In silico design of smaller size enzymatic protein by generative artificial intelligence (ProtGPT2)

Area of Science:

• Protein engineering
• Computational biology
• Biotechnology

Background:

• Designing small proteins is vital for bioprocessing and drug development.
• Traditional methods struggle with stability and function replication.
• Novel computational approaches are needed for efficient protein design.

Purpose of the Study:

• To develop a novel method for designing small proteins using ProtGPT2.
• To fine-tune ProtGPT2 for malate dehydrogenase (MDH) sequence generation.
• To validate the functional and structural integrity of designed small MDH variants.

Main Methods:

• Collected malate dehydrogenase (MDH) amino acid sequence data.
• Fine-tuned ProtGPT2 model for MDH sequence generation (ProtGPT2 for MDH).
• Evaluated generated sequences using population (MSA, tSNE) and individual (BLAST, AlphaFold2, InterPro) analyses.

Main Results:

• ProtGPT2 for MDH generated shorter sequences than natural MDH.
• Population analysis confirmed identification and incorporation of MDH functional motifs.
• Individual analysis revealed 9 novel MDH variants with high structural similarity and identified functional domains.

Conclusions:

• ProtGPT2 for MDH shows potential for designing novel small MDH variants.
• The designed sequences maintain structural and functional characteristics of natural MDH.
• Future experimental validation is required to confirm the utility of this protein design approach.