panCATE includes a collection of scripts used for PanGBank pangenomes clustering, annotation, tokenization, and embedding, conducted in the course of a 6-month internship at LABGeM, Genoscope.
Alignment-based clustering, curated annotations, and AI-based protein embeddings are explored in parallel to achieve a comprehensive characterization of PanGBank pangenomes, with the final aim of supporting the PanGAIMiX project, primarily its third work package (WP3), aimed at developing an annotation-level Large Language Model (LLM).
This repository is intended to be a lightweight index and code archive. Large outputs and raw results are left on the server, as documented in docs/scratch_inventory.md and manifests/; selected analysis figures and compact finding summaries are tracked under analysis/.
scripts/: analysis and utility scripts, organized in 5 themes (annot, clust, embed, tokenize, utils).analysis/: curated figures and compact findings from the internship analyses.docs/: documentation of all scripts, analyses, and outputs.manifests/: compact TSV inventories of the main outputs and results on the server.envs/: conda environments needed.report/: internship report.The clustering and annotation workflow lives separately at: PanGBank-annotator.
analysis/: clustering, annotation, and embedding analyses on GTDB_all v1.0.0.outputs/: outputs by PanGBank release version.data/: GTDB_all v1.0.0 input data used as the primary dataset during the internship.logs/: Slurm logs.archived/: relevant older or non-central results.Check docs/scratch_inventory.md for more details.
The following documents are available in docs/:
docs/scratch_inventory.md: documentation of all server-side directories.docs/outputs.md: release-oriented view of outputs/.docs/environments.md: practical environment guidance.docs/analyses/: curated analyses and findings from the internship.
docs/analyses/annotation.md: annotation analyses.docs/analyses/clustering.md: clustering analyses.docs/analyses/embedding.md: embedding analyses.docs/analyses/tokenization.md: tokenization analyses.docs/analyses/structural_validation.md: structural validation of ESM-based superclusters.Scripts are organized by the major task they support:
annot/: annotation runs, clusters functional consistency, and annotation vocabulary analyses.clust/: clustering runs, method comparisons, validation, and ESM projection support.embed/: ESM embedding computation, embedding export, embedding-space analyses, and pooling approaches exploration.tokenize/: annotation propagation to all genomes, gene tables enrichment, all-genomes-level analyses, and vocabulary optimization.utils/: reusable data setup, representative extraction, compression, and sampling helpers.panCATE includes a collection of scripts used for PanGBank pangenomes clustering, annotation, tokenization, and embedding, conducted in the course of a 6-month internship at LABGeM, Genoscope.
Alignment-based clustering, curated annotations, and AI-based protein embeddings are explored in parallel to achieve a comprehensive characterization of PanGBank pangenomes, with the final aim of supporting the PanGAIMiX project, primarily its third work package (WP3), aimed at developing an annotation-level Large Language Model (LLM).
This repository is intended to be a lightweight index and code archive. Large outputs and raw results are left on the server, as documented in docs/scratch_inventory.md and manifests/; selected analysis figures and compact finding summaries are tracked under analysis/.
scripts/: analysis and utility scripts, organized in 5 themes (annot, clust, embed, tokenize, utils).analysis/: curated figures and compact findings from the internship analyses.docs/: documentation of all scripts, analyses, and outputs.manifests/: compact TSV inventories of the main outputs and results on the server.envs/: conda environments needed.report/: internship report.The clustering and annotation workflow lives separately at: PanGBank-annotator.
analysis/: clustering, annotation, and embedding analyses on GTDB_all v1.0.0.outputs/: outputs by PanGBank release version.data/: GTDB_all v1.0.0 input data used as the primary dataset during the internship.logs/: Slurm logs.archived/: relevant older or non-central results.Check docs/scratch_inventory.md for more details.
The following documents are available in docs/:
docs/scratch_inventory.md: documentation of all server-side directories.docs/outputs.md: release-oriented view of outputs/.docs/environments.md: practical environment guidance.docs/analyses/: curated analyses and findings from the internship.
docs/analyses/annotation.md: annotation analyses.docs/analyses/clustering.md: clustering analyses.docs/analyses/embedding.md: embedding analyses.docs/analyses/tokenization.md: tokenization analyses.docs/analyses/structural_validation.md: structural validation of ESM-based superclusters.Scripts are organized by the major task they support:
annot/: annotation runs, clusters functional consistency, and annotation vocabulary analyses.clust/: clustering runs, method comparisons, validation, and ESM projection support.embed/: ESM embedding computation, embedding export, embedding-space analyses, and pooling approaches exploration.tokenize/: annotation propagation to all genomes, gene tables enrichment, all-genomes-level analyses, and vocabulary optimization.utils/: reusable data setup, representative extraction, compression, and sampling helpers.