Prediction and analysis of redox-sensitive cysteines using machine learning and statistical methods

Abstract

Reactive oxygen species are produced by a number of stimuli and can lead both to irreversible intracellular damage and signaling through reversible post-translational modification. It is unclear which factors contribute to the sensitivity of cysteines to redox modification. Here, we used statistical and machine learning methods to investigate the influence of different structural and sequence features on the modifiability of cysteines. We found several strong structural predictors for redox modification. Sensitive cysteines tend to be characterized by higher exposure, a lack of secondary structure elements, and a high number of positively charged amino acids in their close environment. Our results indicate that modified cysteines tend to occur close to other post-translational modifications, such as phosphorylated serines. We used these features to create models and predict the presence of redox-modifiable cysteines in human mitochondrial complex I as well as make novel predictions regarding redox-sensitive cysteines in proteins.