Date of Award
Fall 2025
Document Type
Open Access Dissertation
Department
Chemical Engineering
First Advisor
Esmaiel Jabbari
Abstract
The use of rhBMP2 has several side effects in its application as a growth factor in biomaterials in treating bone related issues. Existing issues of the protein’s nature to denature, possess a short half-life, expensive synthesis process, coupled with its tendency to diffuse away from the site of regeneration that have been observed to lead to various complications make its application questionable. Peptides have been relied on as an alternate source as an osteogenic molecule inspired by the knuckle epitope (KEP) region of the BMP2 which is a 20-mer peptide. However, the peptide sequence is in a more collapsed configuration in its free state than in its native state where it has an ‘open-arm’ configuration which is attributed towards the low osteogenic activity of the peptide in its free state.
In order to identify new 20-mer peptides that can be osteogenic, machine learning (ML) can be used to predict structural descriptors of different sequences. As these ML models can potentially require hundreds of data for training and testing, using data from quantum or mechanical models isn’t a feasible process. Hence, we developed a coarse-grained Dissipative Particle Dynamics (DPD) model for simulating peptide behavior in aqueous environment known as Structure Independent Molecular Fragment Interfuse Model (SIMFIM). SIMFIM doesn’t impose dihedral constraints on the backbone of the peptide as it doesn’t rely on a priori knowledge of the structure. The non-bonded interaction parameters are determined by considering specific interactions. The electrostatic interactions are incorporated in this model by using a normal distribution of charges around the center of the beads to prevent the collapse of oppositely charged soft beads. The uniquely parameterized DPD force field in the SIMFIM model is optimized for a given peptide with respect to the degree of coarse-grained graining for simulating the peptide over long times and length scales. The SIMFIM model was tested in this work using four peptides, namely, TrpZip2, Rubrivinodin, Lihuanodin, and IC3-CB1/Gai peptides, whose structures were sourced from the Protein Data Bank. The SIMFIM model predicted radius of gyration (Rg) values for the peptides closer to the actual structures as compared to the conventional model, and there was less Root Mean Square Deviation (RMSD) between the predicted and actual structures of the peptides.
This model was then used to simulate ~700 peptides, which were obtained by strategically modifying the free knuckle epitope peptide. A dataset was constructed containing the structural descriptors average radius of gyration () and average end to end distance (). For ML modeling, the residues in the 20-mer sequences, as the input features of the database, were represented by different amino acid descriptor (AAD) scales. The performances of all the models were compared using the R2 and other performance metrics. Permutation importance and SHAP interaction analysis were done to determine which residue positions and properties had the highest contribution to the structural properties of the sequences. These studies led to developing trained and tested (Quantitative Structure Activity Relationships) QSARs for predicting the structural properties of any modified BMP2-KEP sequence for the purpose of discovering novel 20-mer sequences with open-arm structures.
Rights
© 2025, Ricky Anshuman Dash
Recommended Citation
Dash, R. A.(2025). Engineering Configurational Mimetic Osteogenic Peptides Through Mesoscale Simulation and Machine Learning. (Doctoral dissertation). Retrieved from https://scholarcommons.sc.edu/etd/8678