Abstract glowing virtual neural network to represent the use of AI (artificial intelligence) in personalized medicine
Credit: FlashMovie/Getty Images

Researchers have developed a new protein simulation that quickly reveals previously hidden protein binding residues—so called undruggable targets. Known as PocketMiner, the neural network tool may help find new treatments for the 50% of cancer-signaling proteins currently considered to be undruggable due to their lack of targetable protein regions. The research from the University of Pennsylvania is published in Nature Communications.

Protein pockets, areas within protein structures to which certain proteins or molecules can bind, are easily found in about half of cancer-related proteins. But some pockets, known as cryptic pockets, remain hidden with most current technologies.

“In the past, we’ve made some good progress with using atomically detailed simulations of proteins to see how these pockets open up,” explained senior author Gregory Bowman, PhD, a professor of Biochemistry and  Biophysics and Bioengineering at Penn, whose lab has been developing a mixture of simulation-based and experimental tools for finding and targeting cryptic pockets and their buried protein residues. “But it has taken a lot of computer time and human time and there are so many targets we want to go after,” he says.

Earlier machine learning techniques relied on the roughly 100 examples about of cryptic pocket structures in Protein Databank. “That’s not a lot of data to work with, so we tested thousands of proteins in our simulations, a good subset of which we’ve experimentally confirmed are predictive,” added Bowman.

PocketMiner is artificial intelligence that predicts where cryptic pockets are likely to form from a single protein structure and learns from itself. The team developed PocketMiner as a way to speed that process up, to predict if and where cryptic pockets are likely to form from a single structure ‘at the snap of a finger’. The network allows researchers to quickly decide if a protein is likely to have cryptic pockets before investing in more expensive simulations or experiments to pursue a predicted pocket further. “We can rapidly decide if a protein is worth spending time on, if we want to go after a certain signaling cascade, or otherwise follow the proteins we are most likely to have success with,” added Bowman.

Thus far, the Penn team simulated single protein structures and successfully predicted the locations of cryptic pockets in 35 cancer-related protein structures in thousands of areas of the body. These once-hidden targets, now identified, reveal potential new approaches for cancer treatment.

The Penn team trained PocketMiner to predict where pockets are likely to open in molecular dynamics simulations.  They started by gathering snapshots separated by 50 nanoseconds in time starting from the initial protein structure to view and then predict which buried residues become exposed to solvent. “And we applied this information to crystal structures or alpha fold structures or homology models to predict the probabilities that every residue in the protein participates in a cryptic pocket,” said Bowman. “It turned out quite well on our test of experimental structures.”

In this paper, Bowman’s team highlighted two key protein structures within cancer-signaling pathways. The first, WNT2 protein in the Jak/Stat pathway is an integral part of cancer signaling in many solid tumors. The second is PIM2, a particular enzyme that is implicated as a driver of several types of cancer, including those of the lung, prostate, and breast as well as leukemia, and myeloma.

“These are well-known cancer targets and they stood out from our PocketMiner predictions as very likely having cryptic pockets,” says Bowman. When they ran simulations of those proteins they found that the pockets do open up as predicted, heightening their focus for new drug treatments. “Because PocketMiner is trained  to find pockets that open pretty quickly, we think that what we have shown is just the lower bound of undruggable proteins that actually have cryptic pockets where small molecules can bind,” he adds. “These are targets people previously wouldn’t pursue with drug discovery.”

Also of Interest