UHT Discovery

UHT Discovery is a protein discovery pipeline centered on three core steps:

1. blaster: retrieve homologous sequences from NCBI
2. trim: quality-control and length-filter FASTA sequences
3. clust (PLMCLUSTV2): cluster sequences using ESM2 embeddings and NLL scoring

Installation

From Source

git clone <repository-url>
cd uht-discovery-package
pip install -e .

From PyPI

pip install uht-discovery

Quick Start

1. BLASTER

Put one or more query FASTA files in:

inputs/blaster/PROJECT/

Run:

uht-blast --project PROJECT --email your@email.com --hits 500

Key options:

• --project: project directory name (required)
• --email: email for NCBI API usage (required)
• --hits: number of hits to retrieve (default: 100)
• --db: nr, swissprot, or refseq_protein (default: nr)
• --evalue: BLAST E-value cutoff (default: 1e-5)

Outputs:

• results/blaster/PROJECT/ with combined FASTA and BLAST report

2. TRIM

Put FASTA files in:

inputs/trim/PROJECT/

Run (automatic thresholds):

uht-trim --project PROJECT --auto

Run (manual thresholds):

uht-trim --project PROJECT --low 100 --high 500

Key options:

• --project: project directory name (required)
• --auto: infer thresholds automatically
• --low / --high: manual inclusive length thresholds

Outputs:

• results/trim/PROJECT/ with filtered FASTA files, removed-sequence logs, and QC plots

3. PLMCLUSTV2

Put FASTA files in:

inputs/plmclustv2/PROJECT/

Run:

uht-clust --project PROJECT --clusters auto

Or fixed cluster count:

uht-clust --project PROJECT --clusters 6

Key options:

• --project: project directory name (required)
• --clusters: integer cluster count or auto
• --sil-min / --sil-max: silhouette search bounds for auto mode
• --keep-separate: process each FASTA file independently

Outputs:

• results/plmclustv2/PROJECT/ including cluster FASTAs, metrics CSVs, representative sequences, and visualization files

Recommended Workflow

Run the pipeline in this order:

1. uht-blast
2. uht-trim
3. uht-clust

Help

uht-blast --help
uht-trim --help
uht-clust --help