Baker Lab Open Scripts Target pH Switchable Protein Binders

What changed

The Baker laboratory published code for pH dependent protein binders on GitHub in October 2026. The repository baker-laboratory/ph_binder_paper documents two pipelines. One pipeline disrupts binder structure at low pH using placed histidines. The other pipeline builds cross interface hydrogen bonds that weaken when pH drops.

Both paths use ProteinMPNN to redesign sequences around engineered histidine sites. The README lists conda setup, PyRosetta needs, and step by step scripts.

Histidine tags change charge when pH shifts. That trick lets designers build switches that turn off binding in acidic spots. The repo splits an exploder style path from an interface bond path. Teams can run one path or mix both on the same starting binder.

Why it matters

Many therapeutic binders must turn off outside the target tissue. pH in tumors or endosomes differs from blood. Designers who can tune binding with histidine chemistry gain another control knob beyond plain affinity maturation.

Drug teams should treat this as a methods release tied to a paper. It is not a clinic ready drug. It is open tools plus steps for groups that already run Rosetta and MPNN stacks.

Who is affected

Protein engineering groups at biotech startups and CDMO partners evaluate binders daily. Computational chemists who already use ProteinMPNN can fork the scripts. Experimental teams must still run oracle models such as AlphaFold2 or AlphaFold3 on picks.

Open scripts make it easier to test ideas, but they do not replace lab work. pH sensitive binders fail quietly if computer filters disagree with wet tests. Budget one wet loop before you promise dates to BD partners.

What to do next

If you work on conditional binders, clone the repo into an isolated environment. Run the exploder pipeline on one internal pdb as a smoke test. Compare energy scores before you scale to large libraries.

Document which oracle you used for filtering. The README states oracles are not bundled. Skipping that step risks false positives in pH sensitivity.

Set OMP_NUM_THREADS=1 when running ProteinMPNN as the README advises. Too many threads slows batch runs and can skew compare studies across engineers.

The exploder path mutates structure at low pH. The interface path breaks cross target hydrogen bonds when pH drops. Pick the path that matches your target biology before you order DNA.

Reserve GPU nodes for oracle runs if your queue is shared. Structure filters are cheap until AlphaFold jobs fill the cluster.

Write a one page plan before you scale orders. Note who owns wet tests and who owns model picks. Clear roles prevent silent handoffs that waste plates.

What to watch

Good lab notes beat fast slides. Write down each pH step you test.

Watch the peer reviewed paper linked from the repository for wet lab validation tables. Watch whether NVIDIA BioNeMo or other hosts wrap the pipeline. Watch licensing on the ProteinMPNN submodule commit pinned in the repo.

Photo: National Cancer Institute / Pexels

Photo: CDC / Pexels

Share the repo with both compute and wet lab leads in one meeting. Split teams often reorder sequences without telling the modeling group, which wastes cycles on dead designs.

Keep a changelog when you fork scripts for internal use. Upstream Baker updates may fix silent bugs in histidine placement rules that your old fork still carries.

Track pH ranges you target in each assay plate. Without that log, teams cannot explain why two forks of the same binder diverge in the lab.

Sources

• Primary — GitHub, baker-laboratory/ph_binder_paper (October 2026). Installation steps, pipeline names, and ProteinMPNN usage notes.
• Secondary — GitHub, NVIDIA BioNeMo model catalog referencing Baker design tools (October 2026). Context on Baker linked design stacks for teams comparing vendor paths.