In my opinion this formula is the best one:
\begin{equation}
\sin\left[\sum_{j=1}^N a_j\right] = \sum_{\sigma_{1,\dots,N-1}=\pm} \left[\prod_{j=1}^{N} \sin\left[a_j + \left(2 + \sigma_j-\sigma_{j-1} -\delta_{j1}(1-\sigma_{j-1}) - \delta_{jN}(1+\sigma_{j})\right)\frac{\pi}4\right]\right]
\end{equation}
The $\delta$'s are Kronecker deltas.
I hope it will be of use to someone; I used it to write the sine as a product of eigenvalues = a matrix. (This last fact is not relevant to the question, but a useful consequence of the above decomposition, i.e. any symbolic product can be written as a matrix with the individual factors of the product along the diagonal. The above produces a signed sum of these matrices, which gave the decomposition as given above some other application.)
EDIT:
Applied to the case N=5, the formula predicts
\begin{align}
\sin\left[\sum_1^5 a_j\right] = \sum_{\sigma_1,\sigma_2,\sigma_3,\sigma_4=\pm} \sin\left[a_1+(1+\sigma_1)\frac\pi4\right]\sin\left[a_2+(2+\sigma_2-\sigma_1)\frac\pi4\right]\sin\left[a_3+(2+\sigma_3-\sigma_2)\frac\pi4\right]\sin\left[a_4+(2+\sigma_4-\sigma_3)\frac\pi4\right]\sin\left[a_5+(1-\sigma_4)\frac\pi4\right]
\end{align}
which can be verified by using the Mathematica code
pmlist = {-1, 1};
n = 5;
sigma\[LetterSpace]list\[LetterSpace]symb =
Reap[Do[Sow[
ToExpression["\[Sigma]\[LetterSpace]" <> ToString[l]]], {l,
Range[n]}]][[2]][[1]];
sigma\[LetterSpace]list\[LetterSpace]symb\[LetterSpace]appended = \
{sigma\[LetterSpace]list\[LetterSpace]symb[[#]], pmlist} & /@
Range[n - 1]
Sin[Sum[Subscript[a, i], {i, 1, n}]] -
Fold[Sum[#1, #2] &,
Product[Sin[
Subscript[a,
j] + (\[Pi]/4)*(2 +
ToExpression["\[Sigma]\[LetterSpace]" <> ToString[j]] -
ToExpression["\[Sigma]\[LetterSpace]" <> ToString[j - 1]] -
KroneckerDelta[j,
1]*(1 -
ToExpression[
"\[Sigma]\[LetterSpace]" <> ToString[j - 1]]) -
KroneckerDelta[j,
n] (1 +
ToExpression[
"\[Sigma]\[LetterSpace]" <> ToString[j]]))], {j, 1, n}],
sigma\[LetterSpace]list\[LetterSpace]symb\[LetterSpace]appended] // \
TrigExpand