Proteins are built from sequences of amino acids coded by DNA sequences. Over the generations these DNA sequences aren't inherited 100% perfectly but may
occasionally mutate. Given the DNA sequences from many different organisms alive today it is then feasible to ask "What is the most likely amino acid sequence that
this particular organism might have had given the sequences its currently living descendants and the descendants of its ancestors might have had?"
The researchers chose to work with a form of the protein Rhodopsin. This is the light sensitive pigment protein at work in the rods of the retina and is specialized for
high sensitivity in low light level. They took the sequences for 30 organisms alive today (including humans, lampreys, chickens and goldfish) and aligned them using a package called
ClustalW. This process is somewhat akin to performing diffs between a set of files (or an agrep between strings). Protein sequences
are like software in that there are different portions of the sequence that perform different functions. Alignment attempts to line up each of the different sequences in such a way that
the regions that perform a similar function (and so have a similar amino acid sequence) line up together. Alignment algorithms are typically use a dynamic programming approach to find the minimum
number of changes (usually weighted in some way) to convert one sequence into another by changing individual amino acids or inserting/deleting subsequences.
Once the sequences were aligned it was then feasible to estimate the sequences that a particular ancestor had. For example if in an alignment we found that the same amino acid always occurred in the
same place in every sequence we can be almost certain that it was present in the archosaur sequence. If on the other hand a particular amino acid occurred in many close relatives of the archosaurs but not more distant ones
we would be confident, but not certain, that it was the one that appeared in the archosaur. By having a model for the probability of any particular mutation we can determine the sequence
with the highest likelihood. This was carried out using a package called PAML.
Once the most likely sequences were determined the protein was synthesized de novo (ie. from scratch). It could then be tested for light sensitivity. Unlike proteins that trigger behaviors like nesting or control
development pathways like homeobox genes, rhodopsin can be tested directly in the laboratory for light sensitivity giving an immediate determination of the light sensitivity of
archosaur vision. The results of testing indicated that these proteins were similar to modern bovine rhodopsin although their light sensitivity was red-shifted. The precise spectral response suggests
that archosaurs had eyes adapted to vision in low light levels and may in fact have been nocturnal.
This experiment has opened up an exciting new way to investigate the physiology and behavior of prehistoric organisms such as dinosaurs. Armed with a computer and a sequence database researchers can now probe our genetic ancestry in a practical way without relying on fossils. And it allows us to determine organism features that couldn't possibly be determined from fossils.
 BBC Science News (http://news.bbc.co.uk/1/hi/sci/tech/2237114.stm).
 Recreating a Functional Ancestral Archosaur Visual Pigment, Chang, Jonsson, Kazmi, Donoghue and Sakhmar, Mol. Biol. Evol. 19(9):1483-1489, 2002