Leveraging Artificial Intelligence for Gene and Peptide Design

Artificial intelligence (AI) is growing rapidly and has the potential to transform all aspects of biomedical sciences. Generative AI (Gen AI) is a subclass of AI technologies that can produce new content based on configurations and arrangements learned from existing data by applying deep learning technology. Deep learning is a subset of machine learning that utilizes artificial neural networks with too many hidden layers from large datasets for humans to analyze and recognize complex patterns and classifications. Deep learning is particularly effective for speech recognition and natural language processing. Gen AI is likely to revolutionize drug discovery and synthetic biology by accelerating the discovery of novel therapeutic peptides and genes. By analyzing vast datasets of genetic sequences and protein structures, Gen AI models can forecast those alterations in DNA or peptide sequences that will impact their functions and interactions.[1] Gen AI models can be employed to design short-chain therapeutic peptides for targeting a wide range of microorganisms.[2,3] A recent paper published in Science introduced the Evo, a specialized Gen AI model, which is trained on 300 billion nucleotides of sequence information, enabling it to devise new proteins and even entire bacterial genomes.[4] We attempted to use Gene Editing GPT, which has been developed by futuredisruptor.com. The developers claim, “This GPT is designed to aid researchers, enthusiasts, and scientists in the rapidly progressing field of gene editing technologies like CRISPR. It provides powerful insights and research support capabilities.” We asked Gene Editing GPT to “Suggest a small peptide to inhibit glucosyltransferase of Streptococcus mutans.” This is a well-known glucosyltransferase that plays critical roles in the virulence of Streptococcus mutans by generating biofilms containing extracellular polysaccharides that enhance the colonization of bacteria.[5] Gene Editing GPT suggested a short peptide with the following amino acid sequences: RRWQWRMKKLG based on theoretical principles such as targeting the active or substrate-bonding regions of glucosyltransferase. Gene Editing GPT claimed the peptide contains a combination of hydrophobic (such as Tryptophan) and positively charged residues such as arginine and lysine. This structure is designed to bind to negatively charged or hydrophobic regions of glucosyltransferase. The “RRWQW” provides Pi-stacking interactions and electrostatic attractions. The “MKKLG” stabilizes the peptide and potentially enhances binding. We have employed a de novo tool, PEP-FOLD4, to predict and visualize the peptide structure from the amino acid sequences [Figure 1].[6] To assess the potential effectiveness of this peptide, in silico and in vitro tests can be done. The crystal structure of glucansucrase from Streptococcus mutans is well known (www.rcsb.org/structure/3AIC). Computational molecular docking studies with tools such as AutoDock (https://autodock.scripps.edu/) can predict bonding affinity. Also, experimental assays like enzymatic inhibition assays or biofilm formation assays can be used to assess the effectiveness of this peptide in vitro.Figure 1: Predicted peptide structure of amino acid sequences “RRWQWRMKKLG” suggested by Gene Editing GPT to inhibit glucosyltransferase of Streptococcus mutans.It is necessary to address safety and ethical concerns regarding synthetic biology. Before using this peptide as a preventive or therapeutic agent toxicity on human oral cells or systemic toxicity must be assessed. The peptide must selectively inhibit glucosyltransferase of Streptococcus mutans without disrupting the balance of the oral beneficial microbiome. Over-targeting may lead to microbial imbalance (dysbiosis) in the oral cavity and consequently have broader health implications. Additionally, the limitations of Gen AI must be appreciated, which cannot think like humans. Gen AI is a statistical, not cognitive, system and cannot research the background to a topic; it only predicts the next words or other types of outputs based on training data. It can generate convincing answers that cite nonexistent studies or events that are factually incorrect and entirely fabricated, known as “hallucinations.[7] In conclusion, Gen AI remains a transformational tool for gene and peptide design, yet it requires safeguards and human oversight. By combining Gen AI compatibilities with robust experimental assessments and ethical considerations, the biomedical research community can unlock its full potential for biomedical advancements.