Predicting the locations of cryptic pockets from single protein structures using the PocketMiner graph neural network

Abstract

Cryptic pockets expand the scope of drug discovery by enabling targeting of proteins currently considered undruggable because they lack pockets in their ground state structures. However, identifying cryptic pockets is labor-intensive and slow. The ability to accurately and rapidly predict if and where cryptic pockets are likely to form from a protein structure would greatly accelerate the search for druggable pockets. Here, we present PocketMiner, a graph neural network trained to predict where pockets are likely to open in molecular dynamics simulations. Applying PocketMiner to single structures from a newly-curated dataset of 39 experimentally-confirmed cryptic pockets demonstrates that it accurately identifies cryptic pockets (ROC-AUC: 0.87) >1,000-fold faster than existing methods. We apply PocketMiner across the human proteome and show that predicted pockets open in simulations, suggesting that over half of proteins thought to lack pockets based on available structures are likely to contain cryptic pockets, vastly expanding the druggable proteome.

Accelerating cryptic pocket discovery using AlphaFold

“…Interestingly, AlphaFold generates open states of the Niemann-Pick C2 Protein that were not discovered in 2 microseconds of adaptive sampling simulations (Fig. 2B) 1 . However, AlphaFold’s ensemble of TEM β-lactamase structures does not include any open states where the Horn 39 or omega 6 pockets are open (Fig.…”

“…This result makes PM-II an exception to a general trend. We have found that many cryptic pockets can be discovered with a handful of simulations of intermediate length (i.e., 40 ns) 1 . Furthermore, significant progress has been made in developing algorithms for cryptic pocket discovery, including Markov State Models 6,41 , enhanced sampling strategies like SWISH 42 , or adaptive sampling approaches like FAST 43 .…”

“…Cryptic pockets, or pockets absent in ligand-free experimental structures, are a promising means to expand the scope of drug discovery. By one estimate, almost half of all structured domains lack obvious pockets in their experimental structures 1 . These proteins have often been considered ‘undruggable’.…”

“…Molecular dynamics simulations can reveal excited states with cryptic pockets that can then be used for structure-based drug design 2,6 . However, in certain cases, cryptic pockets may not be discovered by simulations because cryptic pocket opening motions may be slow (e.g., Niemann-Pick C2 Protein in Meller et al 1 ). Two classes of slow motions include sidechain ring flipping 7 events and secondary structure rearrangements 8 which can both occur on microsecond and slower timescales.…”

EquiFold: Protein Structure Prediction with a Novel Coarse-Grained Structure Representation

“…Recent studies using deep neural networks to predict protein structure [1,2,3,4,5] have accelerated the development of structure-based methods for protein property prediction and design. However, some models tend to predict one or few conformations that may not be optimal for properties of interest, such as ligand-binding pockets [6] and protein-protein interfaces [7,8,9]. Moreover, these models require either multiple sequence alignments (MSA), protein language model embeddings, or other statistical information distilled from large sequence databases.…”