Throw out an oligopeptide to catch a protein: Deep learning and natural language processing-screened tripeptide PSP promotes Osteolectin-mediated vascularized bone regeneration.

Abstract

Angiogenesis is imperative for bone regeneration, yet the conventional cytokine therapies have been constrained by prohibitive costs and safety apprehensions. It is urgent to develop a safer and more efficient therapeutic alternative. Herein, utilizing the methodologies of Deep Learning (DL) and Natural Language Processing (NLP), we proposed a paradigm algorithm that amalgamates Word2vec with a TF-IDF variant, TF-IIDF, to deftly discern potential pro-angiogenic peptides from intrinsically disordered regions (IDRs) of 262 related proteins, where are fertile grounds for developing safer and highly promising bioactive peptides. After the evaluation of the candidate oligopeptides, one tripeptide, PSP, emerged as particularly notable for its exceptional ability to stimulate the vascularization of endothelial cells (ECs), enhance vascular-osteo communication, and then boost the osteogenic differentiation of bone marrow stem cells (BMSCs), evidenced in mouse critical-sized cranial model. Moreover, we found that PSP serves as a 'priming' agent, activating the body's innate ability to produce Osteolectin (Oln) - prompting ECs to release small extracellular vesicles (sEVs) enriched with Oln to facilitate bone formation. In summary, our study established a precise and efficient composite model of DL and NLP to screen bioactive peptides, opening an avenue for the development of various peptide-based therapeutic strategies applicable to a broader range of diseases.