On Jul 23, 2023, at 7:50 PM, Bret Victor <****************> wrote:

Here's a sketch of a script.  It seems long, but it can probably move fast because it's mostly show-and-tell (or rather, make-and-tell).  Or maybe it's actually really long, I don't know.

The idea is to try to minimize the amount of "single-use content" (stuff that we build just for one part of the talk), and instead work within general-purpose environments where we can conjure up what we want to talk about on the fly.  The two main environments are:

The timeline

Papers.  When we bring out a paper, the paper and all of its references show up on the timeline.  We can then go through and tell the backstory of the paper, talking about the interesting references (and selecting figures from them for big display) and dismissing the uninteresting ones.  The intention is to connect paths of influence and prior work, to see the "tapestry" of the development of these ideas.

People.  In parallel, as we call attention to individual people, their faces and their contributions across their entire careers also show up on the timeline.

Places.  Perhaps events on the timeline also show up on a map, or are grouped by site somehow, so we can track the influence of advisors and labs (e.g. the Shih diaspora).

The table

The main thing I want to try here is simply typing DNA strands directly onto the table, with the keyboard, and having a set of cards which let us manipulate them by hand -- reposition them, pair them, zip and unzip them, and so on.  And other views which show rendered 3D models of the "scene" on the table (maybe both candystriped-cylinder and pdb-style), simulated gels, and other analysis tools.  Most of the constructs in the early papers can be made by typing in sequences straight from the paper (or copy and pasting from the paper), and making the junctions etc by hand, without any canned preparation specific to the paper.  The origami sequences have to be imported, but can hopefully be manipulated and explored in the same environment.  I have a vague image of this in my head, I think it'll work, but for now you'll have to use your imagination as you read this.  Whenever I say "show" in the script below, I usually mean "perform this action on the table".

In later papers, 3D structures become important, and I have in mind a "human-powered 3D printer" -- a tool that guides you in constructing 3D rod or wire models by hand, by projecting what to do on each layer, and prompting you to rotate the piece as needed.  I'd like us to demonstrate 3D structures by holding physical models up to the camera, but models that we've constructed in the moment instead of ones we've brought in.  I don't know if this will work, but I think we could pull off something like this.  I find it really hard to understand 3D architectures from the papers (origami, bricks, etc), and I bet we can do it better.

Another 3D possibility (less general-purpose, but potentially very useful for talking about DNA), is a dynamic virtual helix -- a rod or tube that we've wrapped with two strands of addressable LEDs, so we can light up any phosphate.  We could hold a set of these rods in a honeycomb arrangement, and light up the scaffold and staples in different colors so we can see the crossovers.  As well as animate them, etc.  It would essentially be a "DNA display".  I don't think there's time to make this before the talk, but it's certainly doable and would be helpful for future talks.  (If the phosphates are pushbuttons as well as LEDs, it could even be a controller for editing.)

One thing that I think we'll be cutting from the talk is the autobreak stuff, and particular the mathematical derivation that I was working on.  It doesn't really fit, and I won't have time to finish it anyway.  It will return for a future talk!

As you read the script, it might be helpful to follow along with the papers, since a lot of this is recreating the figures from the papers.  ************************

Introduction to Realtalk

"We're giving this presentation using a research computing environment, so I want to take a few minutes so you understand what you're looking at."

Brief background on Dynamicland.

"People made all kinds of projects." Quick demos of some classic projects.  (Gapminder, Geokit, etc.)

"Last year, we worked with Shawn to reimagine some bio tools."  Quick demo of proteins.

Not a product, but an authoring environment.  "This poster is the implementation of the protein tools."

Edit the page to change protein coloring from rainbow to charge.  Make a card that switches the coloring when it's present.

Make "points-at-highlighted-blue", or something simple but dramatic like that.  Maybe Shawn and I can make two separate things at the same time and combine them.

"We normally work with people gathered around a table.  Something we've never done before, until today, is to try using this computing system to give a lecture to a seated audience.  So this is an experimental presentation, and thank you for participating in the experiment."

"If any of this seems interesting to you, we've got a setup in the room over there, so you can get your hands on it and we can talk about it.  And if it still seems interesting, we'd love to talk about collaborating."

"With all that out of the way, Shawn is going to talk about one one his CCC collaborations."

Cell playgrounds  (Is there a real name for this? Did I just make this up?)

(Maybe initially presented as normal-looking slides, but by dealing out cards.)

The biology questions that Orion wants to answer.

Methods and preliminary data.

"To do this, we had to fabricate precisely-defined physical structures at the nm scale... [specs]."

"How could anything like that possibly be made at all? "

Alternative nanofabrication technologies and why they wouldn't work and/or would be stupidly expensive.

"Instead, we designed these structures in a day, ordered $20 of materials, and fabricated 50 billion of them overnight in a thermal cycler."

"But it took 70 years to get to this point. In this presentation, we'd like to give you a hands-on tour of the development of the field..."

Hybridization (1956)

Put down "1956. Rich & Davies, A New Two Stranded Helical Structure: Polyadenylic Acid and Polyuridylic Acid"

Backstory on creating randomAU vs polyA and polyU.

Live demonstration for the camera with real materials. "Here we have a sample of randomAU." [We don't need this part if randomAU is hard to get.]

"Here we have polyA and polyU." Mix them and observe the change in viscosity and optical density.

"He drew fibers, like this, and their X-ray diffraction pattern looked something like this [show RNA diffraction pattern], from which he inferred that the polyA and polyU strands were forming a double helix, spontaneously, without an enzyme."

"What was going on?"

Type "AAAAAAAAAAAAAAAAAA" and "UUUUUUUUUUUUUUUUUUUU" onto the table. Bring them together and show how they pair, and in the 3D view, how they form a helix.

Type "AUAUAUUUUAUAUAUAA" and "UUAUUUAUAUAUAUAUAAUAU" (randomly), and show how there is limited pairing and steric interference.

Bring out a free-energy analysis card, and show the energetics of polyA+polyU hybridizing at room temperature, and the minimum stable length.

Bring out cards that generate thousands of random sequences and try to pair them, show the distribution of continuously-paired lengths, and show that the fortuitously-paired stretches are not long enough to hold together random sequences.

Why did the experts not expect this or disbelieve it?  Show a simulation that reflects the experts' mental model at the time.  (e.g. add charge repulsion to drive the strands apart.)  What was wrong with the experts' model?  (add the effect of the sodium)

Junctions (1982)

Put down "1982. Seeman, Nucleic Acid Junctions and Lattices"

Backstory via the references on the timeline, etc

Fig 1. Replicational junction.  Type "AGTATC" onto the table.  Swipe a pairing card along it to form the duplex.  Unzip 3 pairs of the duplex, and pair each arm.  (Maybe talk about how DNA polymerase does this.)  Show how this structure isn't likely to form [last sentence of fig 1 caption] by pulling apart the strands and pairing them into the hexamer and trimer.  Point out the sequence symmetry that allows for this.  Show how, even if the structure does form, branch point migration will disassemble it.

Fig 2. Recombinational junction.  (a) Type "CGACTGTA", pair it, duplicate and flip it, unzip 4 pairs of each, cross them over.  Maybe talk a bit about biological recombination.  (b) Make a linked copy of (a), so changes to one are reflected on the other.  Reposition the duplexes to form a cross.  (c) Slide the junction around, and see how it slides in both form (a) and form (b).  Point out the features of the sequence that allow it to slide.

Fig 3. Immobile junction.  Type two strands (top-left, bottom-right), position them like a cross, and make the paired strands (bottom-left, top-right), pointing out where we're crossing over the pairing tool from one strand to the other.  Try to slide the junction, and show why it doesn't work, assuming absolute Watson-Crick pairing.

Put down "1983. Kallenbach, An immobile nucleic acid junction constructed from oligonucleotides" and show their data.

Design rules. (p241) Describe Ned's "weak assumption #1" (Watson-Crick pairing may be imperfect near junctions), and why he thinks he needs a bunch of design rules to ensure that only this structure forms.  Show each of the four design rules by making a counter-example and showing the potential failure it enables.

Maybe: put down "1983. Seeman, Design of immobile nucleic acid junctions" and replicate their thermodynamic analysis, producing melting curves like figs 4, 6, 7.  Try to demonstrate why Ned was so concerned about optimizing fidelity.

Fig 4. Semi-mobile junction.  Modify the immobile junction (e.g., change one of the CG pairs near the junction to GC) and show how it flip-flops.

Fig 6. 2D lattices.  Add some sticky ends, and show how they tile.

Fig 5. 3D lattices.  Show how the helical geometry allows for junctions coming off at various angles.  Make a Realtalk-guided rod model of a lattice.

Cube (1991)

Put down "1991. Chen & Seeman, Synthesis from DNA of a molecule with the connectivity of a cube"

The strand sequences are at the bottom-right of the first page of the paper. Copy and paste them from the paper onto the table, then follow the steps in figure 1 (cyclize, hybridize, ligate, purify) to form the cube.  Close-up inspection of a corner to see the junction.  Show/talk about the challenges in making this work, why all the purification is necessary, etc (to set up the contrast later with origami).  Maybe show the gels (figs 3, 4), and on the table make the intermediate/failure products indicated on the gel (especially if our simulated gel can match the photo).

3D crystal (2009)

Put down "2009. Zheng & Seeman, From molecule to macroscopic via the rational design of a self-assembled 3D DNA crystal"

Type it in: magenta "TCTGATGTGGCTGC", green "GAGCAGCCTGTACGGACATCA", blue "CCGTACA"x3, and connect them up.  We want to understand the 3D angles of the junctions, and their relation to helix geometry.  In addition to the rendered 3D views, make a Realtalk-guided physical model of a triangle, and (somehow) eight of those triangles to construct the rhombohedron (fig 3).  In addition to our "hand-positioned" 3D models, bring up the solved structure from the PDB, and use a program to link them (automatically via the sequences), so we can point to bases on our model and see them on the PDB model, and vice-versa.

[When I rotate the crystal at the bottom of this page, I can't make any sense of it.  Can we make a model that is actually understandable?  I think that will involve snapping together the triangles one at a time in physical space.]

Maybe: if we can find the rest of the crystals in Table 1 (they're not in the PDB or supplementary data), do a quick tour through all of them.

Get a sense for the macroscopic nature of the crystals.  250 um is 0.01", or 0.75 pt, or the size of a period in 7.5 pt Helvetica.  The paper says that's 4.5e13 unit cells (45 teracells), or 36000 cells in each dimension.  Can we bridge the nanoscale and macroscale?  If the crystal were blown up to a 10 ft cube (easy to visualize inside the conference room), the cells would be at 300 dpi, which is sort of imaginable.

The crystal cell is 7 nm across. A modern DRAM cell is about 30 nm across, and a 16GB die might be 8 mm across, with 128 gigacells tiled in 2D...

Origami (2006)

Put down "2006. Rothemund, Folding DNA to create nanoscale shapes and patterns"

Happy face. [A preview of the punchline before we step back.]  Put down a card which puts the M13mp18 scaffold on the table.  Switch the scale bar on the table so we can see it all.  Put down a card which puts the happy face staples on the table, and animate the folding of the happy face.  

"How did Paul come to this idea?" Backstory of Paul's work with DNA computation and algorithmic self-assembly.  On the table, perform some sort of DNA computation primitives, to give a sense of what they were thinking about.  [I don't know any of this very well.]  Maybe put down Paul's 2001 dissertation, or the 2004 Sierpinski triangle paper?  We just want some context for where Paul was coming from.

Double crossovers. Put down "1993. Fu & Seeman. DNA double-crossover molecules".  "The first rigid, engineered DNA structure", which would become the core structural primitive of origami.  Make the DAE (bottom-left) construct in Fig 4.  Use the 3D views to call attention to how crossover placement is tied to helical geometry.  Move one of the double crossovers left and right and show how it rotates so it's pointing the wrong direction.

Back to the happy face, zoom in on one corner of it, remove the staples, and add back a few staples by hand, showing how the double crossover pattern from Ned's paper appears directly in the happy face.  Zoom back out to show how the entire happy face is built out Ned's DAE construct.

Scaffolds. What's different from Ned's paper is the scaffold.  Call attention to how scaffolds are different from the oligos that Ned et al had been working with.  Maybe talk a bit about filamentous bacteriophage as a fortuitously naturally-occurring source of ssDNA?  Or show what's difficult about obtaining long ssDNA?  e.g. why you can't just get scaffold by melting a dsDNA genome.  Maybe set up for mentioning pScaf later.

Shapes!  Show the shapes from Paul's paper, and get a sense for their size and properties.  Maybe show raster filling, seams, etc.

Patterns!  On a particular shape, take a particular staple, type the sequence for a dumbbell hairpin into the middle, fold up the hairpin, and put back the staple.  Show how, for a particular shape, you can make a plate of normal staples and a matching plate of hairpin staples, and by choosing which staples to add, you can draw an arbitrary picture on top of the shape.  Make Paul's "DNA" pattern in fig 3a.  Maybe show how blunt-end stacking creates the aggregates in fig 3b?  Talk about molecular breadboards, and the benefit of an asymmetric fully-addressable structure vs a crystal.

Method.  Mock-perform the process of folding an origami -- ordering oligos and scaffold, combining with buffer and salt, heating and cooling. Live bill of materials on the wall, estimating the cost.  Compare with the extensive processes required by other DNA nanostructures, and other nanotechnologies.

Problems solved by origami.  Show how the cube demands careful stoichiometry and purification. Make examples of how oligo-based design can fail, and how failed structures are difficult to separate out.  Show how scaffold avoids the problems.  Show how staple excess works.  (Show how staples don't bind to each other.  Show, on the gel, how staples and folded structures are well separated.)

Next steps.  Put down "2005. Rothemund. Design of DNA origami".  At the end of the paper, Paul identifies three directions for further research:  CAD tools, 3D origami, and sequence optimization via custom scaffolds.  Put the corresponding Shawn papers on the timeline (2009, 2009, and 2018).

CAD tools (2009)

Put down "2009. Douglas et al, Rapid prototyping of 3D DNA-origami shapes with caDNAno".

Tell the "my advisor told me not to do it" story.  Show a couple quick videos of using cadnano.  Show how it ignited the field, by showing on the timeline all the later paper that cite it, and showing the structures from those papers that were designed in cadnano.  Show the BIOMODs on the timeline, etc.  Show (via screenshots and timeline) the cadnano clones and other CAD tools that arose (see e.g. 2021. Dey. DNA origami primer). 

We're just trying to establish how essential tools are.

3D origami (2009)

Put down "2009. Douglas et al, Self-assembly of DNA into nanoscale three-dimensional shapes"

Make a rod model of the simplest 3D shape, such as a short 6-helix tube.  Show the scaffold routing and crossover positions.

Maybe show the shapes from the paper (as rendered 3D).

Show later structures that were made in this architecture -- goniometer, etc.  Give a sense of the breadth of projects built on 3D origami.

[Maybe] Optimization

Scaffold optimization: pScaf.

Staple optimization: Autobreak.

[We could cut this; it's not really part of the flow of the talk.]

Bricks (2012)

Mention (with the timeline) other architectures that were inspired by origami.

Put down "2012. Ke et al, Three-dimensional structures self-assembled from DNA bricks"

Generate two random 32-base strands, as "north bricks", and pair them appropriately to make a "west brick" (fig S2).

Make a Realtalk-guided wire model of a cube of bricks, enough to show how bricks pair in 3D space.

Advantages and disadvantages of bricks vs origami.

Cell playgrounds redux

Design a brick-based cell playground in Realtalk.  Generate the bill of materials, the protocol on plates and racks, the simulated gel, and any other downstream simulations.  Mock-perform the protocol, enough to convey the idea that "that's all there is to it".

"And that's how anything like this could possibly be made at all."

Meta

Talk about the presentation itself, doing science in Realtalk, the lab of the future.