Constantine Evans from Eric Winfree’s lab talked about theoretical and experimental tile assembly results. The optimal use of the sequences of the sticky ends was important. There are only so many short sequences that are appropriate for sticky ends. You run out of unstructured (linear) combinations of ATGC if you can only have an 8-10 base sequence. If you had infinite, specific partner sequences, you could program tiles to assemble anything, deterministically, just by making each tile a new pixel that only binds its neighbors in an array. But we do not have infinite sequence space. That is why biology does not specify “cell X, Y, Z becomes muscle. Cell X+1, Y+1, Z+1 becomes bone.” In biological development, a cell follows a contingency tree according to the genome and the local stimulus to generate a complex shape. Self assembly might be approached the same way.

Tiles can do that. They can follow simple rules like cellular automata. But errors also propagate. How does one avoid locking in an error? Some sequences are better than others. Different pairs have different sensitivities to sequence issues. This can be used to derive rules for what sticky end pairs are best in a given design. Basically, this is careful analysis of sequence binding and misbinding energy for every sticky end in the system. Careful analysis of misbinding off-rates gives a much better simulation of errors and allows the designer to avoid them.

Next, Lulu Qian presented a method for compiling a reasonably complex programming language into DNA. Her point is to integrate robust molecular components in systems, then develop languages, then to synthetic molecular programs (i.e. life). Current state of the art is 74 circuits as a square root calculator. See-saw gates are Lulu’s specialty. The gist is that it is a reversible reaction that propagates to another reversible reaction. That can be made more robust than most programmable DNA reactions. This system is sufficiently orthogonal and robust that it can be abstracted and arranged like a circuit diagram. Then a computer can make up the DNA sequences to “compile” the code.

She also talked about new work in surface-immobilized DNA Turing machines. If you can get key parts of the DNA-DNA reaction immobilized on a surface (but keep the fuel and other auxiliary strands up in solution) it can make an immobilized state machine. I imagine that is a building block of a more complete Turing machine. She used two 4-way branch migration reactions to convert the state of surface-immobile species A to surface-immobile species B in the same orientation. It was very complicated had a 5 hour reaction time, but she experimentally validated that it worked. This can make complex spatial-temporal reactions like the Game of Life.

Niranjan Srinivas (from Winfree Lab) talked about some careful kinetic analysis of strand displacement reactions. These are the core of the reactions the Lulu Quian has made so much progress with. Kinetic and probabilistic analysis of the strand displacement reaction (with some reasonable assumptions about energetics) are 2-3 orders of magnitude off. Why? The first thought was that it was due to simplified thermodynamics (no secondary structure, average base pair energetics). That did not fit the data. After some work, they seem to have reached the conclusion that the slow step is the initiation of branch migration. Naively, one might expect the energetics of initiating branch migration are the same as the energetics of unbinding one base, but that does not seem to be the case. Could this be used to improve the kinetics? Could we use weak AT or mismatched bases on the ends of the branch migration region?

Matthew Lakin gave a talk about analyzing reaction networks. I gather that massively parallel computer systems have interesting comparisons to DNA. It is neat to seem people from Microsoft here. There is a lot of CS nomenclature that I do not really get. I think it is because I never took Set Theory. I much prefer pseudocode, but that is really limiting. I should probably get over that. DNA reaction systems are comprehensible in terms of set theory (if you ignore reaction rates), so I should probably learn.