Camelia Bhattacharyya, Amity University Kolkata
Proteins are the main functional biomolecules in the human body. These are the ones that are responsible for most of the functions in our body and maintains the structure and regulates other metabolic balances in the body. Thus, proper knowledge of it is a must. But the highly complicated structure of the proteins and the difference each protein showcases in the arrangement of the amino acids through peptide bonds is something of a challenge to be established manually as a 3D structure. This challenge has always interfered in the process of learning the physicochemical properties of proteins over the years. Recently, researchers from Vilnius University (Lithuania) and Chalmers University of Technology (Sweden) have taken this challenge as a positive deal and have answered in the smartest of the ways possible by developing the proteinGAN. Someday the implementation of this research can solve thousands of questions related to proteins.
Now, what really is this proteinGAN? It is the protein generative adversarial network that uses information from the one working with a protein and thus the database searches for similarities, homologs, and gaps, ending up giving a proper structural hypothesis related to the unknown protein under study. This structure designing is something of a delicate kind since a tiny difference in the structure of amino acids can change a functional protein into a nonfunctional one resulting in fatal diseases like cancer, etc. Thus, different permutations and combinations along with proper photography and videography of the studied protein sample are required for this process to render accurate results.
The ProteinGAN runs in 3 steps: the generator which creates pictorial data of the unknown studied protein, the discriminator which studies the data produced by the generator and compares it with the original data stored in the database and generates new data based on its study, the generator now gives the final structural data based on the discriminator’s analysis. This process is repeated several times to create more authentic data.
In the first experimental setup, the team used Malate dehydrogenase (MDH) as the template enzyme and showed how about 24% of the AI-generated protein structure exhibited catalytic activity of the MDH and were soluble too in vitro conditions. There were new mutant variants too which proves the fact that in near future such unknown and non-existing proteins might be created which would solve several biological equations and complexities and bring down mortality rate with time.
Also read:Cathepsin L inhibitor- A potential preventive of Covid-19?
Source:Repecca, D., Jauniskis, V., Karpus, L. et al. Expanding functional protein sequence spaces using generative adversarial networks. Nat Mach Intell (2021). doi:http://10.1038/s42256-021-00310-5
- The Corrosion Prediction from the Corrosion Product Performance
- Nitrogen Resilience in Waterlogged Soybean plants
- Cell Senescence in Type II Diabetes: Therapeutic Potential
- Transgene-Free Canker-Resistant Citrus sinensis with Cas12/RNP
- AI Literacy in Early Childhood Education: Challenges and Opportunities
One thought on “ProteinGAN: The merge of machine learning and proteomics”