I only just noticed this question, so maybe it's too late, but here's an
answer.
Note that some symplectic manifolds (like $\mathbf{CP}^2$) contain no Lagrangian spheres, so
this complex is then empty. Let's ignore that and focus on the other
examples.
For certain simple symplectic 4-manifolds (small blow-ups of the
projective plane or a quadric) this complex should be known explicitly
(I can tell you the vertices and tell you what you need to check to
get the edges). Let $X_k$ be the $k$-point blow-up of
$\mathbf{CP}^2$ with its monotone symplectic form.
- e.g. $S^2\times S^2$ with its monotone symplectic form has
precisely one Lagrangian sphere up to Hamiltonian isotopy (due to
Hind https://arxiv.org/abs/math/0311092), so your complex is just a
point. Same for $X_2$ (https://arxiv.org/abs/0902.0540).
- e.g. For $X_3$ and $X_4$, Lagrangian spheres are uniquely
determined up to isotopy by their homology classes
(https://arxiv.org/abs/0902.0540 again), which gives:
- $X_3$: the classes $L_{12}=E_1-E_2$, $L_{13}=E_1-E_3$,
$L_{23}=E_2-E_3$, $L_{123}=H-E_1-E_2-E_3$ (I'm counting the
two orientations of $L$ as the same sphere, so $[L]$ and
$-[L]$ give the same sphere). Just for homology reasons,
$L_{ij}$ must intersect $L_{jk}$ for $i\neq k$. The only
other thing to check is whether $L_{123}\cap L_{ij}$ can be made
empty (no homological obstruction). If so then your complex is a
$D_4$ Dynkin graph: otherwise it's a disconnected set of 4
points (I didn't think about which case holds, sorry).
- $X_4$: similarly we get $L_{ij}=E_i-E_j$ and
$L_{ijk}=H-E_i-E_j-E_k$. The possible edges are $L_{ij}$ to
$L_{k\ell}$ when $\{i,j\}\cap\{k,\ell\}=\emptyset$ and
$L_{ijk}$ to $L_{ij}$. If all these edges exist then your
complex is obtained from a tetrahedron by putting $L_{ijk}$ at
the four vertices and $L_{ij}$ on the edge between $L_{ijk}$
and $L_{ij\ell}$, then connecting opposite edges.
- e.g. for $X_5$ you have similar homological data, but you can also
focus on a single homology class and get infinitely many knotted
spheres (as observed by Seidel
https://arxiv.org/abs/math/0309012). These are nonetheless
classified (Borman-Li-Wu, https://arxiv.org/abs/1211.5952): the
symplectic Torelli group acts transitively on them, and in this case
the symplectic Torelli group is the pure spherical 5-strand braid
group (https://arxiv.org/abs/0909.5622). I haven't thought about how
the Lagrangian "curve complex" would work in this example, but my
guess is it would be closely related to the curve complex on the
punctured sphere, in the following way. All the spheres are
vanishing cycles for algebraic degenerations of $X_5$ coming from
moving the 5 blow-up points. As three blow-up points become
collinear, the blow-up develops a holomorphic $-2$-curve (proper
transform of the line through these three points). If you take the
anticanonical model then you get a nodal variety (and the Lagrangian
sphere is the vanishing cycle of this node). To get 5 points on
$S^2$ from this, look at the conic through the 5 blow-up
points. You get a holomorphic sphere with 5 marked points. As the
node forms, two of these marked points collide, and you can
associate to the Lagrangian $S^2$ the simple closed curve which
encircles these two marked points. If you can form two nodes
simultaneously then these pairs of points collide simultaneously so
the simple closed curves will be disjoint (as will the vanishing
cycles).
In higher dimensions, little is known. However, by focusing attention
on your favourite collection of spheres and computing Floer cohomology
you should be able to figure out subcomplexes. Perhaps a more
natural/manageable/useful construction here replaces vertices of the
complex with quasiequivalence classes of objects (maybe spherical objects if you want to specify that) in the Fukaya category, joined by an edge when their Floer
cohomology vanishes. This still admits an action of the symplectic
mapping class group (though factoring through the autoequivalence group of the Fukaya category) which is what you'd probably want to use this complex for anyway.
