this is for holding javascript data
Antonino Ingargiola edited Concepts.tex
about 8 years ago
Commit id: 297c251d73a0e2da551d74809ad455a8984d173f
deletions | additions
diff --git a/Concepts.tex b/Concepts.tex
index e9343e1..e19f3f1 100644
--- a/Concepts.tex
+++ b/Concepts.tex
...
different photon detection efficiencies of donor and acceptor channels.
A simple way to mitigate the dependence of the FRET distribution on
the burst size selection threshold is weighting bursts
according proportionally to their size
(i.e. their
information content) Fisher information)
so that the bursts with largest sizes will have the largest weights.
In statistics, weights proportional to the inverse of the variance are used
in the context of least-squares fitting to take into account samples
with different variances. In the case of a burst population with fixed FRET efficiency $E_p$,
bursts $n_a$ counts are commonly described by binomial random variable with $n_d + n_a$ trials
and $E_p$ probability of success. While this model neglects the effect of background, it captures
the main features of a static
FRET population. Under the binomial assumption,
the variance of each burst $E_i$ is
inversely proportional to the burst size $n_{ti}$,
as reported in eq.~\ref{eq:var_e}.
$$\operatorname{Var}\left( \begin{equation}
\label{eq:var_e}
\operatorname{Var}\left( E_i \right) =
\operatorname{Var}\left( \frac{n_{ai}}{n_{ti}} \right)
=
\frac{E_p(1-E_p)}{n_{ti}}$$ \frac{E_p(1-E_p)}{n_{ti}}
\end{equation}
The Therefore weights proportional to $n_{ti}$ are the natural choice (see SI XXX).
In general, this weighting
scheme is used for building efficient estimators for a population
parameter (e.g. $E_p$). But it can be
also used to build weighted histograms or Kernel Density
Estimation (KDE)
plots. When plots which will exhibit FRET subpopulations with optimal width,
yielding more accurate fit of peaks positions and better resolving nearby peaks
(compared to corresponding non-weighted plots using the same burst-size threshold).
Traditionally, without using weights,
for optimal results, the
choice width of
a particular
burst FRET subpopulations peaks are manually optimized by finding an ad-hoc (high)
size-threshold which selects only bursts with the highest size
threshold affects (and thus lowest variance).
This procedure, needs to balance reduction in fitting error due to using bigger bursts
and the increase in fitting error due to the
shape lower number of bursts.
Conversely, the use of size weights allows a more efficient use of
burst information.
For example, by fixing the burst
distribution size threshold to a
lesser extent,
therefore lower threshold low value (e.g. 20 photons)
can be used in a wide variety of cases
without broadening the is possible
to obtain optimal-width FRET sub-populations peaks
of sub-populations, effectively removing the without any need to
an ad-hoc optimization of the burst size threshold search
for
each measurement. an optimal burst-size threshold.
\paragraph{Python details}
FRETBursts has the option to weight bursts using γ-corrected
...
\verb|weights='size'| to histogram or KDE plot functions. The \verb|weights|
keyword can be also passed to fitting functions in order to fit
the weighted E or S distributions (see section~\ref{sec:fretfit}).
Several other Other weighting functions (for example
quadratical) depending quadratically onf the size)
are listed in the \verb|fret_fit.get_weights| documentation
(\href{http://fretbursts.readthedocs.org/en/latest/fret_fit.html#fretbursts.fret_fit.get_weights}{link}).
However, using weights different from the size is not recommended
due to their less efficient use of burst information.
\subsection{Plotting \texttt{Data}}
\label{sec:plotting}