Description: Although AlphaFold 3 supports non-canonical chemical modifications, recent findings highlight performance gaps when predicting peptides with backbone N-methylated amino acids (BNMeAAs) [1, 2]. This project aims to investigate where and why AlphaFold 3 fails on these modified backbones.
Scope of Work:
• Data Curation: Collect a dataset of BNMeAA-containing peptides with resolved ground-truth structures (start with [3] and PDB).
• Structure Prediction: Predict target structures using AlphaFold 3 and sample conformational landscapes using Rosetta simple_cycpep_predict (SCP).
• Comparative Analysis: Quantify structural discrepancies (RMSD, backbone torsion deviations, hydrogen-bonding networks) between AlphaFold 3, Rosetta SCP, and experimental ground truths to identify specific modeling bottlenecks.

References
1. Cao, Zhigang, et al. "HighFold-MeD: a Rosetta distillation model to accelerate structure prediction of cyclic peptides with backbone N-methylation and D-amino acids." Journal of Cheminformatics 17.1 (2025): 167.
2. Cao, Zhigang, et al. "HighFold-MeD2: An Enhanced Boltz-2 Model for Accurate Structure Prediction of N-Methylated and d-Amino Acid Cyclic Peptides." Journal of Chemical Information and Modeling 66.10 (2026): 6057-6066.
3. https://github.com/hongliangduan/HighFold-MeD2