Detecting antibiotic resistance: classical, molecular, advanced bioengineering, and AI-enhanced approaches

Abstract

Antibiotic resistance continues to erode the effectiveness of modern medicine, creating an urgent demand for rapid and reliable diagnostic solutions. Conventional diagnostic approaches, including culture-based susceptibility testing, remain the clinical reference standard but are constrained by lengthy turnaround times and limited sensitivity for early detection. In recent years, significant progress has been made with molecular and spectrometry-based methods, such as PCR and next-generation sequencing, MALDI-TOF MS, Raman and FTIR spectroscopy, alongside emerging CRISPR-based platforms. Complementary innovations in biosensors, microfluidics, and artificial intelligence further expand the diagnostic landscape, enabling faster, more sensitive, and increasingly portable assays. This review examines both established and emerging technologies for detecting antibiotic resistance, outlining their respective strengths, limitations, and potential roles across diverse settings. By synthesizing current advances and highlighting future opportunities, this review emphasizes complementarities among detection strategies and their potential integration into practical diagnostic frameworks, including in resource-limited settings.

Keywords: ESKAPE; antibiotic resistance; artificial intelligence; detection methods; machine learning; multidrug resistance; nanotechnological platforms; pathogens.

Figure 1

Phenotypic, molecular, and artificial intelligence-based methods used in the detection of antibiotic resistance. Phenotypic assays include conventional techniques such as disk diffusion and dilution-based methods, as well as advanced analytical platforms including Raman spectroscopy, Matrix-Assisted Laser Desorption/Ionization Time-of-Flight (MALDI-TOF) Mass Spectrometry, Liquid Chromatography with tandem mass spectrometry (LC-MS/MS), and Fourier Transform Infrared Spectroscopy (FTIR). Molecular methods encompass Polymerase Chain Reaction (PCR)-based approaches, metagenomics, whole-genome sequencing (WGS), DNA microarrays, and CRISPR/Cas technologies. Microfluidic platforms and biosensors represent versatile approaches that can be applied in both phenotypic and molecular contexts. The outermost circle illustrates artificial intelligence models (including multilayer perceptrons, bi-directional long short-term memory, multi-branch architectures, convolutional neural networks, residual neural networks, deep neural networks, autoencoders, and support vector machines). Unlike phenotypic and molecular methods, these approaches are not stand-alone diagnostic tools but serve as computational frameworks that integrate with and enhance conventional methods. Their role is to support data interpretation, increase accuracy, and enable automation. Created in BioRender. Aldea, A. (2025, https://BioRender.com/mgjtdea).

Figure 2

AI-driven analysis of molecular and phenotypic data for AMR prediction. The figure is organized into three panels: (left) Data (Input), showing molecular and phenotypic data types used for analysis; (center) AI Models, illustrating various machine learning architectures applied to each data type; and (right) Result (Output), presenting predicted antimicrobial resistance profiles. The data shown here represent illustrative examples of typical input features and model outputs used in AMR studies. Molecular data include raw DNA sequences, which are preprocessed and analyzed using AI models, including multilayer perceptrons, to identify resistance-associated mutations, such as gyrA Q431E conferring fluoroquinolone resistance (Jamal et al., 2020). Separately, antimicrobial peptide sequences, MIC values, and bacterial genomic features can be integrated into deep learning architectures like Bi-LSTM, CNN, and multi-branch models, to predict MIC values against clinically relevant pathogens (Chung et al., 2024). Phenotypic data include microscopy images, processed using CNN or RNN to predict bacterial resistance or susceptibility phenotypes (Ikebe et al., 2024). In addition, spectral data obtained through Raman spectroscopy are analyzed using autoencoders, DNN, or SVM to distinguish resistant from susceptible isolates based on subtle biochemical signatures (Ciloglu et al., 2021). Created in BioRender. Aldea, A. (2025, https://BioRender.com/u5rqf9p).

Figure 3

Comparative overview of antibiotic resistance detection methods by culture dependency and time to result. This figure provides a schematic comparison of major categories of antibiotic resistance detection methods and their indicative timeframes. Where applicable, an upstream DNA extraction module (1–2 h) is shown, preceding nucleic-acid-based workflows (PCR, DNA microarray, WGS, metagenomics, and CRISPR/Cas). (Left) Culture-dependent methods rely on an obligatory cultivation step of 18–48 h prior to testing. These include traditional phenotypic assays, FTIR spectroscopy, MALDI-TOF MS, and whole genome sequencing, which subsequently yield results within 1–24 h. (Center) Semi-culture-independent methods, exemplified by biosensors, CRISPR/Cas, Raman spectroscopy, LC-MS/MS, PCR, and microfluidic platforms, can be implemented with or without prior enrichment depending on the protocol, providing results within 15 min–7 h. (Right) Culture-independent methods bypass growth entirely, directly analyzing molecular signatures from the sample. These include DNA microarrays and metagenomics, with typical times to result ranging from 7 to 48 h. Arrows indicate the general workflow, highlighting how diagnostic speed improves as culture requirements are reduced, from multi-day phenotypic assays to near real-time molecular or biosensor-based approaches. Created in BioRender. Aldea, A. (2025, https://BioRender.com/z40p4eu).