Hence, this activity is growing but currently is not a major focus for
the pharma industry in general.
The authors recognize that for computational simulation to
impact mainstream drug discovery it has to offer improved gener-
ality and speed. Computational simulation is not currently applica-
ble to all targets of interest and can be extremely time consuming.
The authors are acutely aware that the traditional pharma approach
of high throughput screening has been successful and continues to
deliver candidate molecules and successful drugs. Computational
design is expensive, and a great deal of screening can be done for
equivalent cost. While the authors believe that computational
approaches will at least complement and eventually supersede tra-
ditional drug discovery approaches, particularly in difficult areas
such as targeting protein–protein interactions with small molecule
drugs, we are at the early stages of this transition. While intellectu-
ally challenging and occasionally satisfying, computational
approaches need to be mindful of how successful drugs have been
discovered in the past, and respectful of this heritage in order to
truly impact drug discovery in real time and to influence the design
of candidate molecules.6 Notes
- Apart from a small protein system (e.g., 2–3 residues), with
current computing infrastructure it is often not possible to
successfully calculate free-energies in more than a few
dimensions. - For some large higher dimensional spaces, optimal projection
descriptors for the energy surface (or collective variables) give
useful information but at a much lower resolution (i.e., many
details are missed) than what could be achieved in lower dimen-
sions [117]. This issue of optimal collective variables or reac-
tion coordinates is important for efficient sampling of protein
motions and understanding of protein simulations. - With current computing infrastructure, it is often difficult to
ensure all important conformations are explored by MD, espe-
cially those separated by large energy barriers [166, 167]. - QM resolution protein simulations are possible although, even
at low levels of theory, it is prohibitively computationally slow.
A relatively recent GPU based simulation study using BOMD
produced an 8.8 ps trajectory [168]. - These atomistic MM force fields are designed to capture the
most important interactions while being computationally inex-
pensive; hence the use of harmonic potentials rather than more
realistic options such as the Morse potential.
356 Benjamin P. Cossins et al.