#proteins!

Pls pls can you explain more about the stuff you're doing? To a ... not science-minded guy? Because I'm so interested in like, I think you said you were designing new proteins?? How does that work???

okay thank you i love you <3

hi boots!! i would love nothing more than to talk about my stem stuff<3

so for context- i am currently in undergrad for a biochemistry degree with a minor in global health. my research interests are primarily in infectious diseases, therapeutics + drug design/development, and health systems. currently i intend to go to grad school for either biochemistry, medicinal chemistry, or pathobiology but those doors are still very open and possible to change.

this summer i'm interning at a medicinal chemistry lab that specializes in de novo peptide design.

de novo peptide/protein design is a field that exists at the intersections of biochemistry, bioengineering, and computer science. at its core, it utilizes deep learning/machine learning techniques to generate new protein backbones and sequences.

basically- we use artificial intelligence that is trained off the protein data bank (PDB, a database of real, naturally occurring protein structures) to design new proteins to be used as anything from therapeutics to biodegraders and more!

**notably, yes there are AI models involved but i would like to draw a clear distinction between AI used in scientific settings such as this & genAI chatbots that have co-opted the term "AI" writ large.**

what i'm working on this summer is designing an antibiotic peptide to target a specific gram negative bacterial protein complex.

basically the way we do this is we look at the structure of our target protein given by the PDB (often these structures were determined experimentally via x ray crystallography) and using previous research that has determined where on the complex might be attractive binding sites, we choose specific residues (single amino acids) at these sites that we want our new, designed peptide to bind to.

and we have code that tells the machine learning models a) what our target protein is and b) what our chosen resides are, and it computationally generates possible protein/peptide backbone structures that could bind to these spots.

now the coding & tech side is not so much my thing, so i'm not as in the know on the minutia of the code, but basically we run the code and it gives us backbones. and then we choose the best backbones, get attractive amino acid sequences for these backbones, and the models will tell us which sequences+backbones are most likely to bind how we want them to. at which point we select the best options and synthesize them chemically in the lab!

You’re welcome.