A year of atom-level DNA simulation puts mechanics alongside sequence

An international team including Dr. Daiva Petkevičiūtė-Gerlach of Kaunas University of Technology has published in Nature Communications an atomistic survey of DNA's shape, flexibility and dynamics, built by embedding all 4,096 possible six-nucleotide sequences into 190 longer fragments and simulating them ^{[5][6][11][12]}. The operational point is not the physics but the bookkeeping: if a base substitution changes how a stretch of DNA bends, it can change which proteins can reach it, and that is a functional consequence a sequence comparison alone does not record ^[8][10].

The premise, as Petkevičiūtė-Gerlach describes it, is that the nucleotide sequence fixes not only the information stored in DNA but the molecule's shape, flexibility and ability to move, so two segments can differ in how they bend, deform and interact with other molecules ^[7]. Proteins that regulate gene activity, she says, recognise three-dimensional structure and flexibility as well as the letters themselves ^[8]. Access is the constraint that makes this matter: DNA is wrapped around histone complexes into nucleosomes, which packs it into the nucleus but can restrict entry to particular regions ^[9]. Regions that bend or rearrange easily may be more accessible, stiffer or more tightly packed ones less so, which places mechanical properties in the same control layer as sequence for gene expression ^[10]. Work in this vein is described as reading DNA's "physical code" ^[4].

The cost of doing it at this level is the real news. Six-letter space is 4,096 sequences, which is exactly complete coverage at 4^6 ^[11][19], distributed at an average of roughly 22 hexamers per fragment across the 190 constructs ^[18]. The HexABC project ran molecular dynamics on them, a method that tracks the movement and interactions of every atom in the DNA and the surrounding solution ^[12][13]. That took hundreds of computers across Europe and the United States over several months, followed by more than a year to integrate and analyse the output ^[15][16]. Petkevičiūtė-Gerlach is explicit about the trade: simpler models can process millions of sequences in a day but are less detailed, and the team chose one of the most sophisticated atomistic approaches available instead, which made the calculations far slower and much more expensive in compute ^[14].

The evolutionary claim in the framing deserves scepticism proportional to its appeal. The published account carries the headline assertion that DNA's physical properties may shape which mutations evolution preserves ^[1], but it reports no conservation analysis, no mutation-frequency comparison and no selection test ^[20]. What the project appears to deliver is a reference set of mechanical descriptors for every hexamer, produced under a methodology that, according to Petkevičiūtė-Gerlach, experts across the field agreed to adopt in common, which she counts among the study's most important outcomes ^[17]. Anyone running a variant-scoring pipeline should read that as a candidate new input, not as a result to act on yet.

Two things to watch. First, whether the per-hexamer shape and flexibility parameters are released in a form that cheap sequence models can consume, since the fast methods that handle millions of sequences a day are the only ones that scale to variant screening ^[14]. Second, whether the claimed consensus on methodology survives contact with the next dataset ^[17].