The Problem
Discovering
drugs still involves a lot of real experiments on live cells. Automated and
simulated models of cells could allow speeding up this process. Simulations
require designing models, which represent real structures to predict their
behaviour without doing real experiments and requiring more resources.
If
we consider a cell as an object which we are trying to simulate as precisely as
possible, we might need to include physics and chemistry and even quantum
mechanics of each atom. Full simulation even of just one cell is still beyond
what computers can do effectively today and that’s why biology, physics,
chemistry and computing experts try to find the best balance of how far the
problem can be abstracted, but still keep predictions in satisfactory precision
and still return results within resource constraints. Simulating just certain
function would reduce the problem, but could lose precision, because of
undiscovered relations to other functions. There are some examples of quite
abstract solutions in machine learning, like neural networks, which simulate
brain function of recognizing patterns, without the need to simulate brains to
the atomic level and even taking more space than just couple paragraphs of
code. But even if this example lets us know that it is possible to make quite
big abstractions to work, in drug development it might not be the case. Drugs
are also made of molecules, which have to be exactly defined to be able to
produce them. That causes some difficulties when trying to abstract molecular
processes to get answers about precise molecules. It is difficult to abstract
something and get out precise answers, but it is possible to increase precision
by gradually abstracting less and less.
“Computational modeling of the structure of
protein-peptide interactions is usually divided into two stages: prediction of
the binding site at a protein receptor surface, and then docking (and modeling)
the peptide structure into the known binding site” (M. Blaszczyk et al., 2015,
p1). The “gap between identified hits and
the many criteria that must be fulfilled, can be bridged by investigating the
interactions between the ligands and their receptors“(V. Lounnas et al.,
2013, p1). Finding the point of abstraction at which simulated cell or cell’s
function behaves acceptably close to reality is another interesting and
complicated problem. Experimenting at different levels of abstraction could
allow finding the optimal solution for further development of these
simulations. “With faster and more powerful computers larger and more
complex systems may be explored using computer modelling or computer
simulations” (J. Meller, 2001,
p1). Work with cell receptor and ligand interactions have examples both
in real biochemistry experiments and machine learning. The most unambiguous and
precise language for modelling is mathematics and “mathematical modelling typically requires deep mathematical or
computing knowledge, and this limits the spread of modelling tools among
biologists” (R. Gostner et al., 2014, p16).
The
knowledge gap between experts and beginners keeps increasing, because their
time on actual research is more valuable than time spent explaining their work.
Advancing science caused more narrow fields to arise and as a result it keeps
increasing the knowledge gap even between experts. The need for experts and
guides on their created technologies arise and it could be solved by gradually
educating about the general topic or parts of it and its problems. There are
experts who are doing actual experiments and search for real solutions, but
there are not that many intermediate tools or solutions which introduce to the
bioinformatics by showing how it works and what are its tools.
Proposed solution
Partial
solution could be to advance research about these kind of problems and educate
more people about bioinformatics. Proposed solution offers to educate about the
main topics of bioinformatics and one of the best machine learning techniques-
SVM. The implementation expands this guide in to Java implementation of SVM interpretation called Least Similar Spheres (LSS).
No comments:
Post a Comment