As artificial intelligence becomes ever more powerful the fear of whether it will be a tool for good or evil, or even whether it will replace us, becomes increasingly important.

The Issue

Every two years a group of structural biologists (those who study the structure of large biological molecules) have a competition – the Critical Assessment of techniques for protein Structure Prediction, or CASP – to test the software they’ve developed. Their goal is to see who has developed the most accurate tool for predicting the fold of a protein from the amino acid structure and to share new developments or strategies with the wider community of researchers. At the last competition, in 2018, the Deepmind AI called AlphaFold defeated all other challengers, scoring better than the entire competition combined and far better than the second best.

The Importance of Knowing a Protein’s Structure

Proteins are biological molecules composed of amino acid subunits. They can range in size from moderate sized peptide chains (a dozen amino acids) to immense multi-subunit complexes of thousands of amino acids. They are crucial for everything from the structure of our body to the functioning of our life-sustaining chemical processes.

When the first protein structure was determined by Max Perutz and John Kendrew in the late 1950s, the scientific community was dismayed. In comparison to the elegant simplicity of the structure of DNA, determined x years earlier, it seemed ugly and chaotic. Since that time, however, scientists have discovered that protein structures have their own elegance hidden within the jumble of coils, sheets, and strands of amino acids.

False-coloured 3D structures of three proteins. A cartoon viewing mode is used to highlight structural features.

Knowing a protein’s 3-dimensional structure is important for understanding how it carries out its function and for how we might develop pharmaceuticals for biological issues that might involve that protein. Determining a protein’s fold and thereby a more complete understanding of its function is also an interesting endeavour in its own right.

What Happened?

At CASP 2018, challengers were given the amino acid sequences of 90 proteins for which the structures were known but not yet published. In the case of 43 of these proteins, only the amino acid sequence was known (in addition to the unpublished structure). That means things like function or protein classification could not be used to aid the structure determination (certain protein folds are common motifs in certain classifications of proteins). Of those 43 ‘unknown’ proteins, AlphaFold was more accurate in predicting their structure than the competition 25 times. The next best challengers won in only 3 cases.

For an insider’s look at the recent CASP challenge and detailed thoughts on the significance, I recommend a read of M. AlQuraishi’s blog article.

While AlphaFold used computational techniques that have been well-known for some time, because of its nature and the massive hardware investment behind it, its machine-learning algorithms could be trained on vastly larger data sets than ever before. This success has led some to question everything from are scientists obsolete to why can’t big pharmaceutical companies be as successful as an IT company? Especially if they have similar money and have been in the field far longer than Deepmind’s two years.

As with the increasing number of fields where AI is succeeding, the initial shock of many specialists leads to a re-evaluation of what they thought they knew both about themselves, about their field, and about human potential in this technological world (and also their job security). Eventually, however, those who are not too discourages allow their thoughts to settle and come to the realisation that this is just another, albeit very amazing and powerful, tool we’ve developed. And, if harnessed well, it can allow us to do amazing new things.

In Conclusion

For now at least, it remains humans who are the ones making the decisions based on the AI results. As long as that remains the case, artificial intelligence remains a tool and, like any tool, can be used for good or evil. We could see AI used for incredible advances that we had once thought centuries away. Disease cures, human longevity, understanding of the fundamental nature of reality, interstellar travel, we simply don’t know how powerful AI can be and what wonders we can discover and create with it. On the other hand, the potential for restructuring of our societies and sowing political and social chaos is also immense. With each game-changing technology we create comes renewed fear, renewed hope, and ever greater responsibility both on our leaders and on each and every individual.

Scroll to top