Algorithm sheds light on 'disordered' proteins once considered too difficult to study

Intrinsically disordered proteins (IDPs) do not attain a stable secondary or tertiary structure and rapidly change their conformation, making structure prediction particularly challenging. Although these proteins exhibit chaotic and "disordered" structures, they still perform essential functions.

IDPs comprise approximately 30% of the human proteome and play important functional roles in transcription, translation, and signaling. Many mutations linked to neurological diseases, including amyotrophic lateral sclerosis (ALS), are located in intrinsically disordered protein regions (IDRs).

Powerful machine-learning algorithms, including AlphaFold and RoseTTAFold, cannot provide realistic representations of these 'disordered' and 'chaotic' protein regions as a whole. This is because they have not been trained on such data and because these proteins exhibit inherent dynamic behavior, adopting a range of conformations rather than a single stable one.

Now, a team of researchers from BSRC Fleming and the Centre for Misfolding Diseases at the University of Cambridge has found an efficient way to predict the structures of a significant fraction of all human proteins that were previously considered "dark" and notoriously difficult to observe.

The team developed and used an algorithm called "AlphaFold-Metainference," which was trained on data from available protein structure databases as well as molecular dynamics simulations. The findings of the study were recently published in Nature Communications.

"AlphaFold has transformed structural biology by providing accurate predictions of protein structures. We have now shown how to extend these predictions to IDPs, which make up about a third of the human proteome and are implicated in virtually all major diseases," says Michele Vendruscolo, Professor of Biophysics at the Centre for Misfolding Diseases at the University of Cambridge.

"We were surprised to find that although AlphaFold does not accurately predict the three-dimensional structure of IDPs, it can predict the distances between amino acids with quite good accuracy. We then incorporated this information into molecular dynamics simulations, allowing us to accurately predict the three-dimensional structures these disordered proteins adopt and their motion," explains Dr. Faidon Brotzakis, a senior postdoctoral researcher in the lab of Dr. Georgios Skretas at the Institute for Bioinnovation of the Biomedical Sciences Research Center "Alexander Fleming" (BSRC Fleming) and the study's first author.

The algorithm was tested on proteins containing both disordered and non-disordered regions, including TDP-43 (associated with ALS), ataxin-3 (linked to Machado-Joseph disease), and the prion protein (implicated in Creutzfeldt-Jakob disease).

"We tested the algorithm on a total of eleven IDPs and six PDPs, but we focused particularly on proteins associated with serious diseases. In all cases, the algorithm outperformed AlphaFold in accuracy. In fact, in 80% of cases, it matched or exceeded the accuracy of molecular dynamics simulations. This demonstrates the algorithm's advantage in the structural characterization of IDPs," explains Dr. Brotzakis.

Scientists now have a faster and more accurate way to determine the structures of disordered proteins, especially in cases where experimental data are unavailable. "In the future, we can use this information to discover molecules of pharmaceutical interest that can interact strongly with these proteins and modify their dynamics. This could prevent their problematic folding into toxic forms, such as amyloid fibrils, which are observed in many neurodegenerative diseases," says Brotzakis.

The next steps are to apply the algorithm to other biomolecules, such as DNA and RNA.