(7f) Auto-ASR: A Web Server Providing Automated High-Throughput Ancestral Sequence Reconstruction for Combinatorial Protein Library Design

Proteins present nearly limitless potential as a sandbox in which biochemistry can be invented. However, predictive discovery of precise amino acid sequences of functional proteins can be prohibitively difficult. Ancestral sequence reconstruction (ASR) is a computational method that can aid the discovery of useful proteins by simulating the evolutionary history of a family of related or homologous modern proteins. Recent studies have shown that predicting ancestral protein sequences using maximum likelihood and Bayesian algorithms can yield improved thermostability with diverse function; these are highly desirable properties for protein engineering campaigns pursuing improved or novel function. Unfortunately, many of the existing ASR tools are either too simplistic to yield meaningful results or tortuous to navigate for new users. In this talk, we introduce AutoASR, a software which automates the ASR process in a way that does not preclude high-quality results, or trade quality for speed. By reducing the demand on an investigator’s time, many protein families of interest can be explored, and the most promising candidates selected for further utility and characterization. A particularly novel feature of this software is the design of combinatorial protein libraries based on high quality predictions of ancestral sequences and the posterior probabilities of amino acids at individual positions. This approach has been shown to yield a much higher portion of stable, functional proteins compared to random mutations. To test the functionality of this software, we aim to analyze 100 proteins with very different functions, including membrane proteins, signaling proteins, and binding proteins. The results provided by AutoASR will give insight into which types of proteins are amenable to further exploration through ASR; the best dozen candidate families produced by AutoASR will be compared to a higher-quality, manual ASR to evaluate the possibility of using AutoASR results as a direct sequence prediction tool. The software presented in this paper will provide easy access to the power and insight of ASR.