A Bayesian filtering method for estimating fitness effects of nascent beneficial mutations from barcode-lineage tracking data
Author: Huan-Yu Kuo
Affiliation: University of California, San Diego, CA
Keywords: Theory & Method Development
The distribution of fitness effect of new beneficial mutations (DFE) is a fundamental quantity necessary to understand adaptive evolution. Despite its importance, direct measurements of the DFE remain challenging because they require observing many independent and isolated adaptive mutations. Recently, the DNA barcode lineage tracking (BLT) approach has been developed for this purpose, whereby ~105 neutral DNA barcodes are introduced into the genome and used as markers for tracking the frequencies of clonal lineages during evolution. When a new adaptive mutation arises, it is permanently linked to one barcode (as long as the organism reproduces asexually) and leads to an observable increase in this barcode’s frequency. The selection coefficients of many beneficial mutations can then be inferred from the frequency trajectories of rising barcode lineages with a single experiment.
Although the BLT method for estimating the DFE shows great potential, identifying lineages that carry adaptive mutations and inferring the selection coefficients of these mutations remains difficult. In particular, current methods for analyzing BLT data do not fully account for barcode lineage extinction, a situation that can occur when selection or drift is strong. In addition, these methods classify lineages based on their entire frequency trajectories, which becomes computationally prohibitive with denser temporal sampling. Here, we develop a novel Bayesian method for inferring the DFE from the BLT data that overcomes these challenges. The central concept of our method is the probability distribution on the lineage’s selection coefficient, which is updated one data point at a time. At each time step, we also estimate the global parameters, population’s mean fitness as well as the experimental noise, based on the current knowledge of the fitness of individual lineages. We validate our method using simulations and show that it successfully recovers mean-fitness trajectories and the selection coefficients of individual lineages even under relatively high genetic drift, experimental noise, and strong selection. This work paves the way for rigorously and efficiently estimating distributions of fitness effects of beneficial mutations from BLT experiments.