20 - 026 - Protein Folding and Deep Mind

 The possibility of creating a the Digital Human just got very much closer with the DeepMind artificial intelligence research organisation, owned by Google, managing to digitise the protein folding process. So what does this all mean?

The human body is made of real tangible things. At an organ level (eg heart, liver, brain etc) these are easily visualised and you have probably seen the equivalents in animals at the local butcher. Now imagining zooming into these using an electron microscope which has the power to make a human hair look a mile wide. You going to come across structures made up from atoms and molecules so tiny they could never be handled by us as humans but they exist in the same way that organs exist but they are just a whole lot smaller.

These are the components that make up your human body. All be it very small components. They can be viewed under an electron microscope at the lowest level as tiny spheres, tubes and random three dimensional shapes that wriggle and move around. Importantly they have structures so they maybe strips that are straight or folded or tubes that are straight or bent or blobs that are any shape. They can float independently or have differing attachments to each other. They make up a very random and surreal world and to say it is abstract would be an understatement. Now to represent this real moving world of the body digitally has become the objective of DeepMind. It’s a case of getting what you can view under an electron microscope to be created digitally. Just like digital games reflect real life situations then DeepMind maps digitally represent the working of these really tiny biological components.

You are going to zoom down from human body to a human organ to cluster of human cells to a human cell and then to the molecules and atoms within this cell. Then no doubt in the future into the atom itself. It is the clusters of human cells forming their own unique structures where many new facets of life are to be uncovered. Being real physical structures the way they move and interact within the body can be investigated. Even using Computer Aided Manufacture (CAM) vastly enlarged models can be made of these structures. These can be handled and the way they lock together with other cell structures can be assimilated to learn how things work. By applying the same techniques to humans suffering illness and disease the faulty structures can be examined to determine how the condition is causing illness. But possibly most significantly the same processes can be applied after the introduction of medical interventions (eg drugs) to determine how they change these structures. Dismantling the human body, all be it digitally, like you would dismantle a car to locate the component that is the root cause of a medical problem. Digitisation is the key to this approach.

So let us try and visualise how this digitisation is achieved. There are similarities to computer gaming and computer aided design (CAD). So think of large rectangular box that the human body can be placed inside possibly lying down flat like being in a coffin but not coffin shaped being just rectangular box shaped. Now visualise ruler markings along the edges of this rectangular box. Using three of these edges with their marking points can allow you to locate any point within this box. So the tip of your nose can have an X reading and a Y reading and a depth from the top of the box being the Z reading. The classic three co-ordinates necessary to locate anything in space. Now image the ruler markings going down to the thousandth of a hair width so you can locate a point to a very exact molecular and even lower atomic level. You can define a living cell at the end of your nose. A large number of co-ordinates can define the shape of this cell as each point on the surface of the cell are defined by X-Y-Z co-ordinates. You can even define things within the cell itself along with their location and shape. The outer cell membrane can be defined by both its inner and outer surfaces thereby defining its thickness or even variations in this thickness over its surface. Now you can zoom both in and out. Out from a cell to the part of an organ, then the organ itself and then the organs place within the human body. It is Google Maps for the human body. This covers what I term the physical mapping. Now at every one of these points the chemical makeup of the point can be defined based upon the Periodic Table so you achieve chemical mapping. This allows you to create a chemical map if you like layered over the physical like in Google maps viewing something as a map or a satellite image. The same thing viewed in a physical and a chemical dimension. This creates the fact that within the physical shape the chemical makeup may adopt a different pattern. Seeing the human body in this layered way makes investigating the body’s changes and processes a multi-dimensional experience. The physical shape might change whilst the chemical makeup does not. Whilst the chemical makeup maybe changing whilst the physical remains unchanged. One of the commonest processes, a process representing change, would be for the chemical makeup within a cell to change with this subsequently resulting in a physical change. The key medically would be to spot theses chemical changes early with possibly a drug intervention before the problem presents itself as a physical change. True preventative medicine. It is important to reiterate the fact that the physical structures are created as a digital map and the chemical makeup is recorded as a separate digital map. These are then overlaid over each other. Now this overlaying of different what I have termed dimensions is important. The physical dimension and the chemical dimension. This lends itself to the addition of other dimensions if they can be scientifically be defined by capturing data on them. So for example the electrical activity within the body or other dimensions not previously scientifically identified. These maybe considered unreal at present but subjects like radiesthesia and telepathy come to mind being what I and others (but only a few of us) term the “wireless aspects of life”. Subjects to be explored later in this book.

But what becomes really important is when you can run these digital maps over time so you can view all the changes in the human body both physically and chemically. You can watch tumours grow and drugs destroy them digitally. This leads to the important ability to re-run scenarios over and over again with this forming one of the important tenets of artificial intelligence (AI). The ability to build up digital pathways that can represent reality. Adjustments can be made to data along these pathways and the changes can be recorded. This approach can then be used to build up a whole holistic view as thousands of different pathways are simulated and stored. So intelligence on a subject is grown until it shown to accurately represent a reality.

So what you can see under a digital microscope can now be digitally represented with the human body emulating a moving computer game. These are digital 3D shapes all interacting like they would within a real living biological body. In fact what you get to see digitally is a visual representation of what you would see under a powerful digital microscope. The significance of it being digital is that it can be analysed in terms of its internal structures then by using artificial intelligence processing different scenarios can be run through thousands of times with various adjustments being made to the objects under scrutiny. Essentially you can make changes, possibly by applying a drug, into the equation and then run through a sequence to determine the impact of the drug. Like playing strategy games where decision tree style thinking is important where at each node new decisions can be made about the way forward based upon new data. These biological digital models allowing for many thousands if not millions of different scenarios to be tested leading to different final results. Once a positive result is achieved you can reverse back over the decision tree to analyse all the different node point inputs.

Whilst being really comfortable with these digital models essentially replicating what can be seen using a digital microscope it is when it comes to the underlying chemistry of the model where this adds a new very powerful scientific dimension. Think of hovering a pointer over a particular point within the digital model and then the chemical makeup is displayed at a molecular level. Essentially the elements as defined in the Periodic Table. These can be painted as colours within the digital model so you can visually see the chemical makeup and importantly changes to this make up over time. You can see things going wrong both visually and chemically at the same time. A new tumour developing with cell multiplication with these cells changing in their chemical makeup. Then hopefully the reverse effect after the application of treatments. The thought that one day you could see all this happening in your body real time on your smartphone. Possibly the most significant aspect being the very early identification of these changes so they can be treated well before the patient is aware of any symptoms.

DeepMind is applying both Gaming and Computer Aided Design (CAD) principles to biological components down at a nanometre sizes. The CAD is important because it is simple step to Computer Aided Manufacture (CAM) where the three dimensional printing of components has become a standardised industrial practice. So it is a simple step to create physical three dimensional models of these biological components. Now significantly there is a school of thought that being able to create physical models of these tiny biological components will give us a totally new insight into how our bodies are assembled from the very smallest building blocks of atoms and molecules up to cells and cellular structures up to organs. So this is true biological engineering. The creation of these tiny cells, viruses and bacterium as real models produced through Computer Aided Manufacturing (CAM) but in sizes where they can be handled which allow us to see how they interact manually. The way viruses gain entry into human cells can be witnessed through the workings of these enlarged models. It is important to appreciate that although many of the processes within a human body are chemical it is also the mechanical designs that evolve from these chemical processes that undertake purely mechanical activities. They are interlocking mechanical processes like keys opening a lock or pick axes breaking into the wall of a cell. These are real mechanical activities taking place within the human body and their assessment as such it important as a tenet of medical investigation.

 

So why is this suddenly so significant in respect of the work at DeepMind on proteins?

Life is dependent on proteins. Proteins are made up from chains of organic compounds known as amino acids. These bond together into long chains to make proteins, which in turn create different types of tissue that perform all sorts of functions within the human body. Proteins, the building blocks of life, adopt complex three-dimensional shapes which determine what they do and how they do it. They make up good cells, bad cancer cells, viruses, bacteria and antibodies. If something is alive it will be made up of proteins. Proteins are the machinery of both life and unfortunately death. Failures in the production of these proteins will lead to death. Proteins which are collections of compounds called amino acids are the machinery of life. Machinery is the right term to use because it implies they are physical, which they are, so we need to know what they look like. They cannot just be considered as a chemical compound listing the molecules that make them up since life is about movement and this is the machinery that supports this movement. It is about biological engineering and not pure chemical engineering. At the atomic level biology works by the interlocking structures of atoms in a physical way. We need to know how the one dimensional sequence of atoms we get from chemistry “folds” in a 3D image. Into a real physical component.

Protein molecules can coil around themselves in all sorts of different ways. They can be tangled in a normal way. But also they can be tangled in a way that is abnormal leading to conditions like Alzheimer’s and Parkinson’s. Tangling is a really strong force in natural for good and for bad. I think of how the Christmas tree lights or power cables to electrical appliances or water hoses or blind cords can get so horrendously tangled together. There seems to be some natural forces of nature that sit behind the tangling of rope or strip like materials particularly if they have protrusions like the lights on Christmas tree lights. Going back to the primeval soups of life it is these tangling processes that allowed the build-up of more complex life forms. Algae deposits and damp algae twisting and turning until by chance something more sophisticated is created. The complexity of sperm swimming to the egg to penetrate it to generate the start of a cell reproduction cycle. These basic building blocks of life need to be digitally analysed both physically and chemically to understand when and how life itself sparks into a continuous process of growth and decay. In the case of decay protein malformation and malfunction maybe behind many of the unpleasant conditions associated with the aging process.

 

 

Now deriving the chemical formula of a protein is scientifically relatively straightforward but determining its physical appearance was a long laborious process called crystallography. Prior to this digital break by DeepMind this was the standard way of “seeing” a protein by a process of crystallisation. That is lock it in a stasis in a repeating pattern then you fire x-rays at it in all different directions and from the image they form you can infer its shape. What DeepMind did was create some specific chemical knowledge and then apply this to those proteins where x-ray images existed from the above crystallography process. By constantly adjusting the underlying algorithm they started to get the protein chemical makeup to equal the created x-ray images. So the model was built upon crystallised protein x-ray pictures. So having trained the algorithm to link the protein amino acid structure to the corresponding specific crystallographic image it could then be applied to tens of thousands of proteins that had never been mapped using crystallography. Some of these previously unmapped proteins could then have the crystallographic process applied and this could be compared to the DeepMind image. The success rate of this DeepMind mapped image against the actual crystallographic image showed that the process was a real success. But importantly the algorithm could continue to be adjusted as new crystallographic images became available further refining its accuracy. To the point that the digital only processes were deemed as accurate as crystallography. But the digital approach had another major advantage.

Now in the body proteins are not crystallised but they are free moving structures that warp and bend continually. This makes the modelling of this real world behaviour a lot more difficult. But real success will follow on from when the digital modelling mirrors real life. The computer model equals what you can see under an electron microscope. But then significantly digital model will have captured the underlying logic to these biological processes. Once this structure is a “calculated” one based upon this logic we can then run in a “live” model emulating life within the human body. Knowing this we can go onto design the experiments that make drugs and enzymes and all the molecular tools we need to improve the human condition. These medical interventions can also be modelled real time so we can monitor how they interact with the faulty proteins. So create a sick digital human then test the best medical interventions to make it better.