TABLE 1.

Features, advantages and limitations of 5 different protein structure prediction models.

Method	Feature	Advantages	Limitations
AlphaFold2	A neural network architecture combining attention mechanisms and evolutionary information	1. High accuracy in protein structure prediction 2. Continuous updates and development	1. High computational resources requirements 2. Homologous sequence dependence 3. Lack of fine structure prediction ability
Rosetta	Uses energy functions with fragments, Monte Carlo strategies	High computational efficiency with low search space	1. Limited exploration for intricate topology proteins 2. Low-resolution energy functions
RoseTTAFold All-Atom	Merges sequence-based representations of biopolymers with atomic graph representations of small molecules and covalent modifications	Prediction of proteins, nucleic acids, small molecules, metals, covalent modifications	1. Average accuracy 2. Small training datasets
ESMFold	Utilizes protein language model with training parameters instead of MSA.	1. Faster prediction speed. 2. Efficient exploration of large-scale protein structure space	1. Limited prediction accuracy 2. Less effective with complex structures
RGN2	Uses AminoBERT language model and recurrent geometric network	Prediction of orphan and de novo-designed protein structures	1. Poor prediction with sufficient sequence homologs 2. Hard to predict beta-sheet structures 3. Limited to local dependencies between Cα atoms