The lack of a common standard for calculating scores makes comparison between studies difficult and existing bespoke methods are not applicable to the diverse array of experimental designs currently being used. Both ratio and regression analyses can incorporate corrections for wild-type performance or nonsense variants at the expense of restricting the method to protein-coding targets only. For experimental designs that sample from more than two time points to improve the resolution of changes in frequency, ratio-based scoring is insufficient so a regression-based approach has been used instead. However, while intuitive and easy to calculate, ratio-based scores are highly sensitive to sampling error when frequencies are low. This type of ratio-based scoring has been used to quantify the effect of non-coding changes in promoters as well. Two established implementations of deep mutational scanning scoring methods, Enrich and EMPIRIC, calculate variant scores based on the ratio of variant frequencies before and after selection. Existing methods are diverse in terms of their scoring function, statistical approach, and generalizability. įundamental gaps remain in our ability to use deep mutational scanning data to accurately measure the effect of each variant because practitioners lack a unifying statistical framework within which to interpret their results. Guidelines for the design of deep mutational scanning experiments have been discussed elsewhere. Those approaches enable a different set of biological inferences that we do not seek to address here. Scoring the performance of individual variants is distinct from a related class of methods that quantify tolerance for change at each position in a target protein. Analysis of the change in each variant’s frequency throughout the selection yields a score that estimates the variant’s effect. Barcoding enables accurate assessment of variable regions longer than a single sequencing read. Here, the variable region is either directly sequenced using a single-end or paired-end strategy, or a short barcode that uniquely identifies each variant in the population is sequenced instead. Next, the frequency of each variant in each time point or bin is determined by using deep sequencing to count the number of times each variant appears. Selections can be growth-based or implement physical separation of variants into bins, as in phage display or flow sorting of cells. A selection is applied for protein function or another molecular property of interest, altering the frequency of each variant according to its functional capacity. Model systems that have been used in deep mutational scanning include phage, bacteria, yeast, and cultured mammalian cells. In a deep mutational scan, a library of protein variants is first introduced into a model system. For example, deep mutational scanning has been applied to comprehensive interpretation of variants found in disease-related human genes, understanding protein evolution, and probing protein structure with many additional possibilities on the horizon. Deep mutational scanning has greatly enhanced our ability to probe the protein sequence-function relationship and has become widely used. Deep mutational scanning is a method that marries deep sequencing to selection among a large library of protein variants, measuring the functional consequences of hundreds of thousands of variants of a protein simultaneously. Exploring the relationship between sequence and function is fundamental to enhancing our understanding of biology, evolution, and genetically driven disease.