The future of disease treatment at the intersection of AI and biochemistry

By Steven Boykey Sidley

4 Dec 2023

AI could revolutionize healthcare by unlocking the mysteries of proteins to help us understand and cure diseases.

Like many people I know, I am badly undereducated in biology. Whatever I was taught in high school was instantly forgotten when I left, and I never got near it again. Except that it is in our faces daily — diet, health, diseases, cures — we are all bouncing against a subject that few of us know anything much about, which is how the body actually works. Especially at the microscopic, chemical and cellular level.

So it is with some trepidation that I write about a subject that may be as important as any other — how the confluence of artificial intelligence and biochemistry could perhaps drive the greatest leap in human health and wellbeing in history.

Problems with proteins

Let’s start with the protein, a specific class of molecule resident in our bodies doing a wide variety of stuff, all of which keeps us running smoothly. Different flavours of proteins have different names depending on their form and function, such as collagen, the scaffolding for our skin and bones; haemoglobin to carry oxygen to our blood; insulin to regulate our sugar levels; actin and myosin to power our muscles; enzymes to speed up chemical reactions in our cells. Not to mention the protein’s yeoman service in support of our immune systems and the expression of our genes. (The reader may wonder how I know this, given my opening paragraph. AI-assisted research is the answer.)

All 200 million known proteins have the distinct and critical property of folding into very specific and individual 3D shapes — entangled loops, curls, coils and pleats. And sometimes, mysteriously, they fold incorrectly, causing a horror show of miseries — Huntington’s, Parkinson’s, Alzheimer’s, cystic fibrosis, cancers, mad cow disease and more. There is a long list of human and animal suffering at the end of a misassembled protein. It has to fold perfectly into shape to interact seamlessly with the other chemistries and components of the body; otherwise, it is a dangerously loose cannon.

Trying to understand the structure of proteins has been such a gnarly and pernickety process (costing millions of dollars per protein and taking months or even years) that researchers have made painfully slow progress since the word ‘protein’ was first coined by French chemist Antoine Fourcroy in 1789.

Why do they fold into particular shapes depending on their component chemistry? How do they fold? Why do they sometimes misfold and (more importantly) can they be fixed when they do? So many proteins have been identified, and their 3D spatial dynamics are so mathematically complex, that getting to a deep understanding of their diversity, structure and function has seemed altogether too daunting a challenge.

If researchers could quickly get to the core of these critical little machines, they would have a chance of manipulating them, with the possibility of erasing a swath of terrible diseases and bolstering health and well-being for humankind.

The second act of our story occurs in the 2010s. The ancient Chinese game of Go is considered to be the most complex two-person game ever invented; its billions of different combinatorial possibilities and game strategies make those of chess look trivial. The rules of the game are simple (small black and white stones are moved around a rectangular board by two opposing players to capture territory), but that is where the simplicity ends. The great Go players, most of whom have studied nothing else from an early age, are revered as demi-gods in Japan, because their skill is considered to be both rare and inscrutable.

I do not intend to repeat the story of the victory of an AI called AlphaGo over the greatest ever Go player, Lee Sedol, in 2016; there has been much written about it. AlphaGo was created by Google’s DeepMind, a major player in today’s AI wars. The system was trained on millions and millions of games over a long period, its internal genius was to generalize the best tactics of gameplay, some of which had never been tried (or even imagined) by humans.

Read more in Daily Maverick: Deep medicine: Artificial intelligence is changing the face of healthcare, daily

However, as impressive as this feat was, it does not compare with what came next. The researchers at AlphaGo decided to try something else. They discarded all the games that they had used as input for the AI and simply outlined the rules of Go and the definition of a victory and then let the machine play against itself. Hundreds of millions of times. It did not require any external training at all.

It became the best Go player on earth in a matter of hours. Everyone, including the researchers, was astounded. Its capacity to outstrip a long history of human dedication, ingenuity and creativity in a very short time was profound, albeit for a game. DeepMind immediately looked for other hard problems to which it could be applied, particularly where they concerned human wellbeing.

One of the first candidates was protein-folding, the recalcitrant bete noire of biochemistry research. The AI system was repurposed and re-engineered and given the name AlphaFold (followed by a later upgrade called AlphaFold 2).

It was turned on in 2021. What has happened since?

Dawn of AlphaFold

Here is something to put in your pipe and smoke: this year, the system had unravelled the structure of all 200 million proteins known to science and could describe how the 3D shapes are constructed from their core elements (another set of simple molecules called amino acids).

It can predict structures with precision down to the detail of single atoms. The results are available to anyone; it is open source, a massive database called the AlphaFold Protein Structure Database, in partnership with The European Molecular Biology Laboratory’s (EMBL) European Bioinformatics Institute — the flagship laboratory for life sciences in Europe.

Anyone can test the AI-predicted structures in this database against real data. The original paper describing AlphaFold 2 is now in the top 100 most-cited research papers of the past 10 years. The database has been accessed by over 1 million users from 190 countries.

It is a very big deal for drug development and other forms of health treatment. Prof Ewan Birney, the Deputy Director general of EMBL, had this to say about it: “This will be one of the most important datasets since the mapping of the Human Genome”.

It is easy to see AI through the lens of dystopia and skepticism and it makes for facile clickbait to portray it as such. In this story, perhaps, glimmers a kind of AI to challenge all the pessimism and suspicion that darkens our view of the future. DM

Steven Boykey Sidley is a professor of practice at JBS, University of Johannesburg. His new book It’s Mine: How the Crypto Industry is Redefining Ownership is published by Maverick451 in SA and Legend Times Group in UK/EU, available now.

Categories:

Tags:

DM is much better if you are signed in...

Business Maverick

Problems with proteins

Dawn of AlphaFold