Changes since release 9.10

• sequence_db objects now only store part of the sequence database in memory when using a binary (HDF5) file (databases in PIR or FASTA format still must be read in in their entirety). Functions that use sequence databases now take a window_size argument, that controls how much of the database is read into memory at one time. This results in much lower memory usage when using large databases such as UniProt.

• Limits on the numbers of atoms and bonds per residue have been removed.

• SSBOND and CONECT records are now written to PDB files when models containing disulfides are saved.

• The default behavior of automatically converting common modified residues (MSE, MEX, ABU) to their standard equivalents can now be disabled by setting env.io.convert_modres=False.

• Rather than silently truncating align codes to the internal maximum length (of 40 characters), too-long codes now result in an error.

• alignment.get_suboptimals() now takes an align_block argument to support suboptimal alignments of two previously-aligned blocks of sequences.

• Bundled version of HDF5 updated to 1.8.8.

• Bugfix: Chain.join() now works correctly when joining adjacent chains.

• Bugfix: when converting MSE residues to MET, set the element correctly for the SD atom (to S; previously it was left as SE).

• Bugfix: improve matching of alignment sequence to PDB when using '.' in the alignment header and PDB contains duplicate residue numbers.

• Bugfix: parallel loopmodel runs no longer leak memory.

• Bugfix: alignment.salign() no longer leaks memory on encountering an error.

• Bugfix: on machines without networking, allow parallel runs with local_slaves only (rather than failing to create a job object at all).