Antibody-drug conjugate programs commit enormous resources at candidate selection. Payload class, linker chemistry, DAR, and dosing schedule get locked in long before the first GLP tox study, and unwinding a bad choice later costs a year or more. Yet most discovery teams make these calls with spreadsheet PK at best, because full mechanistic ADC modeling has traditionally required coding and modeling expertise and weeks of setup.

This post walks through a therapeutic index estimate for a representative ADC design in Pylot, start to finish, in about ten minutes. You can follow along in the live demo without creating an account.

The design we want to pressure-test

Suppose your team is weighing a solid-tumor ADC with this profile:

Payload: auristatin class, membrane-permeable
Target antigen: a solid-tumor antigen with moderate normal-tissue expression
Linker: vc (protease-cleavable)
DAR: 3 to 4
Dosing: repeat, every three weeks

The question at this stage is not "what is the exact human PK." It is "does this design have enough room between efficacious tumor exposure and tolerated systemic exposure to justify advancing it."

Step 1: enter the design, not equations

In the ADC Consultant workspace you specify the design in the same terms your project team uses: payload class, target, linker chemistry, DAR range, and dosing schedule. There are no differential equations to write and no control streams to debug. The underlying models are the standard multi-analyte structure the field expects, tracking total antibody, conjugated ADC, and free payload as separate species.

Step 2: read the exposure profiles

Running the simulation produces concentration-time profiles for all three analytes, plus tumor compartment exposure for the payload. Two readouts matter most at this stage:

Tumor payload exposure relative to the potency of the payload class, which drives the efficacy side of the ratio.
Systemic free payload exposure, which drives the classic auristatin toxicities on the tolerability side.

The therapeutic index view puts these on one screen, so the conversation shifts from "what does the curve look like" to "is the window wide enough."

Step 3: anchor against approved ADCs

An early-stage estimate is only as credible as its priors. Pylot anchors simulations against precedents from approved ADC therapeutics, including brentuximab vedotin and trastuzumab deruxtecan class behavior, so your design's numbers sit in context rather than in a vacuum. If your predicted window is narrower than every approved comparator in the class, that is a finding worth taking to the team now, not after candidate nomination.

One limit to know: the efficacy side needs a tumor static concentration (TSC), and the only sourced ones for approved ADCs are plasma values from mouse xenografts (T-DM1, Betts 2020). Pylot classifies the index where a TSC for the same payload exists. For other payloads, this vc-MMAE example included, it shows the ratio but labels it exploratory rather than calling the window wide or narrow.

Step 4: interrogate the design with what-ifs

This is where interactive modeling earns its keep. A few questions we would ask of this design, each answerable in under a minute:

What does DAR 7-8 buy us, and what does it cost? Higher DAR raises payload delivery but typically accelerates clearance and narrows the window.
Does a non-cleavable linker change the free payload picture? Swapping vc for a non-cleavable option shifts payload release from systemic to intracellular, with consequences for both efficacy and bystander effect.
Single versus repeat dosing? Cumulative exposure drives several ADC toxicity domains, so schedule is a design lever, not an afterthought.

Each variation reruns instantly, and the AI guide can explain any result in terms of the model's actual parameters and assumptions. It is grounded in the calculation you just ran and cites its sources; it does not improvise numbers.

What this does and does not replace

A discovery-stage TI estimate does not replace GLP toxicology, allometric scaling with real preclinical data, or a formal population PK effort later in development. What it replaces is the guesswork phase: advancing a design for months before anyone quantifies whether the exposure window is plausible.