Evolutionary accumulation modeling in AMR: machine learning to infer and predict evolutionary dynamics of multi-drug resistance

Abstract

Can we understand and predict the evolutionary pathways by which bacteria acquire multi-drug resistance (MDR)? These questions have substantial potential impact in basic biology and in applied approaches to address the global health challenge of antimicrobial resistance (AMR). In this minireview, we discuss how a class of machine-learning approaches called evolutionary accumulation modeling (EvAM) may help reveal these dynamics using genetic and/or phenotypic AMR data sets, without requiring longitudinal sampling. These approaches are well-established in cancer progression and evolutionary biology but currently less used in AMR research. We discuss how EvAM can learn the evolutionary pathways by which drug resistances and other AMR features (for example, mutations driving these resistances) are acquired as pathogens evolve, predict next evolutionary steps, identify influences between AMR features, and explore differences in MDR evolution between regions, demographics, and more. We demonstrate a case study from the literature on MDR evolution in Mycobacterium tuberculosis and discuss the strengths and weaknesses of these approaches, providing links to some approaches for implementation.

Keywords: AMR; bioinformatics; drug resistance evolution; genome analysis; machine learning; multidrug resistance.

Fig 1

Example insights from EvAM for MDR evolution. These results are from HyperTraPS-CT (A–C) and HyperHMM (D) applied to MDR evolution in tuberculosis (45, 48, 51–53). (A) Data set, comprising a set of phylogenetically linked isolates with drug resistance profiles (gray, resistant; white, susceptible) to each of 10 drugs (codes at end of caption; the horizontal bar corresponds to 10 branch length units). (B) Inferred hypercubic transition network describing likely pathways of MDR evolution from an initial, fully susceptible state (top) to a fully resistant state (bottom). Each step down the network corresponds to the acquisition of resistance to another drug. The width of a line gives the probability of that step; the label describes which feature is acquired in this step and the mean and standard deviation of the timescale of this acquisition (in units of branch length from panel A. The central annotation illustrates a prediction: from a state with INH, RIF, STR resistance, the predicted most likely next step is EMB resistance. The inset shows a validation study from reference ; when such a prediction is made about the next likely step, is that step actually the next that occurs in reality? Horizontal axis gives the predicted ordering of the step that was, in reality, the next step. (C) Interactions between features. Inferred positive and negative influences: does resistance to drug X increase (blue), or decrease (red), the probability of acquiring resistance to drug Y? (D) Comparison of inferred dynamics across countries. Multidimensional scaling plot of inferred transition networks for different countries (clusters 1 and 2 are sets of countries with identical observations), illustrating the structure of between-country similarities and differences in dynamics (52). Insets give some example networks, styled as in panel B. Drug codes: INH (isoniazid); RIF (rifampicin, rifampin in the United States); PZA (pyrazinamide); EMB (ethambutol); STR (streptomycin); AMI (amikacin); CAP (capreomycin); MOX (moxifloxacin); OFL (ofloxacin); and PRO (prothionamide).