Predicting bacterial tolerance for ionizing radiation using protein annotations

Ionizing radiation (IR) is an environmental stress encountered in medical, industrial and military contexts. Exposure of bacteria to IR  causes damage across the cell, including the generation of toxic Reactive Oxygen Species (ROS). The ability of bacteria to repair IR damage, and to prevent ROS spread, varies between species, resulting in variable tolerance for IR. The IR tolerance of bacterium within complex communities is difficult to experimentally determine, since many bacteria can’t be grown in isolation. In this study we describe the construction of a model that predicts the tolerance of an input bacterium’s tolerance to IR using widely available proteome annotations. A literature search identified 29 bacteria with experimentally determined 10% survival doses (d10). Of these, 26 were used for model construction, while 3 were reserved for testing. A guided approach, using the abundance of proteins suspected to play a role in ROS tolerance, and an unguided approach, starting with all possible proteins, were both employed. Each approach was used to construct a linear model, that predicted the d10 of an input bacterium’s proteome, and a logit model, that classified an input bacterium’s proteome as tolerant or intolerant of IR. An unguided linear model (M4) that used the relative abundance of multiple protein domains was found to be the most robust predictor of d10 values from the test set and the hidden set. The proteomes of 50 bacteria from the human microbiome with unknown tolerance for IR were analyzed using M4. Several were identified as highly sensitive to IR.