Keith Dufault-Thompson
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Keith Dufault-Thompson has a total of 41 co-authors. Here are the top co-authors by number of shared publications:
- Xiaofang Jiang
Intramural Research Program, National Library of Medicine, National Institutes of Health
9 shared publications Anisa Habib
Department of Computer Science, Kahlert School of Computing, University of Utah, Salt Lake City , UT, 84112
3 shared publications- Brantley Hall
Department of Cell Biology and Molecular Genetics, University of Maryland, College Park, College Park, MD, USA
3 shared publications - Gabriela Arp
Department of Cell Biology and Molecular Genetics, University of Maryland, College Park, College Park, MD, USA
3 shared publications - Maggie R. Grant
Department of Cell Biology and Molecular Genetics, University of Maryland, College Park, College Park, MD, USA
3 shared publications - W. Zac Stephens
Department of Pathology, School of Medicine, University of Utah , Salt Lake City, UT, 84132
3 shared publications
Latest Publications
- 27 June 2026
- Published by Oxford University Press (OUP) in NAR Genomics and Bioinformatics
- Vol. 8  (3)
LAMBDA: a prophage detection benchmark for genomic language models
Transformer-based genomic sequence models represent an emerging frontier in computational biology. Yet, their embeddings have not yet shown the same level of predictive power as natural and protein language models, highlighting a gap between current implementations and theoretical promise. Existing benchmarks for DNA language models primarily focus on classifying regulatory elements in eukaryotic genomes, leaving open the fundamental question of whether these models learn sequence-level features across whole genomes. We introduce LAMBDA, a benchmark designed to rigorously evaluate genome language model embeddings through phage–bacteria sequence discrimination across four categories of increasing complexity: probing tasks, fine-tuning assessments, diagnostic tests, and genome-wide prophage detection. Our comprehensive analysis of current genomic language models provides insight into the importance of training data selection relative to model size, the need for domain-specific training, and the capabilities and limitations of genomic language models for detecting prophage sequences. This benchmark represents a challenging genomic annotation task in the bacterial domain and addresses a key computational problem with direct relevance to microbiology and medicine.
- 27 June 2026
- Published by Oxford University Press (OUP) in NAR Genomics and Bioinformatics
- Vol. 8  (3)
LAMBDA: a prophage detection benchmark for genomic language models
Transformer-based genomic sequence models represent an emerging frontier in computational biology. Yet, their embeddings have not yet shown the same level of predictive power as natural and protein language models, highlighting a gap between current implementations and theoretical promise. Existing benchmarks for DNA language models primarily focus on classifying regulatory elements in eukaryotic genomes, leaving open the fundamental question of whether these models learn sequence-level features across whole genomes. We introduce LAMBDA, a benchmark designed to rigorously evaluate genome language model embeddings through phage–bacteria sequence discrimination across four categories of increasing complexity: probing tasks, fine-tuning assessments, diagnostic tests, and genome-wide prophage detection. Our comprehensive analysis of current genomic language models provides insight into the importance of training data selection relative to model size, the need for domain-specific training, and the capabilities and limitations of genomic language models for detecting prophage sequences. This benchmark represents a challenging genomic annotation task in the bacterial domain and addresses a key computational problem with direct relevance to microbiology and medicine.
- Open Access
Discovery of BilV reveals a multienzymatic basis for bilirubin reduction across vertebrate gut microbiomes
Gut bacteria reduce bilirubin to urobilinogen, allowing it to be excreted through feces and urine, but studies have long noted a heterogeneous mixture of partially reduced bilirubin-derived intermediates, suggesting that multiple enzymes are involved. Here we identify bilirubin vinyl reductase (BilV), a novel Old Yellow Enzyme family reductase encoded in the genomic neighborhood of the known bilirubin reductase ( bilR ). Using heterologous expression and LC-MS/MS, we show that BilR acts on the methine bridges in the bilirubin reduction pathway; co-expression with BilV enables vinyl-group reduction and complete conversion to urobilinogen. In bacterial genomes, bilV co-occurs primarily with the bilR -insertion subtype and is largely absent alongside bilR -short. Analysis of 1,197 gut metagenomes across 14 vertebrate species reveals that this differential co-occurrence shapes pathway availability across hosts: carnivores and omnivores carry balanced bilR and bilV , whereas avian microbiomes, dominated by bilR -short, are depleted for bilV . These findings establish that bilirubin reduction to urobilinogen involves two enzymes with complementary regioselectivity, and that their distribution across vertebrate gut microbiomes varies in concert with host bile pigment chemistry.
Open AccessDiscovery of BilV reveals a multienzymatic basis for bilirubin reduction across vertebrate gut microbiomes
Gut bacteria reduce bilirubin to urobilinogen, allowing it to be excreted through feces and urine, but studies have long noted a heterogeneous mixture of partially reduced bilirubin-derived intermediates, suggesting that multiple enzymes are involved. Here we identify bilirubin vinyl reductase (BilV), a novel Old Yellow Enzyme family reductase encoded in the genomic neighborhood of the known bilirubin reductase ( bilR ). Using heterologous expression and LC-MS/MS, we show that BilR acts on the methine bridges in the bilirubin reduction pathway; co-expression with BilV enables vinyl-group reduction and complete conversion to urobilinogen. In bacterial genomes, bilV co-occurs primarily with the bilR -insertion subtype and is largely absent alongside bilR -short. Analysis of 1,197 gut metagenomes across 14 vertebrate species reveals that this differential co-occurrence shapes pathway availability across hosts: carnivores and omnivores carry balanced bilR and bilV , whereas avian microbiomes, dominated by bilR -short, are depleted for bilV . These findings establish that bilirubin reduction to urobilinogen involves two enzymes with complementary regioselectivity, and that their distribution across vertebrate gut microbiomes varies in concert with host bile pigment chemistry.
- Open Access
LAMBDA: A Prophage Detection Benchmark for Genomic Language Models
Transformer-based genomic sequence models represent an emerging frontier in computational biology. Yet, their embeddings have not yet shown the same level of predictive power as natural and protein language models, indicating a gap between current implementations and theoretical promise. Existing benchmarks for DNA language models primarily focus on classifying regulatory elements in eukaryotic genomes, leaving open the fundamental question of whether these models learn sequence-level features across whole genomes. We introduce LAMBDA, a benchmark designed to rigorously evaluate genome language model embeddings through phage-bacteria sequence discrimination across four categories of increasing complexity: probing tasks, fine-tuning assessments, diagnostic tests, and genome-wide prophage detection. Our comprehensive analysis of current genomic language models provides novel insights into the importance of training data quality relative to model size, the need for domain-specific training, and the application of genomic language models for detecting prophage sequences. This benchmark represents a challenging genomic annotation task in the bacterial domain and addresses a key computational problem with direct relevance to microbiology and medicine.
Open AccessLAMBDA: A Prophage Detection Benchmark for Genomic Language Models
Transformer-based genomic sequence models represent an emerging frontier in computational biology. Yet, their embeddings have not yet shown the same level of predictive power as natural and protein language models, indicating a gap between current implementations and theoretical promise. Existing benchmarks for DNA language models primarily focus on classifying regulatory elements in eukaryotic genomes, leaving open the fundamental question of whether these models learn sequence-level features across whole genomes. We introduce LAMBDA, a benchmark designed to rigorously evaluate genome language model embeddings through phage-bacteria sequence discrimination across four categories of increasing complexity: probing tasks, fine-tuning assessments, diagnostic tests, and genome-wide prophage detection. Our comprehensive analysis of current genomic language models provides novel insights into the importance of training data quality relative to model size, the need for domain-specific training, and the application of genomic language models for detecting prophage sequences. This benchmark represents a challenging genomic annotation task in the bacterial domain and addresses a key computational problem with direct relevance to microbiology and medicine.
- Open Access
- 18 August 2025
- Published by Oxford University Press (OUP) in Bioinformatics
- Vol. 41  (9)
The impact of tokenizer selection in genomic language models
Motivation: Genomic language models have recently emerged as a new method to decode, interpret, and generate genetic sequences. Existing genomic language models have utilized various tokenization methods, including character tokenization, overlapping and nonoverlapping k-mer tokenization, and byte-pair encoding, a method widely used in natural language models. Genomic sequences differ from natural language because of their low character variability, complex and overlapping features, and inconsistent directionality. These features make subword tokenization in genomic language models significantly different from both traditional language models and protein language models. Results: This study explores the impact of tokenization in genomic language models by evaluating their downstream performance on 44 classification fine-tuning tasks. We also perform a direct comparison of byte pair encoding and character tokenization in Mamba, a state-space model. Our results indicate that character tokenization outperforms subword tokenization methods on tasks that rely on nucleotide-level resolution, such as splice site prediction and promoter detection. While byte-pair tokenization had stronger performance on the SARS-CoV-2 variant classification task, we observed limited statistically significant differences between tokenization methods on the remaining downstream tasks. Availability and implementation: Detailed results of all benchmarking experiments are available in https://github.com/leannmlindsey/DNAtokenization. Training datasets and pretrained models are available at https://huggingface.co/datasets/leannmlindsey. Datasets and processing scripts are available at doi: 10.5281/zenodo.16287401 and doi: 10.5281/zenodo.16287130.
- 18 August 2025
- Published by Oxford University Press (OUP) in Bioinformatics
- Vol. 41  (9)
Open AccessThe impact of tokenizer selection in genomic language models
Motivation: Genomic language models have recently emerged as a new method to decode, interpret, and generate genetic sequences. Existing genomic language models have utilized various tokenization methods, including character tokenization, overlapping and nonoverlapping k-mer tokenization, and byte-pair encoding, a method widely used in natural language models. Genomic sequences differ from natural language because of their low character variability, complex and overlapping features, and inconsistent directionality. These features make subword tokenization in genomic language models significantly different from both traditional language models and protein language models. Results: This study explores the impact of tokenization in genomic language models by evaluating their downstream performance on 44 classification fine-tuning tasks. We also perform a direct comparison of byte pair encoding and character tokenization in Mamba, a state-space model. Our results indicate that character tokenization outperforms subword tokenization methods on tasks that rely on nucleotide-level resolution, such as splice site prediction and promoter detection. While byte-pair tokenization had stronger performance on the SARS-CoV-2 variant classification task, we observed limited statistically significant differences between tokenization methods on the remaining downstream tasks. Availability and implementation: Detailed results of all benchmarking experiments are available in https://github.com/leannmlindsey/DNAtokenization. Training datasets and pretrained models are available at https://huggingface.co/datasets/leannmlindsey. Datasets and processing scripts are available at doi: 10.5281/zenodo.16287401 and doi: 10.5281/zenodo.16287130.
- Top Cited PapersOpen Access
- 8 July 2025
- Published by Springer Nature in Nature Communications
- Vol. 16  (1) , 6285
Identification of gut bacteria reductases that biotransform steroid hormones
The metabolism of steroid hormones by the gut microbiome is increasingly recognized as a key factor in human health; however, the specific enzymes mediating these transformations remain largely unidentified. In this study, we identify Δ4-3-ketosteroid 5β-reductase, 3β-hydroxysteroid dehydrogenase/Δ5-4 isomerase, and Δ6-3-ketosteroid reductase enzyme families encoded by common human gut bacteria. Through phylogenetic reconstruction and mutagenesis, we show that 5β-reductase evolved to specialize in converting both natural and synthetic 3-ketosteroid hormones into their 5β-reduced derivatives, while Δ6-3-ketosteroid reductase adapted to produce Δ6-reduced derivatives. We also find that the novel 3β-hydroxysteroid dehydrogenase/Δ5-4 isomerase is fused with 5β-reductase in multiple species, streamlining the conversion of pregnenolone, a 3β-hydroxy-5-ene and steroid hormone precursor, into epipregnanolone. Through metagenomic analysis, we reveal that these enzymes are prevalent in healthy populations and enriched in females compared to males. These findings lay the groundwork for mechanistic investigations into how microbial steroid metabolism modulates host hormonal physiology.
- 8 July 2025
- Published by Springer Nature in Nature Communications
- Vol. 16  (1) , 6285
Top Cited PapersOpen AccessIdentification of gut bacteria reductases that biotransform steroid hormones
The metabolism of steroid hormones by the gut microbiome is increasingly recognized as a key factor in human health; however, the specific enzymes mediating these transformations remain largely unidentified. In this study, we identify Δ4-3-ketosteroid 5β-reductase, 3β-hydroxysteroid dehydrogenase/Δ5-4 isomerase, and Δ6-3-ketosteroid reductase enzyme families encoded by common human gut bacteria. Through phylogenetic reconstruction and mutagenesis, we show that 5β-reductase evolved to specialize in converting both natural and synthetic 3-ketosteroid hormones into their 5β-reduced derivatives, while Δ6-3-ketosteroid reductase adapted to produce Δ6-reduced derivatives. We also find that the novel 3β-hydroxysteroid dehydrogenase/Δ5-4 isomerase is fused with 5β-reductase in multiple species, streamlining the conversion of pregnenolone, a 3β-hydroxy-5-ene and steroid hormone precursor, into epipregnanolone. Through metagenomic analysis, we reveal that these enzymes are prevalent in healthy populations and enriched in females compared to males. These findings lay the groundwork for mechanistic investigations into how microbial steroid metabolism modulates host hormonal physiology.
- Open Access
- 20 May 2024
- Published by Oxford University Press (OUP) in Glycobiology
- Vol. 34  (7)
The three-sided right-handed β-helix is a versatile fold for glycan interactions
Interactions between proteins and glycans are critical to various biological processes. With databases of carbohydrate-interacting proteins and increasing amounts of structural data, the three-sided right-handed β-helix (RHBH) has emerged as a significant structural fold for glycan interactions. In this review, we provide an overview of the sequence, mechanistic, and structural features that enable the RHBH to interact with glycans. The RHBH is a prevalent fold that exists in eukaryotes, prokaryotes, and viruses associated with adhesin and carbohydrate-active enzyme (CAZyme) functions. An evolutionary trajectory analysis on structurally characterized RHBH-containing proteins shows that they likely evolved from carbohydrate-binding proteins with their carbohydrate-degrading activities evolving later. By examining three polysaccharide lyase and three glycoside hydrolase structures, we provide a detailed view of the modes of glycan binding in RHBH proteins. The 3-dimensional shape of the RHBH creates an electrostatically and spatially favorable glycan binding surface that allows for extensive hydrogen bonding interactions, leading to favorable and stable glycan binding. The RHBH is observed to be an adaptable domain capable of being modified with loop insertions and charge inversions to accommodate heterogeneous and flexible glycans and diverse reaction mechanisms. Understanding this prevalent protein fold can advance our knowledge of glycan binding in biological systems and help guide the efficient design and utilization of RHBH-containing proteins in glycobiology research.
- 20 May 2024
- Published by Oxford University Press (OUP) in Glycobiology
- Vol. 34  (7)
Open AccessThe three-sided right-handed β-helix is a versatile fold for glycan interactions
Interactions between proteins and glycans are critical to various biological processes. With databases of carbohydrate-interacting proteins and increasing amounts of structural data, the three-sided right-handed β-helix (RHBH) has emerged as a significant structural fold for glycan interactions. In this review, we provide an overview of the sequence, mechanistic, and structural features that enable the RHBH to interact with glycans. The RHBH is a prevalent fold that exists in eukaryotes, prokaryotes, and viruses associated with adhesin and carbohydrate-active enzyme (CAZyme) functions. An evolutionary trajectory analysis on structurally characterized RHBH-containing proteins shows that they likely evolved from carbohydrate-binding proteins with their carbohydrate-degrading activities evolving later. By examining three polysaccharide lyase and three glycoside hydrolase structures, we provide a detailed view of the modes of glycan binding in RHBH proteins. The 3-dimensional shape of the RHBH creates an electrostatically and spatially favorable glycan binding surface that allows for extensive hydrogen bonding interactions, leading to favorable and stable glycan binding. The RHBH is observed to be an adaptable domain capable of being modified with loop insertions and charge inversions to accommodate heterogeneous and flexible glycans and diverse reaction mechanisms. Understanding this prevalent protein fold can advance our knowledge of glycan binding in biological systems and help guide the efficient design and utilization of RHBH-containing proteins in glycobiology research.
- 19 March 2024
- Published by American Society for Microbiology in mSystems
- Vol. 9  (3) , e0003624
Annotating microbial functions with ProkFunFind
Analyzing microbial genomes has become an essential part of microbiology research, giving valuable insights into the functions and evolution of microbial species. Identifying genes of interest and assigning putative annotations to those genes is a central task in genome analysis, and a plethora of tools and approaches have been developed for this task. The ProkFunFind tool was developed to bridge the gap between these various annotation approaches, providing a flexible and customizable search approach to annotate microbial functions. ProkFunFind is designed around hierarchical definitions of biological functions, where individual genes can be identified using heterogeneous search terms consisting of sequences, profile hidden Markov models, protein domains, and orthology groups. This flexible and customizable search approach allows for searches to be tailored to specific biological functions, and the search results are output in multiple formats to facilitate downstream analyses. The utility of the ProkFunFind search tool was demonstrated through its application in searching for bacterial flagella, which are complex organelles composed of multiple genes. Overall, ProkFunFind provides an accessible and flexible way to integrate multiple types of annotation and sequence data while annotating biological functions in microbial genomes. IMPORTANCE: Genome sequencing and analysis are increasingly important parts of microbiology, providing a way to predict metabolic functions, identify virulence factors, and understand the evolution of microbes. The expanded use of genome sequencing has also brought an abundance of search and annotation methods, but integrating the information from these different methods can be challenging and is often done through ad hoc approaches. To bridge the gap between different types of annotations, we developed ProkFunFind, a flexible and customizable search tool incorporating multiple search approaches and annotation types to annotate microbial functions. We demonstrated the utility of ProkFunFind by searching for gene clusters encoding flagellar genes using a combination of different annotation types and searches. Overall, ProkFunFind provides a reproducible and flexible way to identify gene clusters of interest, facilitating the meaningful analysis of new and existing microbial genomes.
- 19 March 2024
- Published by American Society for Microbiology in mSystems
- Vol. 9  (3) , e0003624
Annotating microbial functions with ProkFunFind
Analyzing microbial genomes has become an essential part of microbiology research, giving valuable insights into the functions and evolution of microbial species. Identifying genes of interest and assigning putative annotations to those genes is a central task in genome analysis, and a plethora of tools and approaches have been developed for this task. The ProkFunFind tool was developed to bridge the gap between these various annotation approaches, providing a flexible and customizable search approach to annotate microbial functions. ProkFunFind is designed around hierarchical definitions of biological functions, where individual genes can be identified using heterogeneous search terms consisting of sequences, profile hidden Markov models, protein domains, and orthology groups. This flexible and customizable search approach allows for searches to be tailored to specific biological functions, and the search results are output in multiple formats to facilitate downstream analyses. The utility of the ProkFunFind search tool was demonstrated through its application in searching for bacterial flagella, which are complex organelles composed of multiple genes. Overall, ProkFunFind provides an accessible and flexible way to integrate multiple types of annotation and sequence data while annotating biological functions in microbial genomes. IMPORTANCE: Genome sequencing and analysis are increasingly important parts of microbiology, providing a way to predict metabolic functions, identify virulence factors, and understand the evolution of microbes. The expanded use of genome sequencing has also brought an abundance of search and annotation methods, but integrating the information from these different methods can be challenging and is often done through ad hoc approaches. To bridge the gap between different types of annotations, we developed ProkFunFind, a flexible and customizable search tool incorporating multiple search approaches and annotation types to annotate microbial functions. We demonstrated the utility of ProkFunFind by searching for gene clusters encoding flagellar genes using a combination of different annotation types and searches. Overall, ProkFunFind provides a reproducible and flexible way to identify gene clusters of interest, facilitating the meaningful analysis of new and existing microbial genomes.
- Open Access
- 3 January 2024
- Published by Springer Nature in Nature Microbiology
- Vol. 9  (1) , 173-184
BilR is a gut microbial enzyme that reduces bilirubin to urobilinogen
Metabolism of haem by-products such as bilirubin by humans and their gut microbiota is essential to human health, as excess serum bilirubin can cause jaundice and even neurological damage. The bacterial enzymes that reduce bilirubin to urobilinogen, a key step in this pathway, have remained unidentified. Here we used biochemical analyses and comparative genomics to identify BilR as a gut-microbiota-derived bilirubin reductase that reduces bilirubin to urobilinogen. We delineated the BilR sequences from similar reductases through the identification of key residues critical for bilirubin reduction and found that BilR is predominantly encoded by Firmicutes species. Analysis of human gut metagenomes revealed that BilR is nearly ubiquitous in healthy adults, but prevalence is decreased in neonates and individuals with inflammatory bowel disease. This discovery sheds light on the role of the gut microbiome in bilirubin metabolism and highlights the significance of the gut–liver axis in maintaining bilirubin homeostasis.
- 3 January 2024
- Published by Springer Nature in Nature Microbiology
- Vol. 9  (1) , 173-184
Open AccessBilR is a gut microbial enzyme that reduces bilirubin to urobilinogen
Metabolism of haem by-products such as bilirubin by humans and their gut microbiota is essential to human health, as excess serum bilirubin can cause jaundice and even neurological damage. The bacterial enzymes that reduce bilirubin to urobilinogen, a key step in this pathway, have remained unidentified. Here we used biochemical analyses and comparative genomics to identify BilR as a gut-microbiota-derived bilirubin reductase that reduces bilirubin to urobilinogen. We delineated the BilR sequences from similar reductases through the identification of key residues critical for bilirubin reduction and found that BilR is predominantly encoded by Firmicutes species. Analysis of human gut metagenomes revealed that BilR is nearly ubiquitous in healthy adults, but prevalence is decreased in neonates and individuals with inflammatory bowel disease. This discovery sheds light on the role of the gut microbiome in bilirubin metabolism and highlights the significance of the gut–liver axis in maintaining bilirubin homeostasis.
- Open Access
- 2 January 2024
- Published by Oxford University Press (OUP) in GigaScience
- Vol. 13
Large-scale genomic survey with deep learning-based method reveals strain-level phage specificity determinants
Background: Phage therapy, reemerging as a promising approach to counter antimicrobial-resistant infections, relies on a comprehensive understanding of the specificity of individual phages. Yet the significant diversity within phage populations presents a considerable challenge. Currently, there is a notable lack of tools designed for large-scale characterization of phage receptor-binding proteins, which are crucial in determining the phage host range. Results: In this study, we present SpikeHunter, a deep learning method based on the ESM-2 protein language model. With SpikeHunter, we identified 231,965 diverse phage-encoded tailspike proteins, a crucial determinant of phage specificity that targets bacterial polysaccharide receptors, across 787,566 bacterial genomes from 5 virulent, antibiotic-resistant pathogens. Notably, 86.60% (143,200) of these proteins exhibited strong associations with specific bacterial polysaccharides. We discovered that phages with identical tailspike proteins can infect different bacterial species with similar polysaccharide receptors, underscoring the pivotal role of tailspike proteins in determining host range. The specificity is mainly attributed to the protein’s C-terminal domain, which strictly correlates with host specificity during domain swapping in tailspike proteins. Importantly, our dataset-driven predictions of phage–host specificity closely match the phage–host pairs observed in real-world phage therapy cases we studied. Conclusions: Our research provides a rich resource, including both the method and a database derived from a large-scale genomics survey. This substantially enhances understanding of phage specificity determinants at the strain level and offers a valuable framework for guiding phage selection in therapeutic applications.
- 2 January 2024
- Published by Oxford University Press (OUP) in GigaScience
- Vol. 13
Open AccessLarge-scale genomic survey with deep learning-based method reveals strain-level phage specificity determinants
Background: Phage therapy, reemerging as a promising approach to counter antimicrobial-resistant infections, relies on a comprehensive understanding of the specificity of individual phages. Yet the significant diversity within phage populations presents a considerable challenge. Currently, there is a notable lack of tools designed for large-scale characterization of phage receptor-binding proteins, which are crucial in determining the phage host range. Results: In this study, we present SpikeHunter, a deep learning method based on the ESM-2 protein language model. With SpikeHunter, we identified 231,965 diverse phage-encoded tailspike proteins, a crucial determinant of phage specificity that targets bacterial polysaccharide receptors, across 787,566 bacterial genomes from 5 virulent, antibiotic-resistant pathogens. Notably, 86.60% (143,200) of these proteins exhibited strong associations with specific bacterial polysaccharides. We discovered that phages with identical tailspike proteins can infect different bacterial species with similar polysaccharide receptors, underscoring the pivotal role of tailspike proteins in determining host range. The specificity is mainly attributed to the protein’s C-terminal domain, which strictly correlates with host specificity during domain swapping in tailspike proteins. Importantly, our dataset-driven predictions of phage–host specificity closely match the phage–host pairs observed in real-world phage therapy cases we studied. Conclusions: Our research provides a rich resource, including both the method and a database derived from a large-scale genomics survey. This substantially enhances understanding of phage specificity determinants at the strain level and offers a valuable framework for guiding phage selection in therapeutic applications.