SimPep and OP-AND: A deep learning framework and curated database for predicting osteogenic peptides

Abstract

Bone health is a growing concern in aging populations, and bioactive peptides in dairy products offer a promising approach to preventing bone-related diseases. However, the lack of a public database for osteogenic peptides (OPs) has limited the computational detection efforts. In this work, we introduce OP-AND, a curated public database of osteogenic peptides. We also propose a novel hypothesis that peptides derived from proteins involved in osteoclast formation may serve as non-osteogenic. Considering the limited availability of OP data, we present SimPep, a deep learning framework that achieves 86.87% accuracy and 76.88% area under receiver-operating characteristic curve score using 5-fold cross-validation. SimPep’s performance is further evaluated on external datasets, and a pipeline is introduced to select potential OPs for experimental studies. The camel milk alpha s1-casein peptide ‘MKLLILTCLVAVALARPKYPLRYPEVF’ is highlighted as a top candidate for future exploration. The OP-AND database is available in https://github.com/CBRC-lab/SimPep_and_OP-AND.