pyprego

Python implementation of the prego R package — a PWM Regression Optimizer for motif discovery in DNA sequences.

Installation

pip install -e .

Optional dependencies:

pip install pymisha   # for genomic interval integration
pip install logomaker  # for sequence logo plots

Quick Start

Continuous regression (find motifs correlated with a response)

import pyprego

# sequences: list of equal-length DNA strings
# response: 1D or 2D numpy array (one row per sequence)
result = pyprego.regress_pwm(sequences, response)

# Result contains:
result.pssm       # PSSM DataFrame (pos, A, C, G, T)
result.spat       # Spatial model DataFrame (bin, spat_factor)
result.pred       # Predictions for each sequence
result.consensus  # Consensus motif string
result.r2         # R-squared per response dimension

# Predict on new sequences
new_scores = result.predict(new_sequences)

Binary classification (find motifs that discriminate two classes)

result = pyprego.regress_pwm(
    sequences, binary_response,  # 0/1 vector
    score_metric="ks"
)
result.ks    # KS test statistic
result.pred  # Predictions

Multiple motifs

result = pyprego.regress_pwm(sequences, response, motif_num=3)
result.models      # List of individual motif models
result.multi_stats # Statistics for each motif
result.pred        # Combined predictions

PWM scoring with known motif

scores = pyprego.compute_pwm(sequences, pssm, spat=spat_model, bidirect=True)
local_scores = pyprego.compute_local_pwm(sequences, pssm)

K-mer screening

kmers = pyprego.screen_kmers(sequences, response, kmer_len=8)
print(kmers.head())  # Top correlated k-mers

PSSM utilities

pyprego.pssm_cor(pssm1, pssm2)       # Correlation between PSSMs
pyprego.pssm_match(pssm, motif_db)   # Match against database
pyprego.bits_per_pos(pssm)            # Information content
pyprego.consensus_from_pssm(pssm)     # Consensus sequence
pyprego.pssm_rc(pssm)                 # Reverse complement
pyprego.pssm_trim(pssm)              # Trim low-info edges

Model export/import

from pyprego.export import export_regression_model, load_regression_model

export_regression_model(result, "model.json")
loaded = load_regression_model("model.json")
new_scores = loaded.predict(new_sequences)

API Compatibility with R prego

pyprego implements the same functions as the R package:

Testing

# Fast tests (~6 seconds)
pytest tests/ --ignore=tests/test_high_level.py --ignore=tests/test_regression.py --ignore=tests/test_integration.py

# Full suite (includes slow regression tests)
pytest tests/

Architecture

• NumPy-based: All computation uses NumPy arrays (no GPU/PyTorch dependency)
• pandas DataFrames: PSSMs and spatial models use DataFrames matching R conventions
• Optional pymisha: Genomic functions work when pymisha is installed
• GPU-ready design: Clean array interfaces allow future torch tensor swap

License

MIT