bioRxiv Science⌕ Search
Home/Archive/bioRxiv

bioRxiv · 10.1101/2024.05.06.592800

GPT-4 based AI agents - the new expert system for detection of antimicrobial resistance mechanisms?

Gieske, C. G.·Bressan, M.·Fiechter, F.·Hinic, V.·Mancini, S.·Nolte, O.·Egli, A.
View original source ↗Export metadata ↓

Abstract

BackgroundEUCAST recommends a two-step process for beta-lactamases in Gram-negative bacteria. Screening with minimal inhibitory concentrations (MICs) or inhibition zone diameters for potential extended spectrum beta-lactamase (ESBL), plasmid-mediated AmpC beta-lactamase, or carbapenemase production is followed by confirmatory tests. GPT-4 and its newly released customized GPT-agent may support the initial EUCAST-screening process. We aimed to validate a customized GPT-agent to identify potential resistance mechanisms. MethodsWe used 225 Gram-negative isolates. Based on phenotypic resistances against beta-lactam antibiotics, we formed four categories: "none", "ESBL", "AmpC", or "carbapenemase". We included 862 phenotypic categories. Next, we customized a GPT-agent with EUCAST-guidelines, expert rules, and EUCAST-breakpoint table (v13.1). We compared routine diagnostic outputs (reference) to (i) EUCAST-GPT-expert, (ii) medical microbiologists, and (iii) GPT-4 without customization. We determined performance as sensitivities and specificities to flag suspect resistance mechanisms. ResultsThree human readers showed concordance in 814/862 (94.4%) phenotypic categories and used in median eight words (IQR 4-11) for reasoning. Median sensitivity and specificity for ESBL, AmpC, and carbapenemase were 98%/99.1%, 96.8%/97.1%, and 95.5%/98.5%, respectively. Three independent prompting rounds of the GPT-agent showed concordance in 706/862 (81.9%) categories but used in median 158 words (IQR 140-174) for reasoning. Median sensitivity and specificity for ESBL, AmpC, and carbapenemase prediction were 95.4%/69.23%, 96.9%/86.3%, and 100%/98.8%, respectively. In the non-customized GPT-4, 169/862 (19.6%) categories could be interpreted. Of these 137/169 (81.1%) categories agreed with routine diagnostic. The non-customized GPT-4 used in median 85 words (IQR 72-105) for reasoning. ConclusionHuman experts showed higher concordance and shorter argumentations compared to GPT-agents. Human experts showed comparable median sensitivities and higher specificities compared to GPT-agents. GPT-agents showed more unspecific flagging of ESBL and AmpC, potentially, resulting in additional testing, diagnostic delays, and higher costs. GPT-4 and GPT-agents are not IVDR/FDA-approved, but validation of LLMs is critical and datasets for benchmarking are needed.

Source connections

Explore related subjects

microbiologyPreprint

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Gieske, C. G., Bressan, M., Fiechter, F., Hinic, V., Mancini, S., Nolte, O., Egli, A.. 2024-05-06. GPT-4 based AI agents - the new expert system for detection of antimicrobial resistance mechanisms?. https://doi.org/10.1101/2024.05.06.592800

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related preprints

bioRxiv2026

A conserved cysteine-histidine-glutamate metal site identifies DUF501 (Rv1025), an essential uncharacterised protein family of Mycobacterium tuberculosis, as a candidate metalloenzyme and drug target

A substantial fraction of the Mycobacterium tuberculosis proteome remains functionally uncharacterised. Rv1025, a 155-residue protein carrying the domain of unknown function DUF501 (Pfam PF04417), is essential by transposon mutagenesis and vulnerable by CRISPR interference, an attractive but neglected drug target, yet has never been functionally described. The family (4,370 proteins, no Gene Ontology term, no solved structure) is uncharacterised across all organisms and essential in three Actinobacterial genera. A Foldseek search of the AlphaFold model against complete structural databases finds no significant homolog, indicating a novel fold. The operon eno-divIC-Rv1025-ppx2 is conserved across the Actinobacteria phylum, yet AlphaFold-Multimer finds no direct complex between Rv1025 and its neighbour DivIC. Instead, conservation across 8,700 homologous sequences reveals a near-invariant Cys113-His115-Glu59 cluster forming a pocket. Holo AlphaFold3 predictions with Zn, Fe and Mn confidently place a divalent metal on this triad at 2.25-2.47 A; mutating the triad relocates the metal, and an independent backbone-geometry predictor recovers the same site, confirming specificity. The triad is universal across the family: present in all 1,472 near-complete bacterial sequences of the Pfam alignment, with no non-conservative substitution among the 2,228 sequences examined, a defining feature of bacterial DUF501 rather than a mycobacterial peculiarity. We propose that DUF501 is a metal-binding protein and candidate metalloenzyme, the first functional hypothesis for this family, whose conserved, essential metal pocket is a promising drug target. As the predictions build on a conservation-defined site within a fully computational study, they are supportive rather than proof of metal occupancy and warrant experimental validation.

microbiology↗
bioRxiv2026

Mycoplasmal endosymbionts of Trichomonas vaginalis are associated with reduced risk for Chlamydia trachomatis endometrial infection in asymptomatic, coinfected, women.

Trichomonas vaginalis is a protozoan parasite that causes trichomoniasis, the most common curable non-viral sexually transmitted infection, and Chlamydia trachomatis is a bacterial pathogen that can ascend to the upper genital tract and cause pelvic inflammatory disease, infertility, and ectopic pregnancy. T. vaginalis harbors bacterial endosymbionts, including Candidatus Malacoplasma girerdii, an obligate symbiont, and Metamycoplasma hominis, which can live freely or symbiotically. In a 16S rRNA sequencing study of the cervicovaginal microbiome of women at high risk for chlamydial infection, Ca. M. girerdii abundance was one of 13 features predicting lack of chlamydial spread to the endometrium, despite no direct association between T. vaginalis infection and reduced chlamydial ascension. Investigating the relationship between these microorganisms further, we found that T. vaginalis vaginal abundance correlated positively with chlamydial burden in women whose infection was confined to the cervix, while a nonsignificant inverse relationship was seen in women with endometrial spread. Among participants with high chlamydial burden, Ca. M. girerdii was detected exclusively in women without endometrial infection. Both endosymbionts trended toward more frequent detection, and higher abundance, in coinfected women without endometrial spread, while M. hominis abundance correlated strongly with T. vaginalis burden in this group. These findings suggest that mycoplasmal endosymbionts of T. vaginalis, rather than T. vaginalis itself, are microbial factors limiting chlamydial ascension, and point to a three-way interaction between parasite, endosymbiont, and bacterial pathogen that shapes upper genital tract C. trachomatis infection risk.

microbiology↗
bioRxiv2026

Understanding the physiological alterations of Vibrio cholerae upon exposure to L-ascorbic acid

The scourge of cholera remains a major global public health threat. It affects up to 4 million people worldwide and causes tens of thousands of deaths each year. The disease is experiencing a concerning resurgence in many parts of Africa, the Middle East, and Asia. To effectively tackle cholera and circumvent rising antimicrobial resistance, targeted biological and preventive approaches, complementing traditional rehydration, are urgently needed. In this regard, our group has demonstrated the efficacy of L-ascorbic acid in controlling the growth and pathogenesis of Vibrio cholerae in vitro. The present work further provides a mechanistic elucidation of the L-ascorbic acid-mediated physiological changes in V. cholerae and also bolsters such a non-antibiotic approach to control cholera.

microbiology↗