Processed HepG2 drug-perturbation proteomics data for DeepHeal

Abstract

This dataset supports the 18-profile HepG2 analysis in “DeepHeal: An interpretable proteomics-first framework for computational drug repurposing”. It contains supplied processed protein log2 fold changes versus DMSO, the corresponding 18 × 9,960 aligned feature matrix, Average-194 and SupCon-257 frozen consensus representations, profile coverage, compound metadata, and 36 saved RealTabPFN-v2.5 predictions. HepG2 cells were treated with 18 compounds at 10 micromolar for 24 hours in biological duplicate. Seventeen candidate profiles are provided in one processed table (6,442 protein rows), and MI-773 in a separate table (7,520 protein rows). The supplied contrasts are averaged profiles. Individual replicate intensities and raw mass-spectrometry files are not included. This is distinct from earlier 367-compound and legacy 17-profile datasets. The aligned matrix uses the archived HCT116 protein order, zero-fills absent/nonfinite features, and sets absolute log2FC values of 0.15 or less to zero. Both branches use frozen model components; no fitting or threshold tuning was performed on HepG2. Average-194 is primary (14/18 positive calls), and SupCon-257 is sensitivity (15/18). These are assumed-positive-panel recovery results, not specificity or balanced accuracy estimates. Scores are uncalibrated screening readouts. The 36 rows represent two analyses of 18 profiles. Version 1.0.0 packages the existing frozen analysis outputs from 5 October 2026. README, a field dictionary, experimental metadata and SHA-256 checksums accompany the data. Source code and instrument raw data are outside this processed-data release. Contact: Hua Li (lihua@fjtcm.edu.cn) and Lixia Chen (syzyclx@163.com). Funding: National Natural Science Foundation of China (U24A20784); Fujian Provincial Science and Technology Innovation Joint Fund Project (2024Y9503); Natural Science Foundation of Fujian Province (2025J01929); Scientific Research Foundation for High-level Talents of Fujian University of Traditional Chinese Medicine (X2024005-talents); and the “Select the best candidates to lead key research projects” program of Fujian University of Traditional Chinese Medicine (XJB2022008).