Minutes 27/01/06

• Fine tuning parameters for experiments
• Reformulating the computational problem for the case of unequal content
• Writing up the report

• Results for comparing topological accuracy of mgr, mgr_2, mgr_uc and mgr_uc_2
• Results for comparing the accuracy of the recovered root for mgr, mgr_uc and mgr_uc_2

Results of the comparison of the percentage of correct edges shows that

• when k = 3, all method achieve high accuracy. When the percentage of deletions increase the accurary of mgr_uc and mgr_uc_2 remain high which that of mgr decreases
• mgr_uc_2 performs as well as mgr for the case of 0% deletions and it out performs mgr when the percentage of deletions is increases.
• mgr_uc_2 generally performs better than mgr_uc

Results of the comparison of the accurary of the recovered scenario by comparing the normalized reversal distance between the recovered root and the identity

• The normalized distance increases significantly when k is increased
• mgr_uc generally performs better than mgr_uc_2
• both mgr_uc and mgr_uc_2 performs better than mgr

mgr_uc performs better in terms of the recovered scenario while mgr_uc_2 performs better in terms of the percentage of correct edges. Therefore one way to combine both approaches is to set a threshold value for mgr_uc, when the score of the best rearrangement falls below the threshold, switch to the method of mgr_uc_2. Prof Leong suggested that we can plot a graph of the score of the best rearrangement and use it to determine a good threshold value.

Prof Leong suggest that we compute the number of rearrangements determined during the “good” rearrangements phase as well as the number of rearrangments determined during the iterative median phase of the MGR algorithm. This will be able to help us understand the behaviour of the MGR algorithm.

Prof Leong pointed out that the iterative median method used in the MGR algorithm is very similar to the NJ method.

Instead of using a fixed k for the number of rearrangement/edge in the model tree, perhaps this can be varied

None

• Threshold value for mgr_uc to mgr_uc_2