Gainous, T. B. | bioRxiv Science

bioRxiv2023

Discovery and engineering of hypercompact epigenetic modulators for durable gene activation

Programmable epigenetic modulators provide a powerful toolkit for controlling gene expression in novel therapeutic applications, but recent discovery efforts have primarily selected for potency of effect rather than contextual robustness or durability thereof. Current CRISPR-based tools are further limited by large cargo sizes that impede clinical delivery and, in gene activation contexts, by brief activity windows that preclude transient, single-dose strategies such as lipid nanoparticle (LNP) delivery. To address these limitations, we perform high-throughput screening to discover novel classes of transcriptional modulators derived from thousands of human, viral, and archaeal proteomes. We identify high-potency activators capable of mitotically stable gene activation in a multitude of cellular contexts and leverage machine learning models to rationally engineer variants with improved activities. In liver and T-cells, novel hypercompact activators (64 to 98 amino acids) derived from vIRF2 core domain (vCD) achieve superior potency and durable activation lasting weeks beyond the current large activators ([~]five-fold larger). In a humanized mouse model, we target a human hypercholesterolemia susceptibility gene and achieve activation persisting five weeks after a single dose by LNP delivery. Our discovery pipeline provides a predictive rubric for the development of contextually robust, potent, and persistent activators of compact size, broadly advancing the therapeutic potential of epigenetic gene activation.

synthetic biology↗

bioRxiv2023

Improving few-shot learning-based protein engineering with evolutionary sampling

Designing novel functional proteins remains a slow and expensive process due to a variety of protein engineering challenges; in particular, the number of protein variants that can be experimentally tested in a given assay pales in comparison to the vastness of the overall sequence space, resulting in low hit rates and expensive wet lab testing cycles. In this paper, we propose a few-shot learning approach to novel protein design that aims to accelerate the expensive wet lab testing cycle and is capable of leveraging a training dataset that is both small and skewed ({approx} 105 datapoints, < 1% positive hits). Our approach is composed of two parts: a semi-supervised transfer learning approach to generate a discrete fitness landscape for a desired protein function and a novel evolutionary Monte Carlo Markov Chain sampling algorithm to more efficiently explore the fitness landscape. We demonstrate the performance of our approach by experimentally screening predicted high fitness gene activators, resulting in a dramatically improved hit rate compared to existing methods. Our method can be easily adapted to other protein engineering and design problems, particularly where the cost associated with obtaining labeled data is significantly high. We have provided open source code for our method at https://github.com/SuperSecretBioTech/evolutionary_monte_carlo_search.

bioinformatics↗