The Genom of Homo sapiens.pdf

TRANSCRIPTIONAL UNITS AND GENE PAIRS 467Table 4. Human Chr. 22 Antisense Pairs in Which Both Members of the Pair Are Expressed in the Same Human Tissue andGenomic Organization Differs in the MouseType of genomic organization Pairs Genes and TUs in antisense pairs Expression of both members of pairhuman mouse 9 FLJ32500 BID kidney; skinHIRA tu_BF589683 skin (cancer)MIF BC036909 brainADORA2A tu_AI198582 testisPISD tu_BI915399 brainTIMP3 tu_BM990787 trabecular bone; placenta; uterusUNC84B tu_AW504307 B-cellsUNC84B tu_BG057310 colon (tumor)ACR BC050343 testishuman mouse 4 RFPL1 RFPLL1ANT brainneither RFPL3 RFPL3ANT brainFLJ30933 tu_AA844700 testistu_AA383102 tu_BU585030 testishuman mouse 2 FLJ30119 SNRPD3 skin (cancer)SF3AI BC018040 brainExon skip in mouse relative to human 1 AL365514 HSC3 testisa chance combination of two independent events (p =0.01 by χ 2 ). Therefore, for a given expressed feature, participationin one UGP type increases the probability thatthe feature is also involved in the other.It is notable that more than half of our antisense data setdoes not overlap the antisense pairs identified in the independentand parallel study by Yelin et al. (2003), whichwas performed while the present work was in preparation,on 5q31 and chr. 22. At 5q31, 9 of the 13 pairs identifiedby us outside the PCDH clusters, along with 1 of the8 pairs within the PCDH clusters, were also identified byYelin et al. (48%). On chr. 22, 34 of our 77 pairs werealso identified by Yelin et al. (44%). This suggests thatour manual and automated approaches, respectively, arecapable of identifying UGPs missed by independently designedalgorithms.UGP Distribution along Chr. 22On the basis of the 5q31 results, we hypothesized thatUGPs on chr. 22 would cluster, rather than be randomlydistributed along the chromosome or distributed proportionalto gene density. We operationally defined a “UGPisland” as a region with at least two UGPs of the sametype, and in which any two consecutive UGPs are 2 independentantisense partners were not allowed. For eachinterval, 10,000 simulations were performed. The simulationswere implemented with a Perl script we designedspecifically for this purpose. The actual biological complexityof an interval was defined as the number of UGPsof both types, and the number of islands of UGPs of eithertype, within the interval. The proportion of total simulationsper interval in which the actual biological complexityof that interval was met or exceeded wastabulated.Simulation results suggest that nonoverlapping intervalsof chr. 22 containing ~50 transcript models per intervalcan be assigned to one of two types. In intervals ofthe first type, which contain most UGP islands, the actualbiological complexity is far greater than that expected bychance, as it is matched or exceeded in