BPS2026 – Deep-learning structure elucidation from double-mutant deep mutational scanning

Deep learning has revolutionized the field of protein structure prediction. AlphaFold2, a deep neural network, vastly outperformed previous algorithms to provide near atomic-level accuracy when predicting protein structures. Despite its success, there still are limitations which prevent accurate predictions for numerous protein systems, particularly dynamic proteins, large protein complexes and proteins with disordered regions. We recently showed that sparse residue burial restraints from deep mutational scanning (DMS) can be incorporated into deep learning protein structure prediction to enhance results. DMS is a versatile, high throughput technique that systematically maps genetic variations to phenotypes. While successful, our previous method only used information from single-mutant DMS. However, it is widely accepted that double-mutant DMS data (assessing the phenotypic consequences of simultaneously mutating pairs of residues) provides significantly more structural and functional information than single-mutant DMS. Hence, we are hypothesizing that double-mutant DMS data has the potential to transform data-guided structure prediction by allowing prediction models to better capture cooperative residue interactions that impact protein structure. With the rapid generation of a growing body of double-mutant DMS data, there is an urgent need for computational tools to elucidate structure from the data. Here, we are proposing to develop and train the first such deep learning structure prediction tool guided by double-mutant DMS data. We will use NVIDIA resources to train, fine-tune, and benchmark the model. Once the developed model will likely outperform existing structure prediction tools. We will make the model freely and publicly available to support structure elucidation from a growing and enthusiastic DMS community.