Search

Recent progress in high-throughput DNA synthesis and sequencing has enabled the development of massively parallel reporter assays for strain characterization. These datasets map a large number of DNA sequences to protein expression levels, sparking increased interest in data-driven methods for sequence-to-expression modeling. Here, we highlight advances in deep learning models of protein expression and their potential for optimizing strains engineered to produce recombinant proteins. We review recent works that built highly accurate models and discuss challenges that hinder adoption by end users. There is a need to better align this technology with the constraints encountered in strain engineering, particularly the cost of acquiring large amounts of data and the requirement for interpretable models that generalize beyond the training data. Overcoming these barriers will help to incentivize academic and industrial laboratories to tap into a new era of data-centric strain engineering.

Patient X: A 67-year-old Caucasian man slips on a patch of ice. He has abrasions to his hands and has sustained significant damage to his hip. At the emergency room, he informs clinicians he takes atorvastatin, metformin, and glimepiride to treat hypertension and Type 2 Diabetes Mellitus (T2DM). X-rays reveal a fractured hip, which will require total hip replacement surgery.

Dozens of SARS-CoV-2 vaccines have been approved for public use, yet there remains a risk that the virus evolves to escape vaccine protection. This motivates the development of universal vaccines capable of protecting against current and potentially new strains of the virus. A key challenge is the lack of computational tools to design new viral proteins capable of vaccine escape, which could serve as good targets for the development of universal vaccines. Here, we designed VAE capable of generating SARS-CoV-2 spike proteins with variable immune visibility to the cell-mediated immune response. We compared our model with two simpler generative models; a random-mutator and an 11-gram language model. All three models can generate stable, structurally valid sequences, yet only the VAE model can generate low immunogenicity sequences that interpolate smoothly along the principal variance directions of known natural sequences. This model provides an effective computational tool for the generation of spike protein sequences useful for universal vaccine design. We provide its source code at https://github.com/hcgasser/SpikeVAE.