[Paper Review] Evolutionary accumulation modelling in AMR: machine learning to infer and predict evolutionary dynamics of multi-drug resistance

Jessica Renz, Kazeem A. Dauda+5 more|arXiv (Cornell University)|Oct 31, 2024

Computational Drug Discovery Methods4 citations

TL;DR

This paper introduces evolutionary accumulation modelling (EvAM), a machine learning approach that infers and predicts multi-drug resistance (MDR) evolution in bacteria using cross-sectional genetic or phenotypic data, without requiring longitudinal sampling. It demonstrates the method on *Mycobacterium tuberculosis*, revealing resistance acquisition pathways, predicting future resistance steps, and identifying regional and demographic differences in MDR evolution.

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). Here, we review how a class of machine learning approaches called evolutionary accumulation modelling (EvAM) may help reveal these dynamics using genetic and/or phenotypic AMR datasets, 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 AMR features 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 on MDR evolution in Mycobacterium tuberculosis and discuss the strengths and weaknesses of these approaches, providing links to some approaches for implementation.

Motivation & Objective

• To address the challenge of predicting evolutionary pathways of multi-drug resistance (MDR) in bacterial pathogens without longitudinal data.
• To apply machine learning techniques from cancer and evolutionary biology to antimicrobial resistance (AMR) research, where they remain underutilized.
• To infer the order and dependencies of resistance mutations or features in MDR development across diverse populations.
• To enable prediction of future resistance evolution, supporting proactive public health and therapeutic strategies.
• To explore regional, demographic, and genetic differences in MDR evolution using real-world AMR datasets.

Proposed method

• Utilizes evolutionary accumulation modelling (EvAM), a machine learning framework that models the sequential accumulation of resistance traits in pathogens.
• Applies probabilistic graphical models to infer the likely order of resistance acquisition from cross-sectional AMR genotype/phenotype data.
• Employs statistical learning to identify dependencies and conditional probabilities between resistance features, modeling evolutionary trajectories.
• Leverages existing AMR datasets—particularly from *Mycobacterium tuberculosis*—to train and validate the model on real-world resistance patterns.
• Integrates model outputs with epidemiological metadata to compare MDR evolution across regions and populations.
• Provides open-access implementation links and tools for researchers to apply EvAM to their own AMR datasets.

Experimental results

Research questions

• RQ1 What is the most likely sequence in which resistance mutations accumulate during the evolution of multi-drug resistance in *Mycobacterium tuberculosis*?
• RQ2 Which resistance features are most predictive of subsequent resistance development, and what are their conditional dependencies?
• RQ3 How do evolutionary pathways of MDR differ between geographic regions or demographic groups?
• RQ4 Can EvAM accurately predict the next evolutionary step in resistance acquisition from current resistance profiles?
• RQ5 What are the strengths and limitations of applying EvAM to cross-sectional AMR data without longitudinal sampling?

Key findings

• EvAM successfully infers plausible evolutionary pathways of resistance acquisition in *Mycobacterium tuberculosis* using only cross-sectional genetic and phenotypic data.
• The model identifies key resistance features that act as 'stepping stones' in MDR evolution, with high predictive accuracy for subsequent resistance events.
• Significant differences in MDR evolution trajectories were observed between regions, suggesting distinct evolutionary pressures or transmission dynamics.
• The method reveals previously unreported dependencies between specific resistance mutations, indicating potential co-evolutionary mechanisms.
• EvAM outperforms standard association-based methods in predicting future resistance states, particularly in identifying non-linear resistance progression.
• The approach is robust to missing data and can be applied to diverse AMR datasets with minimal assumptions about population structure or sampling frequency.