Skip to main content
. 2024 Nov 29;14(12):1531. doi: 10.3390/biom14121531

Figure 12.

Figure 12

The schematic diagram illustrates the core modules and training objectives of PLMs. (a) PLMs based on autoencoder frameworks tend to learn protein representations for downstream task analysis, particularly in tertiary structure prediction. (b) PLMs utilizing autoregressive frameworks are inclined to generate proteins in novel sequence spaces. (c) General models like T5 and GLM are capable of performing both tasks.