SPR or BLI for AI-Designed Proteins? Why I’d Start With Alto
Suppose 12 AI-designed proteins arrive on Monday and your team can afford to make more of three. By Friday, you need to know which three. You also need enough of the most promising one left to check whether its binding h
Suppose 12 AI-designed proteins arrive on Monday and your team can afford to make more of three. By Friday, you need to know which three. You also need enough of the most promising one left to check whether its binding holds up when you change the experiment.
For that lab, I’d start with Nicoya Alto digital SPR. It offers much of the ease that makes biolayer interferometry (BLI) appealing, with the real-time binding data of surface plasmon resonance (SPR). The feature that tips the decision for me is simple. You load the starting sample, and Alto makes the serial dilutions on its cartridge. In a published Nicoya assay, 2 µL of antibody stock was enough for a five-concentration binding series.
What did those 2 µL actually buy?
Nicoya tested how a mouse antibody bound to an Fc receptor. On Alto, the team loaded 2 µL of a 300 nM antibody stock per sample well. The cartridge prepared five concentrations spanning 1.23 to 100 nM. For the 150 kDa antibody in that experiment, that works out to about 90 nanograms of starting antibody per loaded well.
A contract lab measured the same interaction by BLI. Here’s what Nicoya reported in its side-by-side comparison.
| What the team measured | Alto digital SPR | BLI workflow |
|---|---|---|
| Binding affinity using the same one-site fit | 8.27 nM | 9.36 nM |
| Hands-on operator time | 28 minutes | 66 minutes |
The BLI estimate was 7.67 nM when the fit accounted for how quickly antibody reached the sensor. Both methods put the interaction in a similar affinity range. What would change my week is the 38 minutes of hands-on time saved in this assay. Nicoya attributed 25 minutes of savings to dilution preparation alone. If my lab receives a fresh batch of designs every Monday, that’s the step I’d want the instrument doing for me.
Both workflows can be automated. Where does the work happen?
BLI has a real strength when a lab already has plates of diluted samples ready to run. The Sartorius Octet RH96 supports robotic loading of plates and biosensor trays, and samples can be recovered afterward. Sartorius lists 40–80 µL per well in its tilted 384-well plate format. That’s diluted well volume, so I’d compare the micrograms taken from the original stock across the whole experiment.
Alto is the better fit for me when the lab starts with small tubes of concentrated protein and needs binding curves at several concentrations. Its 16-channel cartridge handles the dilution series and moves the samples and buffers to the sensors. With the Alto Automation Suite, an Opentrons Flex can prepare cartridges, and a lab’s scheduling software can check instrument status, start runs and retrieve results. The next batch can be in preparation while the current one is collecting data.
That difference becomes clearer when you plan a real workload. Twenty-four designs at five concentrations in duplicate means 240 concentration measurements before extra controls or reruns. My throughput question would be how many of those 24 designs get a result the team can use by Friday, and how much protein it takes to get there.
What does SPR-quality data look like?
In a separate Nicoya study comparing Alto with conventional SPR, researchers measured an antibody fragment binding to green fluorescent protein. They reported affinities of 8.6 nM on Alto and 10.0 nM on conventional SPR. Figure 3 shows the measured curves at five concentrations alongside the fits. You can see how binding builds while the sample is present and how quickly the complex comes apart afterward. That’s the evidence I’d want behind a number going back to a design team.
There’s a good reason to save protein for a second look. In the BindCraft study, 13 of 53 designs showed binding to PD-1 in a BLI screen that included an unrelated protein as a specificity control. The strongest hit released so slowly, and benefited so much from its two binding arms, that the authors reported an apparent affinity tighter than 1 nM. That less-than sign belongs in the dataset. Entering 1 nM would give the next model a number nobody measured.
For a finalist like that, I’d use some of the protein saved on the first run to repeat the measurement with less target on the sensor and more time to watch it release. On a crowded surface, a molecule can detach and catch a neighboring target, making it appear to stay bound longer. If the ranking holds up on the second surface, I have a stronger reason to make more of that design.
For smaller batches, Revo 4 and Revo 8 bring the same practical idea to a four- or eight-channel benchtop system, with automated on-cartridge dilutions and a 2 µL minimum sample input. Alto is where I’d start for a team running recurring design cycles and building toward a connected, automated lab.
The experiment I’d buy Alto to run
I’d put one real batch in the demo, including a known binder, an unrelated control and a candidate the team can barely spare. Then I’d ask to see the curves, the hands-on time and what’s left in that last tube.
That last tube is why I’d choose Alto for this work. A first screen can tell me which design looks best. If automated dilutions leave enough of it to change the surface, repeat the measurement and see whether it still comes out on top, I can tell the design team why it deserves another round. That’s a better starting point for the next set of AI-designed proteins.
Desmond Gilmour
Email: [email protected]
Phone: +1 519-546-9876
I work at Nicoya Lifesciences. The views expressed here are my own and do not necessarily reflect those of Nicoya. This is a personal publication, not an official company communication.
Originally published by Dev.to AI. Aggregated on AIWithGhost for educational purposes — full credit and traffic to the original publisher.