Interpretable neural architecture search and transfer learning for understanding CRISPR–Cas9 off-target enzymatic reactions

Finely tuned enzymatic pathways control cellular processes, and their dysregulation can lead to disease. Developing predictive and interpretable models for these pathways is challenging because of the complexity of the pathways and of the cellular and genomic contexts. Here we introduce Elektrum, a deep learning framework that addresses these challenges with data-driven and biophysically interpretable models for determining the kinetics of biochemical systems. First, it uses in vitro kinetic assays to rapidly hypothesize an ensemble of high-quality kinetically interpretable neural networks (KINNs) that predict reaction rates. It then employs a transfer learning step, where the KINNs are inserted as intermediary layers into deeper convolutional neural networks, fine-tuning the predictions for reaction-dependent in vivo outcomes. We apply Elektrum to predict CRISPR–Cas9 off-target editing probabilities and demonstrate that Elektrum achieves improved performance, regularizes neural network architectures and maintains physical interpretability. Developing predictive mechanistic models in biology is challenging. Elektrum uses neural architecture search, kinetic models and transfer learning to discover CRISPR–Cas9 cleavage kinetics, achieving high performance and biophysical interpretability.