my first real startup. That is, the first time that I’ve actually incorporated a company. Simmunology Limited exists since last week.
The path to incorporation has been an interesting and sometimes terrifying ride. (Aside: understatement will typically be an element of my style, this should be borne in mind when reading my prose.) I finished up a postdoc in computational neuroscience at TU Berlin last December (2017). Since then I have been trying to establish a place for myself in the world in which I get to use my, not inconsiderable, experience to choose my own directions.
I left academia with a strong feeling that not-all is right in the academic world. In fact it is very, very wrong. But more importantly, I don’t think that the kind of cross-disciplinary talents which I bring to the table will ever be fully utilised in that sector, at least not as it exists today.
I came up with two pet projects which I thought I could work on to bring an impact to the world. The first, is a behavioural modelling AI system which, in honour of recent news stories, I like to call Cambridge Analytica for Social Good. This is a system which through initial expert interventions and later through automated discovery procedures is able to guide patients towards behavioural patterns which are more appropriate for their illness. I wanted to target metabolic syndrome and autoimmune illnesses. Today, the most successful interventions for these illnesses are largely behavioural, rather than medicinal. In my design, the system learns the most appropriate nudges which will influence the individual behaviour of the patient towards more desirable patterns of behaviour. The terrible name that I gave it at the time, was pathways to health. I accidentally met with a company working with cancer patients, in Berlin, and they are now committed to rolling out an implementation of my design as part of their patient-interaction system.
The second, and infinitely more ambitious, pet project is what we are now calling Simmunology. Again, I am interested in autoimmune and metabolic syndromes. These illnesses are extremely complicated and traditional drug development techniques have had very limited successes in developing drug targets for their treatment. My background in mathematical sciences has typically focused on complex adaptive systems. I see most of biology as a set of homeostatic processes maintaining the system on an attractor, which we call life. This is overstating things somewhat in favour of a particular type of maths. Life is considerably more complicated that what I am describing. But the metaphor in my mind is useful. More importantly, my metaphor is considerably more complicated than most current models of biological functioning and, indeed, drug functioning.
Today, drug development begins with a disease. Pharmacologists and medicinal chemists identify biochemical pathways which they believe may be critical in this disease. This identification is based on expert knowledge, and the process is usually pretty effective (remember the understatement disclaimer). They then work backwards towards targets which they think they can use to influence this pathway. When a target is identified, it is characterised, and the process of drug discovery can begin.
In this process, we’ve worked all the way backwards from disease to a point where the topic of discovery is typically molecules (for simplicity I will describe the traditional process for discovery of NMEs – new molecular entities – and ignore biologicals for now). The impact of computers and mathematics in the pharma pipeline is most acutely felt at this earliest stage. This is the point at which physicists, computational chemists, and a few computer scientists are allowed to have their fun. They design a molecule which will fit the target receptor and hand this design over to the medicinal chemists. A further impact of computational scientists, this time statisticians, is felt in the characterisation of pharmacokinetics/pharmacodynamics (PK/PD) which you can think of as how long the molecule takes to be broken down in the body. This is particularly important as the first step towards discovering toxicity and dosage.
Now, from the point described above until part of the way through the clinical trials there is little to no further mathematical or computational impact on the drug design process. This despite the obvious impact of computerisation on a multitude of other fields.
Despite mentioning it, my complaint is not about the lack of computerisation per-se. It’s rather about the fallible nature of the original pathway choice and the fact that you don’t really find out if you made a mistake until after you’ve developed some relatively expensive assay results (case-specific experimental results performed in a lab), or worse, until you’ve given them to an animal.
Simmunology is building a model of the human immune system. You can think of this as a simulation, in large part this is exactly what it is. We provide three basic benefits to pharma companies: 1) you can try out your lead candidate (that’s the favourite molecule) on our simulation and find out what it will do to a human subject in-silico, you even get to rewind the simulation and discover why a particular outcome occurred; 2) you speed-up your drug development, less need to cycle through multiple rounds of improving a candidate molecule and trying it out on wet-lab assays, do some of the cycling in-silico and reject other leads out-of-hand due to unsuitability. I talked with a professor from the UK today who reminded me that of 5 compounds that typically make it through to human clinical trials, 4 will fail. We’d like to stop those 4 compounds from ever making it that far through the process. We would also like to stop the adverse effects which occur in the test subjects, on the road to rejection, and the wasted animals who underwent testing in order to bring those compounds that far in the first place. Finally, 3) you get to play with the biochemical/immune pathways during the original process of choosing drug targets. Look for network effects! Today, many illnesses are not targettable. This means that no drug can uniquely target a cellular process which will ‘cure’ this illness. We will change this by enabling the impacting of multiple targets via combination therapies, and the immediate visualisation of expected outcomes if you target multiple parts of the immune network via multiple mechanisms.
This is the drug development of the future. Please, join us.