Amyloid fibrils form and precipitate in more than 50 incurable human diseases, including Alzheimer’s (AD) and Parkinson’s disease. Only a small number of human proteins and protein variants are known to form amyloids, limiting our ability to understand, predict and engineer amyloid aggregation from sequence. The initial events in amyloid formation are particularly challenging to study by classic biophysical methods, due to the transient and high-energy nature of transition states, but they are also the most critical steps to understand, as these events control the rate of the aggregation reaction and can be targeted to prevent or slow down amyloid formation for therapeutic purposes. 

We have developed a multiplex assay of variant effects (MAVE) approach that is able to quantify the rate of aggregation of thousands of protein sequences in parallel. We have shown that this MAVE accurately classifies insertions, deletions, and missense variants in Amyloid Beta (Aß), the protein which is mutated in familial forms of AD and which aggregates in all forms of AD. In this talk I will show how, by employing the same approach on different amyloids, we find that mutational effects in one single amyloid are not enough to predict mutational impact in another amyloid forming sequence. This provides a rationale for generating MAVEs for all human amyloids, an approach which will greatly impact clinical variant classification. 

Finally, I will show how, by expanding our strategy to cover hundreds of thousands of random sequences, it is now possible to train an interpretable model that is able to predict the ability of any sequence to form amyloids, demonstrating how the power of massive experimental random sequence-space exploration is now opening the way for a systematic and global understanding of amyloid formation.

If you would like to attend the seminar, please register here.