We use massively parallel DNA synthesis-selection-sequencing experiments to understand the ‘energetic structures’ of proteins and RNAs and to produce data at scale for training AI models. The goal is to cost-effectively generate large training datasets to predict, understand, and engineer the biophysical properties of macromolecules from sequence. This includes stability, binding affinity (proteins, DNA, drugs), aggregation, and allostery. A notable advance has been the creation of the first comprehensive allosteric maps of proteins. We have piloted large-scale mutagenesis of human proteins and used combinatorial mutagenesis to understand protein evolution and the energetic structures of reaction transition states, using amyloid nucleation reactions as a model system. Finally, we use pre-mRNA splicing as a model system for using massive data generation and machine learning to obtain mechanistic understanding.
Our previous work focussed on fundamental questions in genetics, primarily concerned with understanding and predicting phenotypic variation in both genetically diverse and genetically identical individuals. This has included work on incomplete penetrance, inter-generational epigenetics, noise, genetic interactions (epistasis), mutation rates, and genetic prediction. My strategy has been to pick fundamental questions and then choose tractable model systems to address them. As a result, we have used diverse experimental, computational, large-scale, and quantitative approaches.