Three things I know: Future-casting 2 of 3

Last week I published an article about Artificial General Intelligence. This week I want to follow-up with my second of three attempts to predict the future. As I said, last week, this was part of a game which is commonly played in incubators when trying to draw insights from deep-tech founders. This week I want to talk about the Virtual Patient for drug development

I founded my first company over three years ago. We made no secret of our interest in using in-silico methods to build a Virtual Patient for drug development. We didn’t succeed that time but our lack of success had little to do with either technical issues or a lack of a commercialisation option, it was entirely our own fault.

The Virtual Patient

In the years since closing my company I have re-evaluated the virtual patient multiple times. For a while I sincerely doubted the commercial opportunity.

It is very hard to sell software to biologists. Pharmaceutical companies make drugs. They need enough biological and chemical insight to make a drug and then they need to demonstrate that it works. Often software lags biological insight.

This is the technical problem, which drives the sales problem, and it needs to be emphasised.

I spent 15 years in academic research, making mathematical models of biological systems. I have never seen a mathematical model truly precede the biological insight. The closest I have come to seeing math leading biology is actually in one of my own papers. The insights were there from the biological side, but we used the model to constrain the search-space before carrying out the experiments. This is a good use of mathematical modelling in biology. But many of the most famous mathematical models of biology, which are held up as key examples in the teaching of mathematical biology, were derived after the experiments had already been carried out and analysed.

Where math applied to biology excels is in (i) formalisation, and (ii) scaling.

Formalisation is the easy example, my own paper where we constrained the search-space is an example of formalisation leading to better experimental understanding.

Scaling is a harder case to describe. Here I lean towards explaining using some of the current multi-factor models in genomics. We now have enough statistical understanding of genomics that we can make predictions about weak interactions. We are far from perfect, but it is a good starting point for further biological exploration.

I think that the virtual patient approach is ready to contribute a similar level of insight to drug development.

We have formal models of metabolic, immune, circulatory, genomic, and even microbiome systems. Alone these models can only answer questions of formalisation. A good scientist is capable of reading the academic literature and deriving these answers themself.

But if we put together a panel of such models then we can constrain our search for insights and solutions considerably. What does this mean in real world language?

By having many models we can decide with much greater relevance and accuracy which particular lab experiments to pursue.

One model is not enough for this.

Sufficient scale leads to a new way of working, which opens new opportunities and cost savings.

Unlike my prediction about AGI, the Virtual Patient for drug development is both already here and further away than ever. Every pharma company insists on building models in-house. This is costly and loses the insight I am presenting here about the benefits of scale.

Technological revolution almost never comes solely through making a known process faster. Usually it also involves some fundamental change to the problem being solved. The macro-problem remains the production of drugs using cutting-edge biological insights. The tech here merely constrains the search-space far more efficiently than humans can do this.

Finally, the AGI is likely to be a recognisable technology when it emerges. The Virtual Patient will be hidden behind a software interface. The ‘patient’ is not important, it is a metaphor, the purpose of the software will be laboratory decision support.

Leave a Reply

Your email address will not be published.

This site uses Akismet to reduce spam. Learn how your comment data is processed.