A.I. and Gene Regulation

Understanding gene regulation is a bitch. Seriously, this is one of the thornier problems of science today and it is because the complexities of living cells, with the thousands of proteins in each cell, make tracing a single protein’s regulation just as complex as hell. Smart guys Tareen and Kinney have figured out a way for AI to interpret (some) interactions using massively-parallel reporter assays to figure out the biophysical basis for (some) gene regulation…which is more than we have had heretofore. They did this by assigning nodes and weights with explicit physiochemical interpretations. This last is the important bit; many AI algorithms are very difficult to interpret, so the underlying “logic” is impenetrable to humans.  The smart guys made many of the decisions explicit, so they would be better able to understand the “logic” by which the characterizations were derived.

 

 

Homework: Biophysical models of cis-regulation as interpretable neural networks, Ammar Tareen, Justin B. Kinney BioRxiv, doi: https://doi.org/10.1101/835942