to human disease biology.

Effective use of AI requires the effective use of human data throughout the drug pipeline.

Currently, over 250 biotech companies claim to be leveraging AI for drug discovery, addressing a wide range of conditions from cancer to rare genetic disorders.

Ochre Bio, a liver biotech company, is no exception. With liver research sites in Europe, Asia and the US generating among the largest genomic, histological and clinical data sets of human liver disease to currently exist, AI would change the game:

It makes histological image analysis faster, cheaper and more reproducible than a team of expert histopathologists
It identifies non-linear connections in our data that simpler statistical models baulk at
It’s helping us build ‘in silico liver’ general purpose computational models that are good enough to predict which experiments will be a waste of time.
It’s fair to say that we have a close bond with AI. But we love good data even more. Effective AI needs effective data.

For target discovery that means the Ochre team is ‘deep phenotyping’ human livers across the globe: mapping genes, to cells, to histology, to blood and clinical traits. But what purpose would this serve if validating these human findings still relied on animal models further down the pipeline?

For AI-fuelled pipelines to become truly effective, human data needs to move beyond early-stage discovery into later stage validation. We need human models to test our predictions before spending a hundred million dollars on a clinical trial.

To do this, at Ochre we have not only set up validation platforms to culture liver cells and tissues from healthy and diseased human donors, but we have invested considerable resources into a world first: a ‘liver ICU’.

Here, in a New York lab facility, human livers are kept alive outside of the body on perfusion machines for five days. In our ICU, no longer is liver disease invisible – it’s right in front of our eyes.

We are directly studying human organ biology, testing if predicted responses to our therapies hold true.

We’re doing the clinical trial before the clinical trial. Equally exciting is that, with Ochre’s focus on RNA therapies, this approach allows the progression from a prediction to the generation of human data with a new therapy in days rather than years.

Our scientists are failing faster and, in the process, they are learning what types of data and AI make for the most effective therapeutic predictions.
This is a form of closed-loop learning that, interestingly, sees humans and algorithms behaving almost as collaborators.

Significantly reducing liver disease deaths is a finish line that remains all too far away for those currently on transplant waiting lists.

As a clinician and scientist with many years on the frontline of liver disease research, however, I’ve never felt more optimistic that, within my lifetime, liver transplants can become a last resort rather than a common necessity.

Twenty years ago, we witnessed an important step towards this, with the discovery of a cure for hepatitis C.

We are currently witnessing what may be another important step, the rise of incretin therapies that help with blood sugar management and to reduce obesity (an important cirrhosis risk factor).

I hold hope that a further big step over the coming years will be the bringing-to-market of therapies that reduce liver fibrosis and restore healthy liver cell populations.

While it’s difficult to predict when this will happen, it’s perhaps easier to predict how. What will likely make these next therapeutic innovations different will be the central role of human data.

Not only the type of discovery data from large-scale human biobanks, but hopefully the type of human validation data that we are now able to generate from maintaining human cells, tissues and entire organs outside of the body.

Accelerating this shift will be regulators following the lead of the FDA Modernization Act, which no longer requires animal data to support clinical trials.

With support for rich and fail-fast human validation data, there is room to be optimistic that AI can effectively contribute to liver therapies.

Hard to process

Moving with the times

Human league

Room with a view

Final analysis