After several months of internal development and a focused testing period with a small group of oncology research partners, we are opening early access to the Avenzo Target Ranking Engine.
This post is addressed to research teams considering applying. It covers what the early access program actually involves, what you get from it, what we are hoping to learn, and what we are not yet ready to offer.
What Led to This Point
We started building Avenzo because a problem kept showing up in every oncology research program we had been part of: the evidence needed to make a better target decision existed in the literature, but no one had time to find all of it. The synthesis bottleneck was structural, not scientific. The question was whether a software system could address it at the speed and evidence coverage that a discovery team's decision cycle actually requires.
The early development phase was spent building the evidence ingestion and classification architecture: pulling from preclinical literature, trial context data, and computational databases; classifying evidence by type and quality tier; and integrating across source types into a composite target score. That part of the work is largely done. What we are looking for in the early access cohort is real discovery workflow context we cannot simulate internally.
What Early Access Involves
Early access is structured around a specific use case: a solid-tumor target screening or go/no-go decision that a research team has on the near-term horizon. You bring a target list, we run the evidence analysis, and we walk through the output together. The first run is a working session, not a demo. We want to understand how the ranked output maps against your existing assessment, where it adds information you did not have, and where it may have missed something your team already knew.
Access is currently by application. We are not running an open sign-up. The cohort size is deliberately small, because the value of early access is the feedback loop between the platform output and the team's expert judgment. That feedback only works if we have bandwidth to engage with each partner's actual workflow.
The early access period runs for a defined set of analysis runs per team. We are not billing for this phase. What we are asking for in exchange is structured feedback: which outputs were useful, which were surprising, where the evidence classification looked wrong to your team's subject-matter knowledge, and whether the ranked output changed how you thought about any target.
What You Get
Each analysis run returns a ranked evidence report for the targets you submit. The report includes a composite evidence score per target, broken out by the three evidence dimensions: preclinical support depth, clinical context alignment, and computational signal convergence. Each dimension score includes a source count, an evidence quality breakdown by tier, and flags for specific evidence patterns we consider particularly meaningful (strong in-vivo support, contradictory clinical context, thin evidence base despite high citation count, and others).
The output is structured for direct use in a target review meeting. We designed it to answer the questions that come up in those meetings: how much of the preclinical evidence is in-vivo versus cell-line? Is there any clinical context that either supports or cuts against the nomination? Does the pathway context reinforce or complicate the direct evidence? You get source attribution throughout, so the report is defensible to the scientific team, not a black-box score.
What We Are Hoping to Learn
The honest answer is that we are testing the evidence classification model against expert judgment in real research contexts. Internal benchmarking against known outcomes from the published literature gives us one signal on model quality. But it does not tell us how the output performs against a working scientist's assessment of a target they know deeply.
We expect to encounter cases where the model's evidence classification disagrees with the team's reading. Those cases are the most valuable for model improvement. When an experienced solid-tumor biologist looks at our evidence ranking for a target they have studied for two years and says the tier weighting is wrong, we want to understand their reasoning. Not to immediately accept it as ground truth, but to examine whether the model is missing something the literature actually contains or whether this is a case where published evidence and expert knowledge genuinely diverge.
What We Are Not Ready to Offer Yet
Early access is scoped to solid-tumor target identification. We are not yet running analyses for hematologic malignancies, and the model has not been trained or validated against targets in those contexts. If your program is in a hematologic tumor type, early access is not the right fit right now.
The platform is also not yet integrated with internal data sources. The evidence corpus is entirely from indexed external sources: peer-reviewed literature, preprints, trial registries, and public computational databases. We cannot incorporate a team's proprietary in-house experimental data into the scoring model in this phase. Teams with substantial unpublished internal validation should treat the platform output as a supplement to their internal evidence base, not a substitute.
How to Apply
If you are running a solid-tumor target identification program and want to bring a real decision into the early access period, reach out at [email protected]. Describe the program context, the tumor type, and the rough number of targets you would be evaluating. We will respond with whether the fit looks right for this cohort and what the next step would be.
We are looking for teams where this analysis can actually inform a real upcoming decision, not teams looking for a demo. The early access program is designed to produce genuine learning on both sides, and that requires a real research context on the partner side.