Some important changes:

• Implements bond orders for CG using . = $ etc.
• Multiedges are dropped
• Bonds with order 0 are not resolved
• Nodes that only have zero-order bonds are not resolved

This should allow the capturing of virtual sides. The idea is that a VS has only non-zero order if it corresponds to a real fragment. In this case, it is a user's choice how many edges with zero order one provides on top of that. Any edges required to reconstruct the molecule are still mandatory.

Let's look at some examples:

• Glucose has one VS TC4 which sits in the middle of the triangle so it has 3 zero edges and no fragment.
  {[#SP4]1.2[#SP4].3[#SP1r]1.[#TC4]23}.{#SP4=OC[$]C[$]O,#SP1r=[$]OC[$]CO}

• Naphthalene has one VS in the middle of the ring TC5A here. We don't care because it has connectivity to all it's constructing neighboring atoms.
  {[#TC5]1[#TC5B][#TC5A]12[#TC5][#TC5B]2}.{#TC5B=[$]cc[>],#TC5=[$]cc[<],#TC5A=[>][>]cc[<][<]}

• Polythiophene has two VS that connect the rings at the CG level. The CG graph could either write this edge and all constructing edges or connect the VS only to one anchor. In any case, the connection between #C has to be provided in all cases. There is no way to recover the CG connectivity in a sensible way as it dos not map from the all-atom level. This is the best we can do.
  -{[#A]1[#B][#C]12.[#VS].[#VS].[#C]12[#B][#A]1}
  -{[#A]12[#B]3[#C]14.[#VS]23.[#VS]23.[#C]14[#B]2[#A]31}