Genomic code for nucleosome positioning?

Over the past few years, the research on nucleosome positioning has been a hot topic: Segal et al., through a series of high-profile publications, have proposed that "the intrinsic nucleosome sequence preferences contribute substantially to nucleosome organization and chromatin function in vivo" (i.e., there exists a "genomic code") and developed bioinformatics tools to predict patterns of nucleosome positions based on genomic DNA sequence. However, as is clear from the title of a paper by Zhang et al., "Intrinsic histone-DNA interactions are not the major determinant of nucleosome positions in vivo", it is highly debatable if the nucleosome sequence preferences is strong enough to be called a "code", in the canonical sense as for the genetic code.

While I am (generally) aware of the controversy about the "genomic code", my understanding of the issue has become much clearer after reading the correspondences between both sides published in the 17(8), August 2010 issue of Nature Structural & Molecular Biology [NSMB]:

• "Nucleosome sequence preferences influence in vivo nucleosome organization" (pp918 - 920) by Kaplan … and Segal
• "Evidence against a genomic code for nucleosome positioning" Reply to “Nucleosome sequence preferences influence in vivo nucleosome organization” (pp920 - 923) by Zhang … and Struhl
• "A preoccupied position on nucleosomes" (p923) by Pugh

Reading carefully through the text, it is clear how important the little details could be: 20-bp vs 40-bp windows? How significant the preference should be to be claimed as a "code"? Pugh clarifies another source of confusion, in the terminology of nucleosome ‘occupancy’ vs ‘positioning’. In my understanding, the dispute on such details itself speaks strongly against the existence of "genomic code".

It should be noted that early work by Drew & Travers (JMB85) and Satchwell et al. (JMB86) provided evidence that the rotational positioning of DNA about the histone octamer is regulated by DNA sequence-dependent bending preferences. Goodsell & Dickerson quantified the Satchwell trinucleotide model by assigning a set of roll angles to predict DNA bending from base sequence. This bending model, along with five others, was implemented by SCHNArP for building approximate 3D DNA structures (with fixed values for slide, twist etc other parameters). However, this functionality is dropped out from 3DNA, simply because of the arbitrariness/subtleties in picking up the parameters.