Every oncology discovery team has a version of the same story. A target that looked promising at nomination, backed by a handful of high-profile preclinical findings, turned out to have a thin and non-replicated evidence base when someone finally sat down to read everything. The assay work was already underway by then.
The problem is not that the evidence was missing. Usually it was there, scattered across the literature, some of it pointing in the right direction and some of it contradicting the early case. The problem was access and synthesis at the pace of real decision cycles.
Why Volume Is a Poor Proxy for Quality
When evidence synthesis gets compressed into the narrow window between target nomination and resource commitment, the practical shortcuts matter. One common shortcut is treating evidence count as a confidence signal. A target with 200 papers feels more validated than a target with 20. But volume is not quality. A target can accumulate extensive literature in a single cell-line system, or through computational re-analysis of the same source dataset, without a single independent in-vivo replication or any clinical context support.
The opposite case also exists. A target with modest total literature volume can carry stronger validation if its evidence covers multiple independent experimental systems, shows consistent direction across different tumor type contexts, and includes at least partial clinical context alignment. A ranking built on volume alone inverts this picture.
We see the consequences of volume-based prioritization in how early discovery programs allocate time. Teams that enter assay development primarily because a target has extensive literature are betting on an unexamined assumption: that more papers mean more validation. It is a reasonable heuristic when evidence synthesis time is short. It is a poor one when the evidence landscape is actually examined.
The Structural Problem in Evidence Synthesis
There is a real constraint here, and it is worth being specific about it. A thorough manual evidence synthesis for a single solid-tumor target, done properly, takes a skilled scientist roughly five to seven days. That means reviewing primary literature, preprint archives, trial registries, computational database outputs, and pathway context data, then classifying each piece of evidence by type, experimental system, tumor specificity, and consistency with the overall mechanistic picture.
In a typical target screening cycle where a discovery team is evaluating ten or more candidates concurrently, that budget does not exist. What happens instead is that the team reviews the highest-visibility papers, applies qualitative judgment, and advances targets based on a necessarily incomplete picture. The evidence problem is not a failure of scientific rigor. It is a structural consequence of time constraints on evidence retrieval and synthesis.
The gap this creates is systematic, not random. When synthesis is limited by time, the evidence that surfaces first is the evidence that is easiest to find: highly-cited papers, familiar targets with well-known biology, and recent preprints that match the nomination hypothesis. Counter-evidence, confirmatory evidence from older or lower-visibility sources, and negative or null findings are systematically underweighted.
What Ordering Actually Changes
Ranking targets by evidence quality, not volume, changes the composition of what enters assay development. Consider a scenario we encountered in our early-access program. A research team working on a lung adenocarcinoma program had three candidate targets at the top of their manual synthesis list. When the Avenzo evidence analysis came back, the rank order was materially different. The top candidate by volume had strong preclinical literature in two-dimensional cell culture, but essentially no in-vivo support and no clinical context alignment. The third candidate by volume had a smaller total evidence footprint, but it included replicated in-vivo findings across two independent mouse model systems, consistent expression patterns in tumor versus normal tissue from public genomic data, and a clean mechanistic story supported by pathway context evidence.
That kind of inversion is not unusual. The targets that look strongest when filtered by evidence quality are not always the ones that look strongest when filtered by name recognition or citation count. And the inversion matters most before bench resources commit, when the cost of reordering is a changed experimental plan rather than six weeks of assay work on the wrong candidate.
How the Ranking Is Built
The Avenzo evidence ranking model is built around three evidence dimensions that we weight independently before integration: preclinical support depth, clinical context alignment, and computational signal convergence. Each dimension receives evidence classified by quality tier.
The tier structure matters here. Tier 1 evidence in the preclinical dimension is a replicated in-vivo finding with a clear mechanistic interpretation, ideally in a tumor type-specific model. Tier 2 is an unreplicated in-vivo finding or a replicated in-vitro finding across multiple independent cell-line systems. Tier 3 is a single-system in-vitro finding. Tier 4 is computational prediction without independent experimental corroboration. A target with ten Tier 4 items does not outrank a target with two Tier 1 items, despite the volume difference.
Clinical context alignment is scored separately from preclinical evidence because the two can diverge in instructive ways. A target with strong preclinical support but absent or contradictory clinical context signal is categorically different from one where preclinical and clinical dimensions are aligned. We flag that divergence in the output rather than averaging it away.
The Boundary: What This Analysis Cannot Tell You
We are clear with our early-access partners about what an evidence ranking can and cannot do. It tells you about the quality and consistency of what is already known. It does not tell you about druggability, medicinal chemistry feasibility, safety profile, or competitive landscape. A high-confidence evidence ranking means the target's role in the tumor biology is well-supported across independent sources. It does not mean a viable drug can be made against it.
It also does not capture unpublished data. A target that appears thin in the literature may have substantial internal validation in the pipeline of a discovery organization that has not yet published. The Avenzo ranking reflects the published and indexed evidence corpus at the time of query. That is a meaningful signal, but it is one input into a decision, not the whole decision.
The teams we work with use the evidence ranking as a filter at the front of the nomination-to-commitment stage, not as a replacement for scientific judgment. The ranking changes the order in which targets get deeper attention, and it surfaces evidence the team may not have reached in a manual synthesis pass. What happens with that output is still a scientific call.
Starting With the Right Ordering
Oncology drug attrition remains high not because the biology is poorly understood, but because the translation from target nomination to clinical success is fragile. The evidence base at nomination is one lever. It is not the only one, and improving it does not guarantee clinical success. But it is a lever that is under-optimized in most discovery workflows because the synthesis problem is structural rather than scientific.
Getting the ordering right before bench resources commit is the point where evidence ranking has its highest leverage. Not because every high-ranking target will succeed, but because targets that fail for evidence quality reasons are the most preventable failures in the pipeline. The evidence existed. The problem was synthesis at decision speed.
That is the problem Avenzo is built to address. If you are running a solid-tumor target selection cycle and want to see how an evidence-ranked output compares with your current prioritization, reach out to the team.