Python Programming for Biology: Bioinformatics and Beyond

Protein structure prediction

We  cannot  yet  automatically  determine  the  three-dimensional  structure  of  a  protein  just

from its amino acid sequences,

6

but looking at the variations and conservation in protein

structures  as  a  whole  it  is  apparent  that  structural  features  are  more  preserved  than  the

protein  sequences.  Sequences  may  vary  considerably  between  proteins  (that  really  are

related)  and  yet  have  very  similar  three-dimensional  structures;  so  much  so  that  we  are

often unable to produce a convincing alignment using the residue sequence alone. Hence,

if we actually can make a good sequence alignment between two protein sequences then

we  can  generally  be  confident  that  their  structures  are  similar,  i.e.  structure  is  better

preserved  than  sequence.  In  effect  we  are  saying  that  the  two  proteins  had  a  common

ancestor  that  was  structurally  similar  to  both.  This  observation  of  structural  preservation

being better than our ability to detect related sequences is at the heart of a process called

comparative  modelling.  Comparative  modelling  allows  us  to  make  a  good  guess  at  the

structure of one protein if we already know the structure of a related one, and as you might

expect the closer the relationship between the two proteins (evolutionarily speaking) then

the  better  our  guess.  This  structural  guess  is  usually  called  a  model.  In  essence,  an

alignment is made with our query sequence and various potential template sequences that

have a known structure. If the degree of sequence similarity is suitably large then we may

attempt  to  build  a  model  of  our  query  protein  structure  using  information  from  the

template.  Such  information  might  include  the  position  of  sub-structures  like  helices  and

extended strands

7

or the distances and angles between atoms.