1992–2026 | Working manuscript v0.8 | submission-structure draft

Centre for Infectious Diseases and Microbiology (CIDM) / CIDM-PH / ICPMR Westmead
Author list, affiliations and corresponding author details to be finalised before submission.

Structure now prioritises Methods → Results → Discussion; figures, tables and reference governance remain versioned.

Like a person, a scientific institution develops through experience, memory, relationships, learning and correction. Its publications are not the person itself; they are traces of what it has seen, thought, tested and learned.

Target formats under consideration: Clinical Microbiology Reviews / Nature Reviews Microbiology-style review / institutional scientific review.
Status: narrative and methods spine assembled; quantitative figures, technology counts, pathogen × method matrix, and final references remain live.

Abstract

Across 1992–2026, the Centre for Infectious Diseases and Microbiology (CIDM) and CIDM-PH developed a long-lived scientific ecosystem spanning clinical microbiology, reference-laboratory practice, public-health surveillance, molecular diagnostics, medical mycology, bacteriology, virology, mycobacteriology, antimicrobial resistance, genomics, epidemiology and training. We assembled a fixed corpus of 1,947 unique publication objects and treated it as an institutional scientific memory rather than a bibliography alone. The current object-level verified evidence state comprises 265 FULL-text reviewed, 1,538 ABSTRACT-reviewed and 144 METADATA-reviewed publications. A title-derived first-pass technology census identified empirical first appearances of culture/phenotype in 1992, PCR in 1996, sequencing/genotyping in 1997, genomic analysis in 2008, MALDI-TOF in 2011 and metagenomics/microbiome work in 2018, with whole-genome/genomic signals peaking in 2021–2025. We developed a three-direction review model—critical, reverse and positive review—and distinguished within-paper fidelity (abstract–full text), between-paper fidelity (claim–reference–source evidence) and longitudinal fidelity (earlier evidence–later institutional memory). Full-text cases demonstrated the practical value of preserving denominators, validation stage, genotype–phenotype boundaries, transmission uncertainty and methodological transparency. We argue that CIDM is best understood as an evolving system of connected capabilities rather than a sequence of isolated papers, and that AI can help transform such an archive into a corrigible scientific memory when provenance, evidence depth and human responsibility remain explicit.

Keywords: clinical microbiology; infectious diseases; scientific memory; artificial intelligence; source return; genomics; public health; manuscript review; institutional learning

1. Introduction

Scientific institutions are usually described through structures, outputs and achievements: departments, grants, papers, citations, technologies and people. These views are useful, but they can miss a more fundamental property of a long-lived scientific community: it learns over time.

CIDM offers an unusually rich opportunity to examine this process because its work has remained connected to clinical microbiology, reference-laboratory practice, infectious-disease research and public health across more than three decades. The resulting publication record is therefore not only an archive of outputs. It is a longitudinal trace of changing questions, technologies, organisms, collaborations and interpretations.

Like a person, a scientific institution develops through experience, memory, relationships, learning and correction. Publications are not the institution itself; they are traces of what it has seen, thought, tested and learned. This perspective changes the purpose of retrospective review. The aim is not merely to count papers or celebrate historical milestones, but to ask what the institution has learned, what remains uncertain, where its own earlier interpretations require correction, and how accumulated experience can be returned to future work.

The AI era makes this question practical in a new way. Tasks that were previously too laborious at institutional scale—large-scale retrieval, repeated comparison, source return, provenance tracking and systematic re-review—can now be made more sustainable. The challenge is therefore whether an institution can learn to remember itself more accurately without confusing computational fluency with scientific authority.

The central proposition of this review is:

AI can help transform a publication archive into a continuously learning scientific memory, provided that evidence depth, provenance, uncertainty and human scientific responsibility remain explicit.

Figure 1 | CIDM as a living scientific institution. The institutional-life metaphor is used as an organising device rather than a biological claim.

2. Methods: corpus, evidence depth and review architecture

The analytical denominator is fixed at 1,947 unique publication objects spanning 1992–2026. Publication identity, full-text availability, binary storage and scientific review depth are treated as separate variables.

Evidence depth was coded as METADATA, ABSTRACT or FULL. FULL required substantive inspection of the attributable article body; discovery of a route or storage of a file did not itself constitute FULL scientific review.

This architecture made a second problem visible: reviewers themselves are fallible. A limitation of study design is not automatically a writing error; absence of a denominator from an abstract is not evidence that the denominator is absent from the paper; genomic relatedness is not equivalent to direct transmission; predicted resistance is not the same as phenotypically confirmed resistance. The review system therefore has to audit not only the manuscript, but also the reviewer.

Table 2 | Living iReview writing-audit summary. Counts are versioned and may change as full-text review expands.

Analyses that aggregate across the corpus therefore retain evidence depth as a separate variable. In particular, title-derived counts, abstract-informed signals and FULL-confirmed interpretations are not collapsed into one evidential category.

3. Results: what the 35-year corpus reveals

Review proceeded in three directions: critical review identified possible problems or overstatement; reverse review tested whether the reviewer or AI criticism was itself justified; positive review identified traceable practices worth preserving and teaching.

A useful paper-level synthesis therefore asks four questions:

Figure 2 | Three-direction review. Critical, reverse and positive review form one learning cycle.

Figure 3 | Current writing-issue profile among substantive-screen records. Counts are multi-label review signals, not counts of confirmed author errors.

What was done well? What could be improved? What did the reviewer initially get wrong? What should future authors learn?

Key quantitative findings

• Corpus: 1,947 unique publication objects across 35 publication years (1992–2026).

• Evidence depth: 265 object-level verified FULL, 1,538 ABSTRACT and 144 METADATA.

• Technology chronology (title-derived): culture/phenotype 1992; PCR 1996; sequencing/genotyping 1997; genomic analysis 2008; MALDI-TOF 2011; metagenomics/microbiome 2018.

• Peak title-signal windows: PCR/qPCR 2006–2010; sequencing/genotyping 2015–2019; metagenomics/microbiome 2019–2023; whole-genome/genomics 2021–2025.

• Writing audit: among 963 substantive-screen records, 98 papers carried at least one material review signal; 21 carried at least two; 5 were adjudicated as concrete internal inconsistencies.

• Teaching examples are preferentially drawn from FULL-reviewed papers with lawful public full text so readers can inspect the evidence directly.

3.1 Representation fidelity and three-direction review

Repeated source return showed that quality depends not only on the correctness of an observation, but also on how faithfully it is represented as it moves through scientific communication.

Abstract–full-text consistency is therefore treated as a quality dimension in its own right. Full text may add expected detail without contradicting the abstract. The problem arises when the reasonable interpretation changes materially: when an abstract overstates certainty, suppresses an important boundary, contains an internal count inconsistency, or leads readers toward a conclusion that the body later narrows substantially.

The same principle applies across papers. A bibliographically correct citation is necessary but not sufficient for evidentially precise scientific writing. References are not decorative metadata at the end of a manuscript; they are transmission channels through which scientific memory moves across time.

Figure 4 | Three levels of representation fidelity: within-paper, between-paper and longitudinal.

3.2 From publication archive to scientific memory

A publication archive answers a limited question: what was published? A scientific memory should support harder questions. Who worked on what? Which organisms and methods recur? Which claims are strongly supported? Which findings were later revised? Where did methods change? Which conclusions depended on narrow populations? Where are the contradictions and UNKNOWNs?

When the record is structured this way, the publication corpus becomes a substrate for institutional learning. Past work can be compared with later evidence; recurring problems can be converted into teaching rules; positive practices can be surfaced; and previously invisible connections between methods, organisms, people and public-health questions become inspectable.

The aim is not to automate judgement, but to make comparison, source return and correction scalable enough that human attention can concentrate on consequential scientific questions.

3.3 From pathogens to platforms: 35 years of scientific evolution

The CIDM record can be read along two intersecting axes: what biological problems were being studied, and what technologies made those problems newly observable. Across 1992–2026, the portfolio spans bacteria, mycobacteria, fungi, respiratory and other viruses, parasites and vector-borne infections, antimicrobial-resistant organisms, and emerging infectious diseases.

The early record was grounded in conventional clinical microbiology: culture, organism identification, antimicrobial susceptibility testing, serology and phenotypic characterisation. These methods remain scientifically important because later molecular and genomic observations still require microbiological and clinical anchoring.

PCR and sequence-based methods increased the speed and specificity with which organisms could be detected and distinguished. Conventional PCR, multiplex assays and real-time PCR progressively shifted parts of diagnostic microbiology from growth-dependent detection toward direct molecular recognition. Sanger sequencing added species identification, molecular taxonomy and phylogenetic comparison.

MALDI-TOF mass spectrometry represented a different type of change by transforming routine laboratory identification through rapid proteomic pattern recognition. Whole-genome sequencing then changed the scale of inference again: isolates could be placed within population structures, outbreaks, resistance lineages and transmission hypotheses. Genomics connected laboratory microbiology more directly with epidemiology and public-health surveillance.

Metagenomic and broader computational approaches reduced dependence on a single pre-specified organism or assay. More recently, AI has begun to function not primarily as another diagnostic platform, but as a layer for retrieval, comparison, interpretation and knowledge integration across increasingly complex datasets.

The trajectory is better described as cumulative integration than technological replacement.

Culture and phenotype → molecular diagnostics → sequencing and typing → proteomic identification → whole-genome epidemiology → metagenomic/computational integration → AI-assisted scientific memory

Earlier approximate technology dates are best treated as provisional until replaced by reproducible corpus-derived evidence. This manuscript now includes a title-derived first-pass chronology across all 1,947 publication objects and labels it explicitly as TITLE/METADATA evidence rather than silently treating it as final prevalence.

Figure 5 | Conceptual technology trajectory. Final first-appearance years and peak periods will be recalculated empirically from the fixed 1,947-object corpus.

Table 3 | Provisional technology evolution. *First-appearance years are historical working estimates and will be recalculated from the fixed 1,947-object corpus.

3.4 Empirical Microbe × Methods census

To replace earlier approximate technology timelines with a reproducible first-pass analysis, we classified all 1,947 publication titles using pre-specified, multi-label pathogen/domain and method keyword families. This analysis is intentionally conservative: it is a TITLE/METADATA census, not a claim that titles capture every organism or method used in the underlying papers. Its value is that the rule can be rerun identically across the fixed corpus and progressively strengthened later with abstract and full-text evidence.

The earliest title-level signals were culture/phenotype in 1992, PCR in 1996, sequencing/genotyping in 1997, genome-wide/genomic analysis in 2008, MALDI-TOF in 2011, and metagenomics/microbiome work in 2018. The highest rolling five-year title-signal windows occurred in 2006–2010 for PCR/qPCR, 2011–2015 for MALDI-TOF, 2015–2019 for sequencing/genotyping, 2019–2023 for metagenomics/microbiome, and 2021–2025 for whole-genome/genomic work. Because these are title signals, the dates should be interpreted as empirically observed first appearances in publication titles rather than definitive dates of first technical use within CIDM.

Table 3A | Empirical technology title-signal census across the fixed 1,947-paper corpus. Multi-label TITLE/METADATA signals; not final method prevalence.

Figure 6 | Microbe × Methods title-signal matrix across the fixed 1,947-paper corpus. Counts are multi-label TITLE/METADATA signals.

Figure 7 | Method title signals across four publication eras. The pattern shows changing emphasis rather than technological replacement.

Figure 8 | Publication output across 35 publication years in the fixed 1,947-object corpus.

The cross-tabulation makes the cumulative nature of CIDM's scientific development visible. Bacterial titles intersected broadly with culture/phenotype, PCR, sequencing and genomics; mycobacterial/TB titles showed particularly strong sequencing/genotyping and genomic signals; fungal work remained linked to phenotype, PCR and sequencing; and outbreak/surveillance titles increasingly intersected with genomic methods. This pattern supports the interpretation that later methods expanded the resolution and scale of earlier microbiology rather than displacing it.

3.5 Evidence depth: separating signal from confirmation

The 1,947-paper iReview census also permits the Microbe × Methods analysis to be interpreted through evidence depth rather than as a single undifferentiated map. At the current verified census, 258 publication objects have FULL-level review evidence, 1,544 have ABSTRACT-level review evidence, and 145 remain at METADATA level. These categories describe what was actually inspected, not what may be available elsewhere.

A higher historical aggregate state of 273 FULL is preserved in the reconciliation ledger, but eight pre-M009 promotions remain aggregate-supported without safely recovered Object_ID-level lineage. The manuscript therefore uses 265 as the paper-facing FULL denominator until those eight identities are closed; no later residual-recovery run has promoted additional papers to scientific FULL.

To create a review-informed overlay, publication titles were combined with each active paper-specific boundary/limitation statement from the iReview census and matched against the same transparent pathogen/domain and method keyword families used in the title-only analysis. This is stronger than title screening alone because it incorporates information recovered during scientific review, while remaining weaker than complete method coding from the full body of every paper.

The resulting matrix should therefore be read in two layers. The main cell value is the number of publications in which the pathogen/domain and method signals were recoverable from the title plus paper-specific review text; the bracketed F value shows how many of those signals occur in the FULL-reviewed subset. This makes evidential asymmetry visible rather than hiding it behind a single count.

Table 3B | Evidence depth in the 1,947-paper iReview census. Evidence depth is distinct from full-text availability.

Figure 10 | Object-level verified evidence depth across the complete 1,947-paper census: 265 FULL, 1,538 ABSTRACT and 144 METADATA.

Figure 11 | Review-informed Microbe × Methods matrix. Main value = title + paper-specific review-boundary signal; [F] = object-level verified FULL subset (n=265).

The FULL subset is especially useful as a validation layer. For example, FULL-reviewed bacterial papers show recoverable signals spanning culture/phenotype, sequencing and genomics, while FULL-reviewed outbreak/surveillance papers are particularly enriched for genomic signals. The comparison does not imply that papers outside the FULL layer lack those methods; it shows where current evidence is strong enough to support a deeper interpretation and where future full-text coding would most improve precision.

3.6 CIDM as an integrated scientific ecosystem

CIDM’s longevity is difficult to explain through publication output alone. Its more distinctive feature is the sustained coexistence of functions that are often separated institutionally: clinical microbiology, reference-laboratory practice, public-health surveillance, infectious-disease research, molecular diagnostics, genomics and training.

This integration matters because each function constrains and strengthens the others. Clinical laboratories keep research connected to real diagnostic problems. Reference-laboratory responsibilities create long-term continuity in organism characterisation, surveillance and quality assurance. Public-health work expands the scale from individual patients to populations and outbreaks. Research introduces new methods and hypotheses. Genomics increases resolution across all of these domains. Training allows accumulated expertise to move between generations rather than remain with individual investigators.

The result is not simply a set of parallel programmes, but an ecosystem in which methods, people and questions repeatedly cross institutional boundaries. A diagnostic assay may begin as a laboratory problem, become a research study, enter reference-laboratory practice, generate surveillance data, and later be reinterpreted using genomic or computational methods.

A deduplicated union of current CIDM/CIDMLS and CIDM-PH professional email rosters identifies approximately 168 email-evidenced identities. This is an indicative communications-roster count rather than an FTE establishment, but it demonstrates that CIDM is better understood as a broad operational and research community than as a small academic group.

CIDM’s scientific identity is therefore better understood as an evolving system of connected capabilities than as a sequence of individual publications.

3.7 International context: reference, not ranking

International comparison is most useful as context rather than ranking. A limited number of centres combine long-term clinical microbiology, reference-laboratory functions, public-health surveillance, translational research and pathogen genomics within connected systems. Examples include the Doherty Institute/VIDRL/MDU-PHL, Erasmus MC, Statens Serum Institut, Institut Pasteur, and national reference-laboratory systems such as RKI/RIVM.

CIDM does not need to be described as the largest or the best to establish its value. Its distinctiveness lies in the long-term integration of clinical, reference-laboratory, public-health, research, genomic and training functions within one evolving Australian ecosystem.

Table 4 | International reference landscape. Reference, not ranking; workforce scopes differ and are not directly comparable.

*CIDM scale = 157 CIDM/CIDMLS email-roster identities + 37 CIDM-PH canonical identities − 26 overlapping people = 168 unique identities; communications-roster count, not FTE.

3.8 Full-text cases as teaching examples

The final article should not rely only on aggregate counts. Wherever possible, representative cases should come from papers with lawful public full text that has been directly reviewed, with direct links that allow readers to inspect the source themselves.

Diagnostic systematic review and meta-analysis (CIDM-WORK-1772)
Positive exemplar: protocol/search transparency, QUADAS-2 appraisal, random-effects pooling, subgroup denominators and visible residual heterogeneity.
Open article

Candida tropicalis genomic and phenotypic susceptibility study (CIDM-WORK-1439)
Positive exemplar: WGS interpreted alongside phenotypic susceptibility and explicit sampling/genotype–phenotype boundaries.
Open article

NSW routine Mycobacterium tuberculosis sequencing study (CIDM-WORK-1644)
Bounded positive exemplar: explicit 1,831-patient denominator, routine sequencing workflow and visible inference limits.
Open article

WGS-guided MDR/XDR-TB management systematic review (CIDM-WORK-1502)
Bounded positive exemplar: evidence synthesis with heterogeneity and limited direct outcome comparisons retained.
Open article

Large-scale Salmonella Typhimurium AMR genomics (CIDM-WORK-1668)
Bounded positive exemplar: very large genome-scale analysis while preserving predicted-resistance versus phenotype distinction.
Open article

One Health E. coli genomic analysis (CIDM-WORK-1704)
Bounded positive exemplar: 5,471 genomes across human, animal, food and environmental sources while retaining sampling and transmission boundaries.
Open article

These examples are intentionally not presented as awards for perfect papers. Their purpose is to show concrete, inspectable practices that can be reused, challenged and improved.

Table 5 | Full-text teaching cases prioritised for the final manuscript.

3.9 Abstract–full-text consistency

Repeated full-text upgrading showed that abstract and full text should be assessed as a representation pair. Full text frequently adds denominators, subgroup structure, model assumptions, validation boundaries and uncertainty. Such additions are normal when they deepen but do not reverse the abstract. A problem arises when omission materially changes the interpretation of the work.

Table 6 | Seed abstract–full-text consistency cases.

This distinction is important for AI-assisted review. A system that simply flags missing abstract details will generate false criticism. Evidence depth must therefore be part of reviewer confidence.

3.10 References as the architecture of institutional memory

Scientific citations do more than identify prior work: they transmit claims about what evidence shows. During that transmission, population identity, task, construct, design, direction of association, uncertainty and level of inference can be compressed or lost even when a reference is genuine and topically relevant.

A bibliographically correct citation is therefore necessary but not sufficient for evidentially precise scientific writing. The relevant auditable unit is the Claim <-> Reference <-> Source Evidence relationship. (Jergas and Baethge, 2015; Pavlovic et al., 2021; Sarol et al., 2024).

For an institutional history, this matters especially because later papers are part of the mechanism by which earlier evidence is remembered. If later claims overstate or distort earlier sources, citation accumulation does not necessarily become scientific memory. Source return therefore provides a practical method for testing whether evidence survives its reuse across generations.

4. Discussion: AI and the next stage of scientific memory

This work would have been difficult to sustain at institutional scale using conventional manual review alone. AI contributed not by replacing expert judgement, but by making large-scale retrieval, comparison, reconciliation and repeated correction more feasible.

The useful unit of partnership is therefore not 'human versus AI' but a corrigible workflow in which both remain accountable to evidence and reality. Human researchers contribute purpose, responsibility, contextual judgement and willingness to revise. AI contributes scalable retrieval, comparison, traceability, connection and repeated checking. This caution is consistent with concerns that AI can increase productivity while also creating illusions of understanding (Messeri and Crockett, 2024).

The opportunity of AI is not merely to produce scientific text faster. It is to help humans express more precisely what the evidence allows them to say, what remains uncertain, and why.

Past science → structured memory → AI-assisted learning → better research → better manuscripts → better diagnostics → better public health → new science

Figure 9 | From publication archive to continuously learning scientific memory.

Table 8 | Ten manuscript-writing lessons derived from the living audit.

4.1 Interpretation

The CIDM record suggests that the scientific value of a long-lived institution lies partly in continuity. New technologies matter, but they acquire meaning through accumulated microbiological knowledge, patient context, reference-laboratory practice, public-health responsibility and training.

The same continuity creates a responsibility to review how knowledge is represented. Within-paper fidelity, between-paper fidelity and longitudinal fidelity are not independent problems. Together they determine whether an institutional archive functions as a reliable memory or merely as a collection of documents.

The three-direction review model also changes the culture of quality improvement. Critical review without reverse review risks reviewer overreach. Reverse review without positive review risks becoming purely defensive. Positive review without critical review risks celebration without learning. The three become most useful when they are treated as one learning cycle.

The principal limitation of the current framework is that the evidence depth remains heterogeneous. Full-text review is expanding but is not yet complete for every publication, and historical full-text access is uneven. Technology first-appearance years, pathogen × method counts and some longitudinal comparisons therefore remain live analyses rather than final claims. These uncertainties should remain visible rather than being silently converted into precision.

5. Limitations and future directions

The principal limitation is heterogeneous evidence depth: 265 of 1,947 objects are currently object-level verified FULL. Title-derived Microbe × Methods counts are reproducible but conservative, and review-informed overlays remain partly abstract-based. Technology first-appearance dates therefore represent first recoverable publication signals rather than definitive dates of first use. The next phase is to expand lawful full-text recovery, deepen method coding from article bodies, reconcile the eight historical aggregate FULL lineage holds, and continue versioned correction rather than freezing uncertain numbers.

The next phase is to complete the Microbe × Methods Observatory, quantify the 1992–2026 transitions directly from the fixed analytical corpus, continue lawful full-text recovery, expand positive and reverse review exemplars, and connect paper-level evidence to a living website and versioned scientific-memory layer. The goal is not a final immutable archive, but a corrigible system in which new evidence can update interpretation without erasing historical versions.

6. Conclusion

CIDM’s 35-year record is more than a publication count. It is the visible trace of an evolving scientific organism: one that has accumulated skills, relationships and experience; changed the tools with which it observes infectious disease; made errors and corrections; trained new generations; and repeatedly returned knowledge to clinical and public-health practice.

In the AI era, the next step is not simply to digitise that history. It is to make the history inspectable, comparable, correctable and reusable. A scientific institution that can remember itself more accurately may also be better able to decide what to ask, what to preserve, what to challenge and what to do next.

Data supplement (working): CIDM_Microbe_Methods_Observatory_v0.1_TitleDerived_2026-10-08.xlsx. This workbook preserves the title-derived keyword rules, method census, pathogen/domain census, Microbe × Methods matrix and period counts used for Figures 6–8.

Additional working data supplement: CIDM_Microbe_Methods_EvidenceDepthOverlay_v0.2_Verified265_2026-10-08.xlsx. This preserves the ALL, FULL, ABSTRACT and METADATA Microbe × Methods matrices used to generate Figures 10–11.

References

1. Jergas H, Baethge C. Quotation accuracy in medical journal articles—a systematic review and meta-analysis. PeerJ. 2015;3:e1364. doi:10.7717/peerj.1364. [Link]

2. Pavlovic V, Weissgerber T, Stanisavljevic D, et al. How accurate are citations of frequently cited papers in biomedical literature? Clin Sci (Lond). 2021;135:671–681. doi:10.1042/CS20201573. [Link]

3. Sarol MJ, Ming S, Radhakrishna S, et al. Assessing citation integrity in biomedical publications: corpus annotation and NLP models. Bioinformatics. 2024;40:btae420. doi:10.1093/bioinformatics/btae420. [Link]

4. Messeri L, Crockett MJ. Artificial intelligence and illusions of understanding in scientific research. Nature. 2024;627:49–58. doi:10.1038/s41586-024-07146-0. [Link]

5. Filippidis P, El Khoury C, Cruciani M, et al. Diagnostic performance of panfungal PCR on tissue specimens for the diagnosis of invasive fungal diseases: a systematic review and meta-analysis of the Fungal PCR Initiative (FPCRI). J Clin Microbiol. 2026;64:e00721-26. doi:10.1128/jcm.00721-26. [Link]

6. Keighley C, Gall M, van Hal SJ, et al. Whole Genome Sequencing Shows Genetic Diversity, as Well as Clonal Complex and Gene Polymorphisms Associated with Fluconazole Non-Susceptible Isolates of Candida tropicalis. J Fungi (Basel). 2022;8:896. doi:10.3390/jof8090896. [Link]

7. Zhang X, Lam C, et al. Genomic characteristics of prospectively sequenced Mycobacterium tuberculosis from respiratory and non-respiratory sources. iScience. 2024;27:110327. doi:10.1016/j.isci.2024.110327. [Link]

8. Hazra D, Lam C, Chawla K, et al. Impact of Whole-Genome Sequencing of Mycobacterium tuberculosis on Treatment Outcomes for MDR-TB/XDR-TB: A Systematic Review. Pharmaceutics. 2023;15:2782. doi:10.3390/pharmaceutics15122782. [Link]

9. Kaur S, Partridge SR, et al. Genomic dissection of the antimicrobial resistance epidemiology of Salmonella Typhimurium. Front Microbiol. 2025;16:1683995. doi:10.3389/fmicb.2025.1683995. [Link]

10. Watt AE, Cummins ML, Donato CM, et al. Parameters for one health genomic surveillance of Escherichia coli from Australia. Nat Commun. 2025;16:17. doi:10.1038/s41467-024-55103-2. [Link]

11. Akerman A, Fichter C, Milogiannakis V, et al. Cross-sectional and longitudinal genotype to phenotype surveillance of SARS-CoV-2 variants over the first four years of the COVID-19 pandemic. eBioMedicine. 2024;110:105415. doi:10.1016/j.ebiom.2024.105415. [Link]

12. Australian Government / NSW Health. Centre for Infectious Diseases and Microbiology – Public Health (CIDM-PH). Institutional information page. [Link]

13. Westmead Institute for Medical Research. Centre for Infectious Diseases and Microbiology. Institutional information page. [Link]

Reference governance note: final submission should source-return every manuscript claim that materially depends on an external source. Bibliographic correctness is necessary but not sufficient for evidential conformance.