SeqEngine is a fully automated, end‑to‑end pipeline for feature extraction and feature selection from biological sequences (protein and DNA). It extracts a comprehensive set of 610 features for proteins and 400 features for DNA, implements seven feature selection methods (univariate filters, wrapper, and embedded), and outputs multiple curated feature subsets along with a detailed text report containing sequence statistics, feature rankings, and performance metrics. Designed for researchers who need a reproducible, interpretable, and efficient tool to pre‑process sequence data for machine learning.
The figure below shows the complete workflow of SeqEngine:
- Comprehensive Feature Extraction:
- Protein: 610 features including AAC, dipeptide, PAAC, CTD, Moran autocorrelation, BLOSUM62, and physicochemical properties.
- DNA: 400 features including NAC, DNC, TNC, k‑mer (k=4), PseDNC, Moran autocorrelation, and ENAC.
- Multiple Feature Selection Strategies:
- Univariate Filters:
f_classif,mutual_info,chi2 - Wrapper: Recursive Feature Elimination (RFE) with Random Forest
- Embedded: Random Forest importance, Lasso (L1), Gradient Boosting importance
- Univariate Filters:
- Human‑Readable Output: Feature names like
AAC_Alanine,CTD_Comp_Hydrophobicity_G1,Kmer_AAAA—no more crypticF1,F2. - Batch Processing with
tqdmprogress bars for large datasets. - Parallel Execution using multiple CPU cores.
- Comprehensive Report: Sequence statistics, feature statistics, method‑by‑method performance (accuracy, precision, recall, F1, MCC), top‑20 ranked features, and execution times.
- Multiple Output Formats: Consolidated feature CSV, per‑method selected feature CSVs, aggregate ranking summary, and a full text report.
pip install seqengine