IMP Manual
for IMP version 2.7.0
|
The IMP.rmf module acts as a link between RMF files and IMP data structures. Currently, three types of data are linked:
For each type of data, there is a set of functions declared in foo_io.h
and defined in foo_io.cpp
. These functions support four major operations
The links between IMP data structures are done via IMP::rmf::LoadLink and IMP::rmf::SaveLink. These are data structures associated with the RMF file (the RMF library provides a mechanism for this) that store persistent state that is available as long as the file is open. There is one load and one save link class per type of data.
atom::Hierarchy data is stored as RMF::REPRESENTATION nodes in the file.
There is exactly one node in the RMF file per IMP::Particle in the hierarchy. The link classes are implemented in terms of a bunch of helper classes, each of whichs handle a particular type of data.
The main tricky part is handling rigid bodies. In particular, the particles defining the rigid body are not always part of the IMP hierarchy, and older RMF files (where rigid bodies were written differently) must also be handled.
Rigid bodies that are not part of the IMP::atom::Hierarchy are handled by
When writing coordinates of things that are rigid bodies, the internal coordinates are written to the RMF rather than the global coordinates. This makes the files more compact. It works since the rigid bodies are written as RMF::decorator::ReferenceFrame nodes in the file, which effectively provides a transformation to the coordinates of things underneath it.
Restraints are stored as RMF::FEATURE nodes in the file.
Restraints are more or less write only, in contrast to the other types. There are create functions, but since the RMF doesn't contain enough data to actually create the correct IMP type, these just create dummy restraints whose score is simply the value stored in the file (and which don't have any derivatives). As a result, it is not clear that they can be used for much beyond verifying scores and some simple analysis.
Each IMP::Restraint added to the RMF is written as follows:
Mapping between the sub-restraints created by create_current_decomposition() and RMF nodes is slightly non-trivial. It is done based on the set up input particles for each restraint returned by create_current_decomposition(). For each such restraint, the link first checks in a map if there is already a node with that set of inputs, if so, it reuses that. If not, it creates a new one.
Geometry is stored as RMF::SHAPE nodes in the file.
The functionality provided in IMP::display::Geometry doesn't map entirely naturally on to RMF's geometry support. On the IMP side, things are implemented in terms of an ephemeral decomposition of complex geometric objects into simple ones (eg, a bounding box can be decomposed into edges, but each time you ask for the decomposition, you will get a different set of IMP::Objects). The RMF side expects a hierarchy whose structure stays constant across frames. As a result, each type of geometry object has to be special cased in IMP.rmf. Currently there is support for
For each piece of data stored in the RMF, a decision has to be made whether to store it once per file (using the RMF set_static_foo() methods), once per frame (using the RMF set_frame_foo() methods), or automatically (using RMF set_foo() methods). In IMP.rmf, in general things are stored once per file with the exception of coordinates/transformations of things that are not rigid parts of rigid bodies, which are stored once per frame. This could be changed to using automatic for almost everything, but that is
In particular, it might make sense for radii to become automatic.
One must be very careful not to store any RMF::NodeHandle or RMF::FileHandle objects in the adaptors as these will keep the RMF file open indefinitely (since the handles will keep the file alive and the file keeps the adaptor alive and the adaptor keeps the handle alive).