When a structure is registered, 3decision runs fpocket (paper) to find its pockets. fpocket fills the empty spaces at the protein surface with alpha spheres, groups them into pockets, and calculates descriptors for each pocket from the spheres and the atoms they touch. 3decision also adds the pockets defined by bound ligands. So a structure has two kinds of pockets:
These descriptors are the columns of the pocket table.
The druggability score, from 0 to 1, estimates how likely the pocket is to bind a small drug-like molecule. It is computed from fpocket descriptors with a logistic function, first described in Understanding and Predicting Druggability and retrained since.
A low score doesn't mean that nothing binds. A peptide binding site binds peptides, for example, but peptides aren't drug-like.
A drug-like ligand here has a molecular weight between 150 and 2000 Da, at least one ring, and fewer than 95% single bonds.
Each residue is counted once.
The flexibility score, from 0 to 1, is based on the mean B-factor of the pocket atoms, normalized against the other pockets of the same protein. In crystal structures, B-factors don't only reflect flexibility, so use this score with care.