AI-designed binders

Measure kon, koff and KD for up to 64 designed binders per chip, directly from lysate. Each design gets a DNA code, the coded designs are pooled, and the chip sorts the pool by itself.

The question

Which of my designed binders bind the target, how fast, and how tightly, without purifying each one?

A protein design model can propose hundreds of binders in a day. Testing them is slower. Purifying each design is often the step that limits the cycle. Methods that measure one binder at a time scale with the number of designs. An endpoint assay ranks designs by one number and hides how fast each one binds and lets go. The next design round is better informed by rate constants for every design.

The setup

ExpressEach design is expressed on its own with a SpyTag or a Twin-Strep-tag, and keeps its design ID
CodeA different coded capture adapter goes into each lysate, with binder in excess over adapter. Covalent: SpyTag with the Oligo Adapter SpyCatcher. Affinity: Twin-Strep-tag with the Oligo Adapter Strep-Tactin® XT
PoolUp to 64 coded lysates go into one well. More pools fill a 96-well plate
ChipA DNA chip; each adapter's oligo finds its own address by DNA-directed immobilization (DDI). 64 designs need a 64-plex chip, which is available on request. At 32-plex, each design gets two molograms
MeasureThe target flows over the chip in a concentration series, for example by single-cycle kinetics. Then the chip is regenerated and the next pool is loaded
Check per constructTag accessibility, expression level and compatibility of the lysate with the assay
Express, code, pool, sort, measure

Eight designs on an 8-plex chip are shown, eight molograms each. A 64-plex chip takes 64 designs, one mologram each. Play the workflow or step through it.

What a screen looks like

The figure below simulates one pool of 64 designs. Most designs in a first round do not bind; a few do, with different rate constants. Each dot is one design on a map of kon against koff. Diagonal lines mark equal KD.

Rank 64 designs

Each binding design gets a single-cycle run: five target injections of 120 s, from 1 to 81 nM in 3-fold steps. There is no regeneration in between, and 600 s of buffer follow. Click a design to see its sensorgram. Set a KD cutoff to mark hits. Then switch to an endpoint ranking: one number per design, the response at the end of the last injection. Press "Same endpoint" to compare two designs that an endpoint assay cannot tell apart.

How to read it

How many designs per plate

The nominal capacity is 96 pools × 64 positions = 6,144 designs per plate, before controls and replicates. Real throughput is lower. A reference takes an address, and a lower multiplexing level gives each design more molograms. The number of pools per chip depends on how well the assay regenerates.

Designs per plate

Why it works

The capture is the purification. Only the tagged design binds the adapter, and only the adapter's oligo binds the address. Everything else in the lysate is washed away or adsorbs evenly on ridges and grooves, where it adds little coherent signal. So the lysate never needs to be purified, only to carry enough binder.

Binder in excess over adapter does two jobs. First, every adapter leaves its lysate with a binder on it. So the adapter sets the capture level on the chip, not the expression level of each design. Second, no adapter reaches the pool with an empty capture site that could take up a different design.

The covalent SpyCatcher route cannot exchange after pooling. An affinity capture holds only while the binder stays bound. On a multiplexed chip, a design that lets go can bind a neighboring address, so the attachment must be stable.

Each code is an address. DDI sorts the pool: every adapter oligo hybridizes only to its complementary capture strand, so the chip reads the design ID from the position. Stripping the duplex clears the chip for the next pool.

Read more: DNA-directed immobilization, strip and reload, what a curve shape tells you, and why only patterned mass builds the signal.

More