Comprehensive Evaluation and Optimization of Martini Simulations for Dipeptide Self-Assembly.

Accurate prediction of peptide self-assembly remains challenging due to the strong dependence of coarse-grained molecular dynamics (CGMD) simulations on parameter settings. Here, we establish a systematic computational-experimental framework for evaluating and optimizing Martini-based simulations of dipeptide self-assembly. Using 40 chemically diverse dipeptides, we quantitatively analyzed the effects of simulation time, peptide concentration, system size, backbone bead type assignment (H/E/C in martinize), and terminal charges on aggregation propensity (AP). We found that secondary structure and terminal charge are critical determinants of aggregation behavior by altering coarse-grained particle types and Lennard-Jones interactions, while their influence diminishes with increasing peptide length. Experimental validation by transmission electron microscopy confirmed that simulations with uncharged termini and β-sheet secondary structure assignment, which specifies the corresponding backbone bead types in the coarse-grained model, achieved the highest predictive accuracy. Applying these optimized parameters, we performed a comprehensive screening of all dipeptides and identified multiple new self-assembling candidates. This work provides mechanistic insight into parameter-dependent variability in Martini simulations and offers practical guidelines for reliable modeling of short-peptide self-assembly.