Forecasting bacterial survival-success and adaptive evolution through multi-omics stress response-mapping, network analyses and machine learning

Whether a bacterial pathogen establishes an infection and/or evolves antibiotic resistance depends on successful survival while experiencing stress from for instance the host immune system and/or antibiotics. Predictions on bacterial survival and adaptive outcomes could thus have great prognostic value. However, it is unknown what information is required to enable such predictions. By developing a novel network-based analysis method, a bacterium's phenotypic and transcriptional response can be objectively quantified in temporal 3D-feature space. The resulting trajectories can be interpreted as a degree of coordination, where a focused and coordinated response predicts bacterial survival-success, and a random uncoordinated response predicts survival-failure. These predictions extend to both antibiotic resistance and in vivo infection conditions and are applicable to both Gram-positive and Gram-negative bacteria. Moreover, through experimental evolution we show that the degree of coordination is an adaptive outcome - an uncoordinated response evolves into a coordinated response when a bacterium adapts to its environment. Most surprisingly, it turns out that phenotypic and transcriptional response data, network features and genome plasticity data can be used to train a machine learning model that is able to predict which genes in the genome will adapt under nutrient or antibiotic selection. Importantly, this suggests that deterministic factors help drive adaptation and that evolution is, at least partially, predictable. This work demonstrates that with the right information predictions on bacterial short-term survival and long-term adaptive outcomes are feasible, which underscores that personalized infectious disease diagnostics and treatments are possible, and should be developed.