Biology subjects

Ratcliff, J.

Publications and source records attributed to Ratcliff, J..

8 recordsLinked to original sources

bioRxiv2026

MicrobeRT: A Framework and Suite of Fine-tuned Language Models for Genomic Sequences

Accurate and reproducible metagenomic analysis is essential for effective infectious disease surveillance. Targeted diagnostics such as culturing, antigen assays, and multiplex PCR panels identify only predefined organisms and can fail to identify atypical or novel infections. Shotgun metagenomics offers hypothesis-free detection, but adoption is limited by workflow complexity, dependence on mutable reference databases, and high computational cost. To combat these challenges, we introduce MicrobeRT, an integrated suite of genomic language models fine-tuned to classify metagenomic sequencing reads across three tasks: (i) species level taxonomic identification, (ii) antibiotic resistance gene detection, and (iii) pathogenicity assessment. The framework builds on three pre-trained genomic language models - Nucleotide Transformer (Dalla-Torre et al 2025), DNABERT (Ji et al 2021, Zhou et al 2024), and HyenaDNA (Nguyen et al 2024) - leveraging their large-scale genomic pre-training, adaptability to downstream tasks, and, in the case of HyenaDNA, ability to capture long-range sequence dependencies. The models are fine-tuned on large, curated datasets spanning microbial, viral, fungal, and human genomes, including more than 500,000 genomes and 250,000 antibiotic resistance gene sequences. To evaluate the performance of our fine-tuned genomic language models across classification tasks, we developed a benchmarking framework that compares MicrobeRT against state-of-the-art (SOTA) bioinformatics tools. Across both simulated and empirical datasets, the models achieve strong performance while reducing computational overhead. MicrobeRT is released as open-source software, enabling further fine-tuning and reproducible integration of alternative models and classifiers. We also demonstrate seamless, drop-in integration by incorporating our fine-tuned taxonomic classification models as an optional add-on within the open-source TaxTriage pipeline (Merritt et al 2026). MicrobeRT demonstrates the feasibility of three clinically relevant multi-label analyses, offering a practical path to scalable, reproducible metagenomic diagnostics and enabling broader deployment of AI/ML-enhanced infectious disease testing.

bioinformatics↗

bioRxiv2026

Systematic exploration of predicted quaternary structures within pandemic-relevant viral proteomes

Mechanistic understanding of viral protein-protein interactions enables global health security and pandemic preparedness, yet experimental characterisation remains difficult, costly, and often restricted to specialised laboratories. Here, we describe an in silico campaign using AlphaFold2 and AlphaFold-Multimer to predict monomers and dimer structures within 2,812 viral proteomes from 23 viral families relevant to human health. We report high-confidence predicted structures of 5,279 hetero-, and 2,749 homo-dimers. Structural clustering of interfaces reduces the set fivefold to 1,598 consolidated by form and function, of which 471 (29.5%) lacked detectable similarity to experimentally determined interfaces in the PDB. These clusters yielded two notable findings: new data on vaccinology-relevant glycoproteins, and accurate prediction of the specificity and cleavage-site recognition of viral proteases, which are targets for anti-viral development. Viral monomers and high-confidence dimers are available for interactive browsing in the Pandemic Preparedness Portal of the AlphaFold Database (https://alphafold.ebi.ac.uk/), and the full dataset is available for bulk download.

bioinformatics↗

bioRxiv2026

Uncovering the multifaceted mechanism of action of a historical antimicrobial

Natural products have provided most of our modern pharmacopoeia, serving as active molecules or scaffolds for active molecules. Their use in drug development is often inspired by their traditional or historical medical use. For many decades, this discovery pipeline has focused on identifying a single molecule responsible for much of the biological activity of a raw natural product preparation (e.g. a whole-plant extract) and scoping this molecule for clinical potential. However, it is increasingly realised that historical/traditional remedies with significant biological activity may owe this activity to the combined action of multiple molecules. Concomitantly, microbiologists increasingly argue that effectively fighting antimicrobial-resistant infections will rely on combination therapies that combine multiple antimicrobials and/or adjuvant molecules. We previously reconstructed a complex historical remedy, Bald's eyesalve. Our reconstruction of this remedy had strong antibiofilm activity, which relied on the presence of multiple ingredients. Here, we report that Bald's eyesalve has multiple antibacterial effects on exemplar Gram-positive (Staphylococcus aureus) and Gram-negative (Acinetobacter baumannii) pathogens. Bald's eyesalve disrupts bacterial membrane integrity; inhibits expression of genes associated with bacterial adhesins, virulence factors and efflux pumps in both S. aureus and A. baumannii; inhibits quorum sensing in S. aureus; and causes downregulation of genes involved in de novo nucleotide biosynthesis in S. aureus. Lastly, we show that this multifaceted mechanism of action makes it difficult for S. aureus, A. baumannii, and Pseudomonas aeruginosa to evolve resistance against Bald's eyesalve. Bald's eyesalve could be used to identify a defined cocktail of natural products suitable for preclinical testing as a multi-target antibacterial preparation to which resistance may arise more slowly than current single-molecule antibiotics.

microbiology↗

bioRxiv2025

Evaluating the Effectiveness of Parameter-Efficient Fine-Tuning in Genomic Classification Tasks

Foundation models are increasingly being leveraged for biological tasks. To address the high memory requirements of fine-tuning large pre-trained language models, parameter efficient fine-tuning (PEFT) methods are also being increasingly utilized. Previous studies have shown minimal, if any, loss in performance when using PEFT on binary classification tasks. However, the impact of using PEFT on tasks with large classification spaces has not been systemically evaluated. In this work, we apply PEFT to the problem of taxonomic classification using pre-trained genomic language models as the classification backbone. We explore various training strategies--including PEFT, full fine-tuning, and partial fine-tuning--for classifying sequences at the superkingdom, phylum, and genus levels. We find that PEFT-trained models significantly underperform compared to those trained via full fine-tuning or partial fine-tuning. Additionally, we demonstrate increased performance of pretrained models over those randomly initialized.

bioinformatics↗

bioRxiv2025

TaxTriage: An Open-Source Metagenomic Sequencing Data Analysis Pipeline Enabling Putative Pathogen Detection

MotivationTaxTriage is a comprehensive pathogen identification workflow designed for both short- and long-read untargeted DNA and RNA sequencing data. Combining read classification, mapping, and de novo assembly approaches, putative pathogens are identified through comparisons to curated pathogens and abundance expectations from healthy cohort data. Flexible installation options are enabled using Nextflow (NF), including cloud deployment via NF Tower (Seqera Platform) and local installation on a variety of systems, including standalone installations without external internet access. Final analysis summaries are compiled into an Organism Discovery Report, which lists likely pathogens and supporting data, including a custom confidence score. ResultsEvaluation of published in silico, clinical, and outbreak datasets identified performance comparable to alternative cloud-based processing pipelines for expected pathogen and co-infection detection with similar sensitivity and increased specificity. To support both public health and veterinary diagnostics communities, customization options have been incorporated to enable improved performance for host species of interest. Availability and ImplementationSource code for TaxTriage is freely available at https://github.com/jhuapl-bio/taxtriage.

bioinformatics↗

bioRxiv2023

Improved Resolution of Highly Pathogenic Avian Influenza Virus Haemagglutinin Cleavage Site Using Oxford Nanopore R10 Sequencing Chemistry

Highly pathogenic avian influenza viruses continue to pose global risks to One Health, including agriculture, public, and animal health. Rapid and accurate genomic surveillance is critical for monitoring viral mutations, tracing transmission, and guiding interventions in near real-time. Oxford Nanopore sequencing holds promise for real-time influenza genotyping, but data quality from R9 chemistry has limited its adoption due to challenges resolving low-complexity regions such as the biologically critical hemagglutinin cleavage site, a homopolymer of basic amino acids that distinguish highly pathogenic strains. In this study, human and avian influenza isolates (n=45) from Cambodia were sequenced using both R9.4.1 and R10.4.1 flow cells and chemistries to evaluate performance between approaches. Overall, R10.4.1 yielded increased data output with higher average quality compared to R9.4.1, producing improved consensus sequences using a reference-based bioinformatics approach. R10.4.1 had significantly lower minor population insertion and deletion frequencies, driven by improved performance in low sequence complexity regions prone to insertion and deletion errors, such as homopolymers. Within the hemagglutinin cleavage site, R10.4.1 resolved the correct motif in 90% of genomes compared to only 60% with R9.4.1. Further examination showed reduced frameshift mutations in consensus sequences generated with R10.4.1 that could result in incorrectly classified virulence. Improved consensus genome quality from nanopore sequencing approaches, especially across biologically important low-complexity regions, is critical to reduce subjective hand-curation and will improve local and global genomic surveillance responses.

genomics↗

bioRxiv2020

Evaluation of Different PCR Assay Formats for Sensitive and Specific Detection of SARS-CoV-2 RNA

Accurate identification of individuals infected with SARS-CoV-2 is crucial for efforts to control the ongoing COVID-19 pandemic. Polymerase chain reaction (PCR)-based assays are the gold standard for detecting viral RNA in patient samples and are used extensively in clinical settings. Most currently used quantitative PCR (RT-qPCRs) rely upon real-time detection of PCR product using specialized laboratory equipment. To enable the application of PCR in resource-poor or non-specialist laboratories, we have developed and evaluated a nested PCR method for SARS-CoV-2 RNA using simple agarose gel electrophoresis for product detection. Using clinical samples tested by conventional qPCR methods and RNA transcripts of defined RNA copy number, the nested PCR based on the RdRP gene demonstrated high sensitivity and specificity for SARS-CoV-2 RNA detection in clinical samples, but showed variable and transcript length-dependent sensitivity for RNA transcripts. Samples and transcripts were further evaluated in an additional N protein real-time quantitative PCR assay. As determined by 50% endpoint detection, the sensitivities of three RT-qPCRs and nested PCR methods varied substantially depending on the transcript target with no method approaching single copy detection. Overall, these findings highlight the need for assay validation and optimization and demonstrate the inability to precisely compare viral quantification from different PCR methodologies without calibration.

microbiology↗

bioRxiv2020

Broad and strong memory CD4+ and CD8+ T cells induced by SARS-CoV-2 in UK convalescent COVID-19 patients.

COVID-19 is an ongoing global crisis in which the development of effective vaccines and therapeutics will depend critically on understanding the natural immunity to the virus, including the role of SARS-CoV-2-specific T cells. We have conducted a study of 42 patients following recovery from COVID-19, including 28 mild and 14 severe cases, comparing their T cell responses to those of 16 control donors. We assessed the immune memory of T cell responses using IFN{gamma} based assays with overlapping peptides spanning SARS-CoV-2 apart from ORF1. We found the breadth, magnitude and frequency of memory T cell responses from COVID-19 were significantly higher in severe compared to mild COVID-19 cases, and this effect was most marked in response to spike, membrane, and ORF3a proteins. Total and spike-specific T cell responses correlated with the anti-Spike, anti-Receptor Binding Domain (RBD) as well as anti-Nucleoprotein (NP) endpoint antibody titre (p<0.001, <0.001 and =0.002). We identified 39 separate peptides containing CD4+ and/or CD8+ epitopes, which strikingly included six immunodominant epitope clusters targeted by T cells in many donors, including 3 clusters in spike (recognised by 29%, 24%, 18% donors), two in the membrane protein (M, 32%, 47%) and one in the nucleoprotein (Np, 35%). CD8+ responses were further defined for their HLA restriction, including B*4001-restricted T cells showing central memory and effector memory phenotype. In mild cases, higher frequencies of multi-cytokine producing M- and NP-specific CD8+ T cells than spike-specific CD8+ T cells were observed. They furthermore showed a higher ratio of SARS-CoV-2-specific CD8+ to CD4+ T cell responses. Immunodominant epitope clusters and peptides containing T cell epitopes identified in this study will provide critical tools to study the role of virus-specific T cells in control and resolution of SARS-CoV-2 infections. The identification of T cell specificity and functionality associated with milder disease, highlights the potential importance of including non-spike proteins within future COVID-19 vaccine design.

immunology↗