in the folded state of amino acids that are originally distant in the
sequence. In folded conformations, these potentially dangerous
regions are buried and protected inside the tertiary and/or quater-
nary structure, conforming protein cores, establishing protein–pro-
tein interactions or assembling (macro)complexes [2, 32], in such a
way that the formation of non native intermolecular contacts is
minimized. However, in solution, globular proteins might experi-
ment misfolding and/or structural fluctuations that can transiently
expose these hydrophobic regions and trigger aggregation. In a
nutshell, the structural context and dynamics of a globular protein
requires a new set of algorithms that consider their three dimen-
sional scenario to accomplish reliable predictions of aggregation.
AGGRESCAN3D (A3D) [33] is one of the last generation
structural predictors conceived to overcome the limitations of lin-
ear predictors when forecasting the aggregation propensity of glob-
ular proteins. A3D is based on AGGRESCAN [34], a sequential
algorithm that relies on empirical data to build a scale of intrinsic
aggregation values for the 20 natural amino acids. Nevertheless,
A3D takes profit of the atomistic 3D coordinates of protein struc-
tures to correct the aggregation propensity of each amino acid
according to the neighboring structural context. Therefore, both
the residue exposure to solvent and the influence of other residues
in the structural vicinity are also computed for each amino acid
according to a multifactorial equation. Using the A3D server, the
identified aggregation-prone residues can be mutated in situ to
design variants with increased solubility, or to model the impact
of pathogenic mutations in disease-linked proteins. Additionally,
A3D server enables to take into account the dynamic fluctuations of
protein structures in solution, which may significantly modulate the
structural intrinsic aggregation propensity, by using fast simulations
of the protein backbone with the high-resolution coarse-grained
molecular modeling approach CABS-flex [35]. In this way, A3D
assembles in a single application the prediction and modeling of
different features that are highly relevant for the aggregation of
globular states; namely, the modulation of this propensity by the
structural context, the assessment of the impact of clinical or syn-
thetic mutations on solubility and the role of conformational fluc-
tuations. The A3D server can be accessed athttp://biocomp.chem.
uw.edu.pl/A3D/.2 Materials
2.1 Algorithm Inspired by AGGRESCAN, A3D benefits from the intrinsic aggre-
gation propensity scale obtained experimentally by De groot et al.
(2006) and allocates intrinsic aggregation tendencies (Agg) to each
single residue for a given structure. However, in this case,
Agg values are subsequently modified by the amino acid exposure
Predicting the Aggregation of Protein Structures 429