Information contents in areas of ambiguous alignment, departure from randomness and resistance to noise addition – a case study of the trnL intron in Colletieae (Rhamnaceae)

AAGESEN, LONE

Mendoza, Argentina

Jornada; III Reunión Argentina de Cladística y Biogeografía; 2001

CRICYT

Noncoding sequences (introns and intergeneric spacers) have been thought of as more variable than coding regions due to fewer constraints and consequently appropriate for analyses at lower taxonomic level. Fast evolving portions of genomes do, however, often vary in length due to insertions and deletions events, and this divergence may cause problems during alignment. Including areas of ambiguous alignment in a phylogenetic analysis may introduce errors due to the uncertain alignment. With ‘optimization alignment’, Wheeler offered a method to analyse DNA sequence without the intervening step of multiple alignment. The procedure can be combined with sensitivity analysis were a wide range of alignment parameters are examined. In an analysis of the trnL intron at species level within the tribe Colletieae (Rhamnaceae) and area of ambiguous alignment was detected. The remaining part of the sequence contained almost no information at species level. When the ambiguously aligned area was included and submitted to a sensitivity analysis using optimization alignment, five monophyletic groups were found in all 20 parameter sets examined. Though a single signal was detected it is unclear whether a similar signal could be obtained if the information in the area of ambiguous alignment area was purely random. First, how would a randomly ordered signal inserted into a real matrix perform in a sensitivity analysis? Second, how strong is the signal, is it resistant to random added noise or does it disappear when low levels of noise are added to the matrix? Third, would an analysis conducted at a higher taxonomic level, where random mutations have blurred the original signal, be weaker or more sensible to noise, or more similar to a random pattern? The original matrix was cut into three submatrices, tow including nonambiguously aligned areas flanking the third ambiguously aligned part, and the three submatrices were analysed together. Noise was added to the ambiguously aligned part (matrix A) bay reassigning character states to randomly picked cells with the probabilities of assigning a given base or a gap determined by the base and gap composition I the original matrix A. The new matrix subsequently replaced the original matrix A and the three submatrices were reanalysed. Noise level of 10%, 25%, and 100% replacement were explored with 25 repetitions at each noise level. Number of groups appearing in all 20 parameters sets to groups appearing in only a single parameter set were used as comparative statistic to evaluate the results. The original signal departs from the one found in a random submatrix mainly by having a higher number of groups occurring in all parameters sets and notably fewer groups occurring in only a single parameter set. Total amount of groups is also lower in the original matrix. When adding 10% noise to the matrix only a single group appears in the entire parameter space in all 25 repetitions. When 25% noise is added none of the original groups are found under all parameter sets in all 25 repetitions. Preliminary results suggest that when sampling at higher taxonomic level within the family resolution is lost when the ambiguously aligned area is included in the analysis. However, as the flanking areas (the non-ambiguously aligned areas) include more phylogenetic informative sites at this taxonomic level the effect of including a completely random matrix A is less disturbing for the final results.