With AI-IDP, brainQr Therapeutics has introduced a new deep learning approach that predicts the structure of disordered proteins directly from their amino acid sequence. The work by Anton Abyzov and Prof. Dr. Markus Zweckstetter has been published as a preprint on bioRxiv.
Proteins without a fixed shape
Intrinsically disordered proteins have no fixed three-dimensional structure; instead, they constantly switch between many shapes. They make up almost a third of all human proteins, control key processes in the cell and play an important role in neurodegenerative diseases and cancer. Examples include tau and α-synuclein, which clump together in Alzheimer’s and Parkinson’s disease.
Precisely because they lack a fixed shape, these proteins have so far been very difficult to predict reliably. Even AI models such as AlphaFold2, which transformed structure prediction for folded proteins, often wrongly depict them as ordered.
How AI-IDP works
AI-IDP breaks the protein chain down into short, overlapping segments. For each segment, a deep learning model predicts the preferred local structures; the segments are then reassembled flexibly, taking physical constraints into account. The result is not a single structure but a whole ensemble of possible shapes that reflects the protein’s flexibility.
The results agree well with experimental measurements, including for tau and α-synuclein. The method also captures the effects of single mutations, for example in variants of the protein TDP-43 that are linked to the neurological disease ALS. Applied to more than 3,000 disordered protein regions, AI-IDP also shows that short-lived structural elements are widespread in disordered proteins and have been finely tuned over the course of evolution.
A foundation for new drugs
For brainQr, AI-IDP is more than a research tool: the platform is intended to make it possible to target the changing shapes of disordered proteins with drugs. The company’s pipeline builds on this, from the tau stabiliser BQ-141 for Alzheimer’s disease to programmes in ALS.
Read the preprint on bioRxiv (opens in a new window)