Imagine your model gives you two antibodies. Both bind the human epidermal growth factor receptor (EGFR) about equally well at pH 6.5. You repeat the experiment at pH 7.4. One still binds strongly. The other gives you a curve you can barely distinguish from background.
For this particular design brief, I’d take a closer look at the antibody with the flat curve.
The Poreva Open EGFR challenge, which listed November 18, 2026, as its submission deadline, asked for molecules that bind human EGFR at pH 6.5 and show no detectable binding at pH 7.4. Human EGFR affinity accounts for 20% of the score. Once a design clears the affinity threshold, pH selectivity is worth 30%. The organizers expect results in April 2027, but the scoring already tells us something about what makes a good candidate.
Why would weaker binding earn a better score? Healthy tissues express EGFR too, and the space around cells in a solid tumor can be more acidic. An antibody that recognizes EGFR more readily under acidic conditions could favor tumor tissue. The first question a lab can answer is much simpler. Does changing the pH actually change binding enough to influence which antibody moves forward?
One extra column changes the shortlist
A study by Liu and colleagues gives us a real pair of antibodies to compare. Four years before the challenge, the team used surface plasmon resonance (SPR) to measure an engineered antibody called G532 and a closely related control. They reported the equilibrium dissociation constant, or KD. A lower value means tighter binding.
| Antibody | Human EGFR at pH 6.5 | Human EGFR at pH 7.4 |
|---|---|---|
| G532 | 294 nM | 3,900 nM |
| G532 control | 406 nM | 243 nM |
Cover the last column and I’d put those antibodies on the same shortlist. Uncover it and I’d make two different decisions. The control actually bound EGFR more tightly at pH 7.4. G532 bound about 13 times more weakly there. For a team trying to build in a preference for acidic conditions, G532 is the one I’d investigate.
There’s a detail I’d keep attached to that result wherever the data go next. G532 still bound at pH 7.4. A 13-fold shift is a useful lead. Calling it “no detectable binding” would give the next design round the wrong answer.
What I’d measure on Revo and Alto
For a small batch of newly expressed antibodies, I’d start with the same concentration series at pH 6.5 and 7.4. I’d match each sample to its running buffer and include a reference surface and a buffer blank at each pH. A buffer mismatch can shift the SPR signal, which matters when a result near background could decide which antibody advances.
This is where I see a good fit for Nicoya’s Revo. It comes in four- and eight-channel configurations, prepares serial dilutions on the cartridge and can work with as little as 2 µL of sample. When the first expression batch is small, I’d rather use that material to repeat the neutral-pH measurement than use it up preparing dilution series by hand. If the repeat stays flat, I’d record “no binding detected” and the highest concentration tested. That gives the result meaning when someone opens the dataset six months later.
For a promising lead, I’d use Alto’s 16-channel format to ask where the change happens. Nicoya’s kinetic assay guide documents up to eight different running buffers on one cartridge. I’d use that flexibility to add intermediate pH values. An antibody that loses most of its binding by pH 6.8 would send me into the next cell experiment with a different expectation from one that still binds well at pH 7.2. You can’t see that difference from the two endpoints alone.
I’d also look at how the antibody lets go. In a separate Nicoya application study on Alto, the team measured an antibody binding its target in buffer and in undiluted serum on the same cartridge. Its affinity shifted from 9.31 nM to 15.4 nM, while the measured dissociation rate more than doubled in serum. The team changed the sample environment in that study. For an EGFR lead, I’d ask the equivalent question about pH. Does the antibody have more trouble attaching at pH 7.4, or does it attach and then release faster? The binding curves can help tell those possibilities apart.
What I’d send back to the model
In What Should Your AI Model Learn From the Next Plate of Experiments?, I wrote about choosing measurements that improve the next round of designs. For EGFR, I’d give the model a result for each pH, keep the concentration range with each result, and distinguish “below detection” from “never tested.” A single “EGFR binder” label would erase the most useful difference in that table.
If a candidate binds well enough at pH 6.5 and repeatedly stays below detection at pH 7.4, that’s the one I’d spend the next cell experiment on. Then I’d ask whether the preference holds when EGFR sits on a cell surface. That’s a decision the team can act on today, while the much larger question of tumor selectivity in patients remains ahead.
Desmond Gilmour
Email: dgilmour@nicoyalife.com
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.