Blog
September 1, 2026
Presenting on Our Approach to the OpenADMET PXR Blind Challenge Win
.svg)
In July, Inductive took first place among 350+ researchers from pharma, biotech and AI labs in OpenADMET’s 2026 blind challenge to predict activation of PXR, a receptor that detects foreign compounds and induces the enzymes that clear them from the body. When a drug triggers that response, it can accelerate its own metabolism, alter dosing, and interfere with other medications: a failure mode that often isn't caught until mid-to-late lead optimization.
In this webinar, Duncan Muir, who led Inductive's PXR submission, walks through our approach end-to-end. He shares:
- How an early model, trained on the challenge's dose-response data, was strengthened with additional data types and then blended with a 3D model
- How error analysis, aided by our own AI assistant Indy, surfaced a pattern behind where the model was overpredicting
- How we built a proxy model to better represent the weak, hard-to-fit compounds
He closes with a look at where structural and biophysical methods might take PXR modeling next.
This challenge is part of a larger effort at Inductive. Alongside academic and industry collaborators, we're building toxicity and drug safety tools under an ARPA-H program, combining in silico models, mechanistic assays, and higher-order biological systems to catch toxicity earlier in drug discovery and help safer drugs reach clinical trials.
- QC analytical reports (1H/13C/19F/2D NMR, chiral SFC)
- Propose, visualize, and triage forward and retrosyntheses
- Monitor the synthesis queue and flag stalled targets
- QC assay data as it lands and flag suspicious results
- Prioritize repeats and nominate compounds for downstream assays
- Analyze SAR and identify data gaps
- Evaluate IVIVC and suggest diagnostic experiments
- Design, enumerate, and triage analogs
- Launch and analyze physics-based workflows like docking and FEP
- Project human PK and efficacious dose
- Answer questions about a program’s history (what was learned and when)
- Track team progress against the target candidate profile
- Get new scientists up to speed on a program
- Make meeting-ready slides and figures
Prompt: I’m attaching two files with activity data, one with new data from today and one with historical data. Can you look through the new data and identify any QC issues that I should be aware of? Create a csv where every compound + assay run pair is labeled with any relevant issues.
Prompt: Do a nitrogen walk around the core of this molecule: O=C(NCC1=CC=C(C)C(F)=C1)C2=CC=C3C=C(N4[C@H](C)CCCC4)C=CN32
Prompt: Do an OH walk around this whole molecule: O=C(NCC1=CC=C(C)C(F)=C1)C2=CC=C3C=C(N4[C@H](C)CCCC4)C=CN32
Prompt: Replace the core of this molecule with this set of 5,6-ring systems (file attached): O=C(NCC1=CC=C(C)C(F)=C1)C2=CC=C3C=C(N4[C@H](C)CCCC4)C=CN32
[Related posts]
September 17, 2026
Blog
Beacon-2: Completing the reward function for chemical optimization in drug discovery
Dose is the primary reward function of pre-clinical discovery, and we believe that making it computable via Beacon-2 is a step change for AI in drug discovery. Now, instead of optimizing individual properties in isolation, agentic systems can drive the drug discovery process itself.
September 14, 2026
Blog
Meet Indy: an AI chemistry assistant designed by medicinal chemists to do medicinal chemistry
Indy, our medicinal chemistry assistant, supports drug discovery programs in improving the rigor of day-to-day analyses and decisions, by QC-ing assay data and monitoring the synthesis queue, to launching FEP, projecting human dose, and making meeting-ready SAR and synthesis slides.
August 1, 2024
Blog
Are local or global models better? Why not both?
A long-standing debate in cheminformatics is whether global property-prediction models perform better or worse than local QSAR models. We describe a result from our publication with Nested Therapeutics in which we show that the best of both worlds is to train a model on global data and then fine-tune it on local data.
June 7, 2024
Blog
Approaching AlphaFold 3 docking accuracy in 100 lines of code
AlphaFold 3 (AF3) is an exciting leap forward in our ability to predict the structure and properties of biomolecular systems. We explore how AF3 small-molecule docking compares to existing techniques and find that the story is more nuanced than headlines suggest. We conclude with thoughts on where AF3 may ultimately be most useful.
November 20, 2023
Blog
Get with the program: building ADME datasets that drive impact
Research advances often don't translate to practical impact. In ML for small molecule drug discovery, this is at least partly because benchmark datasets don't capture important components of drug programs. We show how this can happen in ADME prediction and provide a path forward for building more realistic benchmarks from existing public data.