A killing machine to help science
By Andrew Jack
Published: December 6 2006 02:00 | Last updated: December 6 2006 02:00
From its offices outside San Francisco, a company listed in London is testing a powerful new experimental drug on the human body to the point of death - without any of the ethical concerns that usually make this impossible.
With the help of super-computers, Entelos is developing a "virtual human" for the pharmaceuticals sector, using simulation technology and complex mathematical algorithms.
"There is no ethical reason why we can't kill a virtual human," says Thomas Paterson, Entelos co-founder and senior vice-president, who used to design aerospace simulations. "You learn from failures. We've known how to cure cancer in rats for God knows how long, but it just doesn't translate."
His approach might have been useful to the pharmaceuticals giant Pfizer, which last weekend had to announce the abandonment of its new anti-cholesterol drug torcetrapib in late-stage clinical trials. It identified toxic side-effects only months before it was set to seek regulatory approval.
In spite of considerable scepticism, Pfizer - along with Johnson & Johnson, Roche, Organon and a number of other leading pharmaceuticals companies - is already beginning to use the Entelos model in the struggle to find new, more effective and less costly ways to develop medicines.
Mr Paterson argues that there are important lessons for the pharmaceuticals sector to learn from other industries in which he has worked before. Defence and aerospace also spend large sums on research and development, but have progressed much further in testing through simulations.
"They simulate aircraft before developing costly prototypes, whereas pharmaceutical products get far further along in development in human trials before researchers realise their idea is fundamentally flawed," he says. "The A380 was extensively flight tested, and you don't hear about major problems once it went airborne.
"There are so many questions that are not adequately pressure-tested. We use the discipline of maths and the power of engineering to get a much earlier predictor."
He argues that while human logic can understand cause-and-effect relationships, it is less well suited to interpreting complex biological processes where everything affects everything else. That is where computer simulations come into play.
The main difference - and greatest challenge - is that engineering systems are designed by humans. "With flight data, we have a lot of clues about underlying mechanisms," he says. "Biology is not designed by us. We don't have the same understanding."
As a result, Entelos's medical applications work in another direction. Its experts analyse clinical trial results and design hundreds of algorithms to simulate the mechanisms and interactions reported. "We can't understand the protein-to-protein interactions, so we take a top-down approach, asking what functionally must be going on," he says.
The power of the approach is that it allows very rapid testing of any formulation or dosage over very long periods. A 30-year trial, for instance, can be run in a few days to forecast long-term problems of drug resistance.
Mr Paterson estimates that it typically takes about a year for his staff to design a computer model for a therapeutic system, with a virtual trial then taking place over a matter of hours. In practice, there is a continual process of adaptation, with clients modifying their questions as they test hypotheses.
In the process, it may be possible to skip over intermediate animal tests, moving drugs more quickly into the clinic for humans. The model can help focus questions, identify biomarkers, such as blood tests to be used in humans, and select which patients will prove most responsive.
"We have been looking at ways to get better value from the tremendous amount of information in our historical clinical trial database," says Joe Alexander, director of global clinical technology for Pfizer's human health technologies division, which began working with Entelos a year ago.
"Many approaches focus on 'data mining'," he says. "The power of Entelos was a 'clear tube' rather than a 'black box' for making predictions. We are using such models to do a better job in identifying what patients would benefit from which treatment."
Many scientists remain sceptical about the power of such simulations and how far they can truly refine or replace the existing research and development processes of new medicines.
Entelos counters that it has already helped its clients modify their research programmes, with important cost savings in the process. By modelling the response of healthy virtual patients to Johnson & Johnson's diabetes drugs, for example, it was able to suggest there would be no adverse effects from higher doses.
The result - borne out in the phase-one trials the company subsequently conducted - allowed the company to use one-third of the patients and conduct the experiments in 60 per cent of the projected time.
It is significant that Entelos has so far not developed models in some of the more complex therapy areas, such as the central nervous system, where the mechanisms are less well understood.
But Pfizer and many of its peers are in need of whatever help they can get. Simulation may at least help provide part of the solution.
Copyright The Financial Times Limited 2006
Comments