PhageLysData: an evidence-aware and AI-ready dataset of phage lytic enzymes and depolymerases

David Medina-Ortiz^1,, Álvaro Olivera-Nappa¹, María Elena Lienqueo², Rafael Opazo³, and Jaime Romero^3,.

¹_{Departamento de Ingeniería en Computación, Universidad de Magallanes, Avenida Bulnes 01855, 6210427, Punta Arenas, Chile.}
²_{Centro de Biotecnología y Bioingeniería (CeBiB), Departamento de Ingeniería Química, Biotecnología y Materiales, Universidad de Chile, Av. Beauchef 851, 8370458, Santiago, Chile.}
³_{Laboratorio de Biotecnología de Alimentos, Instituto de Nutrición y Tecnología de los Alimentos (INTA), Universidad de Chile, El Líbano 5524, Santiago 7830489, Chile.}
^*_{Corresponding authors: David Medina-Ortiz and Jaime Romero.}

PhageLysData is an evidence-aware data resource for integrating, harmonizing, characterizing, and computationally enriching phage lytic proteins collected from heterogeneous public sources.

The resource organizes exact protein sequences into explicit evidence universes, preserves source-level provenance, separates non-predictive support from prediction-derived support, and provides reusable annotation, physicochemical, structural, and numerical-representation layers.

PhageLysData is a data resource and reproducible resource-construction workflow. It is not presented as a predictive model or as a definitive classifier of phage lytic proteins. Its primary purpose is to provide a harmonized, provenance-aware, evidence-aware, and computationally reusable foundation for downstream biological and machine-learning analyses.

The complete public data release is distributed separately through Zenodo. This GitHub repository contains the source-processing workflow, reproducible worked examples, and publication-facing Supplementary Information figure notebooks used to construct, inspect, reuse, and document the resource.

---

Table of contents

• Overview
• Resource at a glance
• Biological scope
• Evidence model and resource universes
• Main resource outputs
• Data availability and Zenodo
• Repository structure
• Software requirements
• Installation
• Reconstructing PhageLysData
• Worked examples
• Publication and Supplementary Information figures
• Quick start
• Reproducibility and traceability
• Citation
• License
• Contact

---

Overview

Public information on phage lytic proteins is distributed across general protein databases, specialized enzyme collections, phage genome resources, prediction-oriented repositories, and manually curated datasets. These sources differ substantially in sequence identifiers, biological terminology, annotation depth, evidence interpretation, prediction status, metadata completeness, and data structure.

PhageLysData addresses these limitations through a staged and auditable workflow that:

• retrieves and preserves source-specific records and provenance;
• standardizes protein sequences and source metadata;
• consolidates identical protein sequences through stable exact-sequence identifiers;
• separates non-predictive biological support from prediction-derived assertions;
• defines explicit resource universes and evidence Tiers;
• integrates frozen UniProtKB annotation information;
• performs homogeneous InterProScan-based functional enrichment across the Core;
• computes reusable physicochemical descriptors;
• links existing experimental PDB structures and retrievable AlphaFold DB models;
• distributes precomputed protein language model embeddings and one-hot representations;
• provides resource-level characterization tables;
• exposes reproducible worked examples for querying, visualization, clustering, classification, and candidate retrieval;
• provides publication-facing notebooks that regenerate Supplementary Information figures from frozen release or notebook outputs.

The stable identifier used throughout the resource is:

sequence_entity_id

Each sequence_entity_id corresponds to one exact normalized amino-acid sequence and is preserved across Core tables, annotation relations, physicochemical descriptors, structural assets, numerical representations, characterization outputs, and worked examples.

---

Resource at a glance

PhageLysData v1.0 integrates:

Resource component	Entities / observations
Source observations	807,366
Exact-sequence entities	759,105
Core	11,867
Prediction Extension	745,092
Context	2,146

The evidence-supported Core is further organized by evidence Tier:

Evidence Tier	Core entities
Tier 1	90
Tier 2	784
Tier 3	9,836
Tier 4	1,157

The deeply enriched Core includes:

Asset / annotation layer	Core entities
UniProtKB-mapped entities	7,347
Any InterProScan match	11,549
InterPro annotation	11,215
Pfam annotation	9,300
GO annotation	6,775
Pathway annotation	9,662
Any existing structural asset	745
Representation-eligible / complete	11,259

The public release additionally contains:

• physicochemical descriptors and sequence-level properties;
• UniProtKB relation tables;
• InterProScan, InterPro, Pfam, GO, and pathway relations;
• mapped PDB structural assets where available;
• retrievable AlphaFold DB models where available;
• 11 precomputed protein language model embedding matrices;
• one-hot sequence representations;
• resource characterization tables;
• manifests, provenance records, validation records, and SHA-256 checksums.

---

Biological scope

PhageLysData focuses on proteins associated with phage-mediated lytic activity and related host-envelope degradation.

The frozen biological ontology contains four non-exclusive target classes:

• endolysin
• depolymerase
• virion_associated_lytic_enzyme
• other_phage_envelope_lytic_enzyme

For Core-level summaries, an additional derived resource label is used:

• multiple_nonexclusive

multiple_nonexclusive is not a fifth biological target class. It identifies Core entities supported by more than one non-predictive target class and prevents those entities from being forced into an artificial single-class assignment.

These labels are harmonized resource-level representations of source evidence. They should not be interpreted as replacements for detailed mechanistic, family-level, domain-level, or evolutionary classifications.

Source-specific biological terminology and provenance remain available so that users can inspect the evidence supporting each exact-sequence entity.

---

Evidence model and resource universes

PhageLysData separates exact-sequence entities according to the evidence supporting their inclusion.

Core

The Core contains entities with at least one valid non-predictive target assertion.

The Core is the principal deeply enriched and characterized resource. Prediction-derived assertions may coexist with Core membership but do not replace or override non-predictive support.

Prediction Extension

The Prediction Extension contains prediction-only candidates whose valid target assertions are exclusively prediction-derived.

These entities are retained for provenance, candidate-space coverage, and future reuse but remain separate from the evidence-supported Core.

Context

The Context universe contains retained exact-sequence entities without a valid target assertion under the frozen resource ontology.

These records preserve source, genome, lysis-system, or other provenance and biological context.

Evidence Tiers

Evidence Tiers apply only to Core entities.

Tier	General interpretation
Tier 1	Direct experimental support under the frozen evidence rules
Tier 2	High-confidence curated or independently supported evidence
Tier 3	Annotation-supported evidence
Tier 4	Supported but ambiguous, incomplete, weakly resolved, or conflicting evidence

Prediction Extension entities are never assigned Tier 1--4.

Resource-universe membership, canonical target class, and evidence Tier are frozen before downstream enrichment and are not modified by physicochemical analysis, annotation enrichment, structural linkage, numerical representation generation, characterization, clustering, or classification.

---

Main resource outputs

The public release is organized as:

PhageLysData_v1.0/
├── core/
├── prediction_extension/
├── context/
├── representations/
├── structures/
├── characterization/
├── metadata/
├── README.md
├── MANIFEST.tsv
└── CHECKSUMS.sha256

Core

The principal Core table is:

core/phagelysdata_core.parquet

It contains one row per Core sequence_entity_id and integrates frozen resource metadata with availability flags for downstream assets.

Additional Core assets include:

core/
├── phagelysdata_core.parquet
├── phagelysdata_core.tsv.gz
├── phagelysdata_core.fasta.gz
├── phagelysdata_core_asset_availability.parquet
├── physicochemical/
└── annotations/

Physicochemical layer

The release includes sequence-level physicochemical and descriptor tables such as:

core/physicochemical/
├── physicochemical_features.parquet
├── modlamp_descriptors.parquet
├── amino_acid_composition.parquet
├── molecular_formula.parquet
└── sequence_qc.parquet

UniProtKB annotation layer

core/annotations/uniprot/

This directory contains entity summaries, accession-level annotations, and relation tables for frozen UniProtKB-derived information.

InterProScan annotation layer

core/annotations/interproscan/

This directory contains independently generated InterProScan relations, including member-database matches, InterPro entries, Pfam matches and architectures, GO terms, and pathway annotations.

InterProScan enrichment was generated with:

InterProScan 5.78-109.0
InterPro 109.0
Pfam 38.2

Prediction Extension

prediction_extension/phagelysdata_prediction_extension.parquet

This table contains prediction-only target entities kept outside the deeply enriched Core.

Context

context/phagelysdata_context.parquet

This table preserves context/provenance entities retained by the integration workflow but lacking a valid target assertion under the frozen ontology.

Structural layer

structures/
├── structural_registry.parquet
├── structural_asset_manifest.parquet
├── ...
└── coordinates/
    ├── pdb/
    └── alphafold/

Existing structural information is included when mapped PDB structures or retrievable AlphaFold DB models are available.

No additional structural prediction is performed as part of PhageLysData v1.0.

The no_existing_structure state therefore means that no successfully retrieved PDB or AlphaFold DB asset was available under the frozen release workflow; it does not assert universal absence from external databases.

Numerical representations

representations/
├── representation_registry.parquet
├── representation_eligibility.parquet
├── one_hot/
└── plm/

PhageLysData v1.0 distributes 11 precomputed protein language model embedding matrices plus one-hot encoding for representation-eligible Core sequences.

The PLM matrices use a common representation contract:

• final hidden layer;
• mean sequence pooling;
• FP32 output;
• maximum eligible sequence length of 1,024 amino acids;
• canonical 20-amino-acid alphabet.

The representation layer includes models from the ESM-2, ESM-C, Ankh, ProtBERT, ProtT5, and Mistral-Prot families.

Characterization

characterization/

This directory contains frozen descriptive characterization tables for:

• resource universes;
• evidence Tiers;
• canonical target classes;
• Tier-by-class composition;
• source and evidence richness;
• sequence-length and physicochemical distributions;
• functional annotation coverage;
• structural coverage;
• representation coverage;
• recurrent InterPro, Pfam, GO, and pathway annotations.

---

Data availability and Zenodo

The complete, versioned PhageLysData v1.0 resource is archived on Zenodo:

DOI: https://doi.org/10.5281/zenodo.22046027
Version: v1.0
Repository: https://github.com/kren-ai-lab/phagelysdata

The Zenodo archive contains the frozen public data release only:

PhageLysData_v1.0/
├── core/
├── prediction_extension/
├── context/
├── representations/
├── structures/
├── characterization/
├── metadata/
├── README.md
├── MANIFEST.tsv
└── CHECKSUMS.sha256

The public release is the primary entry point for resource users and is the authoritative frozen dataset associated with version v1.0.

The larger construction workspace is intentionally not part of the Zenodo data release. Resource construction internally separates:

raw_data/
processed_data/
data_release/

Layer	Purpose
raw_data/	Frozen/native source packages and source snapshots used during construction
processed_data/	Staged construction, enrichment, characterization, and validation history
data_release/	Curated, portable, user-facing PhageLysData release

Worked examples consume the public release only and do not require raw_data/ or processed_data/.

The archived release includes MANIFEST.tsv and CHECKSUMS.sha256 for asset inventory and integrity verification.

Repository structure

The repository separates resource construction, release-facing reuse workflows, frozen notebook outputs, and publication-facing figure generation.

.
├── src/
│   ├── uniprot_collection_data/
│   ├── PhaLP/
│   ├── INPHARED/
│   ├── DePP/
│   ├── DposFinder/
│   ├── DepoCatalog/
│   ├── PhagesDB/
│   ├── post_processing/
│   ├── data_enrichment/
│   ├── data_representation/
│   ├── data_integration/
│   ├── data_characterization/
│   └── data_release/
│
├── notebooks/
│   ├── README.md
│   ├── 00_getting_started.ipynb
│   ├── 01_resource_characterization.ipynb
│   ├── 02_embedding_space_v0_2.ipynb
│   ├── 03_clustering_example_v0_2.ipynb
│   ├── 04_classification_example.ipynb
│   └── 05_candidate_selection_v0_2.ipynb
│
├── notebook_outputs/
│   ├── 01_resource_characterization/
│   ├── 02_embedding_space/
│   ├── 03_clustering_example/
│   ├── 04_classification_example/
│   └── 05_candidate_selection/
│
├── si_figures/
│   ├── S02_physicochemical_properties_by_class.ipynb
│   ├── S03_annotation_asset_coverage.ipynb
│   ├── S04_embedding_space_figure.ipynb
│   ├── S05_hdbscan_clustering_figure.ipynb
│   ├── S06_supervised_classification_figure.ipynb
│   ├── S07_candidate_retrieval_figure.ipynb
│   └── generated/
│
├── CITATION.cff
├── environment.yml
├── LICENSE
├── README.md
└── .gitignore

Directory description

Directory	Description
src/uniprot_collection_data/	UniProtKB query design, acquisition, hydration, evidence extraction, and source-layer closure
src/PhaLP/	PhaLP source audit, extraction, normalization, evidence construction, and closure
src/INPHARED/	INPHARED audit, extraction, normalization, candidate/evidence construction, and closure
src/DePP/	DePP audit, retrieval, normalization, evidence construction, and closure
src/DposFinder/	DposFinder audit, normalization, evidence construction, and closure
src/DepoCatalog/	DepoCatalog source processing and closure
src/PhagesDB/	PhagesDB source processing and closure
src/post_processing/	Cross-source integration ontology, resource-universe definition, physicochemical descriptors, and UniProtKB enrichment
src/data_enrichment/	InterProScan enrichment and existing-structure linkage
src/data_representation/	PLM embedding generation, one-hot encoding, and representation-layer finalization
src/data_integration/	Final Core/resource construction closure
src/data_characterization/	Frozen descriptive resource characterization
src/data_release/	Portable workspace organization and public-release materialization
notebooks/	Release-facing worked examples demonstrating direct resource reuse; see notebooks/README.md
notebook_outputs/	Frozen, lightweight outputs generated by worked examples and used where needed by publication-facing SI figures
si_figures/	Publication-facing notebooks that regenerate Supplementary Information figures from the public release or frozen notebook outputs
si_figures/generated/	Derived PDF/SVG/PNG figures and plotting tables produced by the SI figure notebooks

notebook_outputs/ and si_figures/generated/ contain derived analytical outputs, not authoritative resource data. The authoritative biological and computational resource is the frozen Zenodo release.

Software requirements

A frozen repository-level environment.yml is provided for the public worked examples, publication-facing figure notebooks, and the principal Python-based reconstruction utilities.

Public worked examples and SI figures

The frozen environment includes:

• Python 3.13;
• pandas;
• NumPy;
• SciPy;
• PyArrow;
• Matplotlib;
• scikit-learn;
• umap-learn;
• IPython / ipykernel;
• JupyterLab / Notebook;
• Biopython;
• requests;
• lxml;
• openpyxl;
• PyYAML;
• psutil;
• tqdm;
• modlAMP 4.3.2;
• Java 11, Perl, wget, and curl for supporting reconstruction utilities.

The clustering example uses sklearn.cluster.HDBSCAN; a separate external hdbscan package is not required by the public notebook suite.

No GPU is required to execute the public worked examples or publication-facing SI figure notebooks because the required PLM representations are already distributed with the release.

Full resource reconstruction

Some construction stages rely on external or stage-specific software that is intentionally not bundled into the lightweight repository environment:

• InterProScan 5.78-109.0 with InterPro 109.0 and Pfam 38.2;
• Sylphy for PLM and one-hot representation generation;
• pretrained PLM checkpoints used during numerical-representation generation;
• access to source-specific external databases/services where applicable.

Some reconstruction stages require substantial storage, memory, CPU time, GPU resources, or external service access.

Installation

1. Clone the repository

git clone https://github.com/kren-ai-lab/phagelysdata.git
cd phagelysdata

2. Create the frozen Conda environment

conda env create -f environment.yml
conda activate phagelysdata

Register the environment as a Jupyter kernel if desired:

python -m ipykernel install --user \
    --name phagelysdata \
    --display-name "Python (PhageLysData)"

3. Download PhageLysData v1.0

Download and unpack the public release from:

https://doi.org/10.5281/zenodo.22046027

The resulting directory should contain:

PhageLysData_v1.0/
├── core/
├── prediction_extension/
├── context/
├── representations/
├── structures/
├── characterization/
├── metadata/
├── MANIFEST.tsv
└── CHECKSUMS.sha256

4. Point the notebooks to the public release

Either define:

export PHAGELYSDATA_ROOT=/path/to/PhageLysData_v1.0

or set the explicit path near the top of a notebook:

from pathlib import Path

RELEASE_ROOT_OVERRIDE = Path("/path/to/PhageLysData_v1.0")

The explicit override is often the most reliable option when running notebooks through VS Code or remote Jupyter kernels.

5. Point publication notebooks to frozen worked-example outputs

SI notebooks that regenerate embedding, clustering, classification, or candidate-retrieval figures use frozen tables under notebook_outputs/.

Optionally define:

export PHAGELYSDATA_NOTEBOOK_OUTPUTS=/path/to/phagelysdata/notebook_outputs

Each SI notebook also provides an explicit *_ROOT_OVERRIDE variable when environment-variable discovery is inconvenient.

Reconstructing PhageLysData

PhageLysData is constructed through sequential and inspectable stages.

Source-specific acquisition
        ↓
Source package auditing
        ↓
Sequence and metadata normalization
        ↓
Evidence/assertion construction
        ↓
Cross-source exact-sequence integration
        ↓
Evidence-aware resource universes
        ↓
Core evidence Tiers and canonical target classes
        ↓
Physicochemical characterization
        ↓
Frozen UniProtKB annotation enrichment
        ↓
Independent InterProScan enrichment
        ↓
Existing PDB / AlphaFold DB structural linkage
        ↓
PLM and one-hot numerical representations
        ↓
Construction closure and validation
        ↓
Resource characterization
        ↓
Portable public-release materialization

The principal integration and downstream construction stages are:

Stage	Purpose
26	Build integrated exact-sequence resource
27	Define Core, Prediction Extension, Context, canonical target classes, and evidence Tiers
28	Compute Core physicochemical descriptors
29	Materialize frozen UniProtKB annotation enrichment
30	Prepare, run, and materialize independent InterProScan enrichment
31	Link and validate existing PDB / AlphaFold DB structural assets
32	Generate and finalize PLM / one-hot representation layer
33	Finalize resource construction
34	Characterize the frozen resource
35	Materialize organized workspace and portable public release

Completed stages are treated as frozen unless an actual implementation or data-integrity error is identified.

---

Worked examples

The notebooks/ directory demonstrates how a user can work directly with the public PhageLysData release.

Notebook	Purpose
00_getting_started.ipynb	Load the release, inspect Core metadata, filter entities, and retrieve associated assets
01_resource_characterization.ipynb	Reproduce descriptive resource characterization and release-facing summary outputs
02_embedding_space_v0_2.ipynb	Explore a released PLM representation using PCA and UMAP
03_clustering_example_v0_2.ipynb	Demonstrate unsupervised grouping of the released PLM space
04_classification_example.ipynb	Demonstrate simple multiclass supervised learning using released PLM embeddings
05_candidate_selection_v0_2.ipynb	Perform evidence-aware biological candidate retrieval and integrated annotation inspection

Worked-example design principles

The notebooks:

• consume only public release assets;
• use sequence_entity_id as the stable join key;
• do not access source-specific raw files or processing-stage outputs;
• write derived outputs outside the release directory;
• do not modify the distributed release;
• are intended to demonstrate interoperability and reuse, not state-of-the-art predictive performance.

The classification notebook uses a deliberately simple stratified random train/test split and should be interpreted as a reuse demonstration rather than a homology-controlled predictive benchmark.

The clustering notebook treats computational clusters as exploratory groups and does not equate them with biological families.

The candidate-selection notebook performs deterministic rule-based retrieval and does not claim validated candidate ranking or prioritization.

---

Publication and Supplementary Information figures

Publication-facing figure notebooks are maintained separately under:

si_figures/

They are intentionally separated from the analytical notebooks so that manuscript aesthetics can evolve without changing the frozen biological or analytical results.

Notebook	Publication-facing output	Primary input
S02_physicochemical_properties_by_class.ipynb	Physicochemical distributions by Core class	Public Core release
S03_annotation_asset_coverage.ipynb	Annotation, structural, and numerical asset coverage	Public Core release
S04_embedding_space_figure.ipynb	PCA/UMAP latent-space figure	notebook_outputs/02_embedding_space/
S05_hdbscan_clustering_figure.ipynb	HDBSCAN clustering figure	notebook_outputs/03_clustering_example/
S06_supervised_classification_figure.ipynb	Four-class classification figure	notebook_outputs/04_classification_example/
S07_candidate_retrieval_figure.ipynb	Evidence-aware depolymerase retrieval figure	notebook_outputs/05_candidate_selection/

The SI figure notebooks:

• do not redefine resource membership, canonical target class, or evidence Tier;
• do not rerun PCA, UMAP, HDBSCAN, classification, or candidate selection when frozen outputs are available;
• validate expected frozen counts and analysis contracts before plotting;
• export publication figures in vector PDF and SVG formats plus high-resolution PNG;
• export the plotting tables used for each figure;
• write generated outputs below si_figures/generated/.

A typical generated structure is:

si_figures/
├── S0X_...ipynb
└── generated/
    ├── S02_physicochemical_properties/
    ├── S03_annotation_asset_coverage/
    ├── S04_embedding_space/
    ├── S05_hdbscan_clustering/
    ├── S06_supervised_classification/
    └── S07_candidate_retrieval/

These files are manuscript derivatives and should not be confused with the authoritative Zenodo resource release.

---

Quick start

Clone the repository and create the frozen environment:

git clone https://github.com/kren-ai-lab/phagelysdata.git
cd phagelysdata

conda env create -f environment.yml
conda activate phagelysdata

Download and unpack the PhageLysData v1.0 release from:

https://doi.org/10.5281/zenodo.22046027

Set the release path:

export PHAGELYSDATA_ROOT=/path/to/PhageLysData_v1.0

If recreating publication figures from frozen example outputs:

export PHAGELYSDATA_NOTEBOOK_OUTPUTS=/path/to/phagelysdata/notebook_outputs

Start Jupyter:

jupyter lab

Then open:

notebooks/00_getting_started.ipynb

A minimal Python example is:

from pathlib import Path
import pandas as pd

release_root = Path("/path/to/PhageLysData_v1.0")

core = pd.read_parquet(
    release_root / "core" / "phagelysdata_core.parquet"
)

high_confidence_depolymerases = core.loc[
    core["canonical_target_class"].eq("depolymerase")
    & core["assigned_tier"].isin(["Tier 1", "Tier 2"])
]

print(high_confidence_depolymerases.shape)

For a complete walkthrough of the public notebook suite, see notebooks/README.md.

Reproducibility and traceability

PhageLysData is designed around explicit staged processing and frozen outputs.

Key reproducibility features include:

• exact-sequence entity identifiers derived from normalized sequence identity;
• preservation of source-level provenance;
• explicit prediction-derived evidence tracking;
• frozen Core / Prediction Extension / Context membership;
• frozen canonical target classes and evidence Tiers;
• independent UniProtKB-reported and InterProScan-derived annotation layers;
• explicit representation eligibility rules;
• reproducible numerical-representation settings;
• existing-structure asset manifests and coordinate-file checks;
• stage-level validation reports;
• release-level asset manifests;
• SHA-256 checksums for distributed assets;
• release-only filesystem sanitization to avoid developer-specific absolute paths;
• frozen worked-example outputs supporting publication-figure regeneration without analytical drift.

The public release includes:

MANIFEST.tsv
CHECKSUMS.sha256

These files allow users to inspect the distributed asset inventory and verify file integrity.

---

Citation

If you use PhageLysData, please cite the frozen v1.0 data release:

Medina-Ortiz, D., Olivera-Nappa, Á., Lienqueo, M. E., Ramírez, C., & Romero, J. (2026). PhageLysData: an evidence-aware and AI-ready dataset of phage lytic enzymes and depolymerases (Version 1.0) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.22046027

DOI: 10.5281/zenodo.22046027
Version: 1.0

Machine-readable citation metadata are provided in CITATION.cff.

The associated manuscript uses the same resource title and author list. Once a preprint or journal DOI is available, the publication citation can additionally be linked as the preferred citation without changing the archived PhageLysData v1.0 dataset DOI.

License

Source code

Code, workflows, executable notebooks, and environment specifications in this repository are distributed under the MIT License.

See LICENSE.

Data release

PhageLysData-original components, including the harmonization framework, controlled vocabulary, metadata schema, evidence-resolution rules, documentation, and derived records, are distributed under the Creative Commons Attribution 4.0 International (CC BY 4.0) license.

Third-party sequences, annotations, structural assets, and other source-derived content remain governed by the terms and licences of their original providers.

The authoritative licensing statement for the archived v1.0 data release is provided with the Zenodo record:

https://doi.org/10.5281/zenodo.22046027

Contact

For questions, issues, or reproducibility problems, please use the GitHub issue tracker:

https://github.com/kren-ai-lab/phagelysdata/issues

Repository:

https://github.com/kren-ai-lab/phagelysdata

Data release:

https://doi.org/10.5281/zenodo.22046027

PhageLysData is maintained within the Kren-AI Lab research ecosystem.