Code and reproduction artifact for "Reassessing Sequence-Based Cold-Start Protein–Protein Interaction Prediction: A Leakage-Controlled Benchmark, an Audit-Driven Reproducibility Study, and Calibrated Selective Prediction over Frozen PLM Embeddings"
Year: 2026
Type: software
Abstract: This archive contains the code, exact data splits, and per-experiment result records that reproduce all results in the IEEE Access paper "Reassessing Sequence-Based Cold-Start Protein–Protein Interaction Prediction: A Leakage-Controlled Benchmark, an Audit-Driven Reproducibility Study, and Calibrated Selective Prediction over Frozen Protein-Language-Model Embeddings" by Muhammed Tekin and Murat Gök (Yalova University). It includes: (i) the ppi_c3 Python package and search/training code; (ii) self-contained, checkpointed reproduction drivers, one per analysis (homology control, simple-classifier panel, seven design-axis probes, calibration and split-conformal selective prediction, and the Bernett second benchmark); (iii) the exact Park–Marcotte yeast DIP splits used in the paper and the Bernett split assignment; and (iv) machine-readable per-seed result records. Large or licensed inputs are not bundled and are regenerated or downloaded via the included scripts: the raw DIP interactions, the SGD localization tables, and the multi-gigabyte ESM-2 embedding cache. A helper script downloads the Bernett human gold standard from figshare (10.6084/m9.figshare.21591618). All comparisons use matched random seeds. See README.md for setup instructions and a driver-to-paper map. License: MIT (code); CC BY 4.0 (documentation and derived data).
Authors: TEKIN, MUHAMMED, Gök, Murat
Institution: Yalova University
Language: English
Subfield: Molecular Biology
Domain: Life Sciences
Open Access status: green
OpenAlex ID: W7166694844