The Target Is Only the Starting Point: Reconsidering Indication Selection in Solid Tumors
Our oncology expert breaks down how AI in oncology analyzes target expression and the tumor microenvironment to improve solid tumor indication selection and clinical development.
Precision oncology has become increasingly effective at identifying actionable targets. Yet the presence of a target does not necessarily mean that a tumor is biologically prepared to respond. Muaiad Kittaneh, MD, MBA, FACP, Head of R&D and Scientific Strategy at Syneos Health®, explains why development teams must look beyond target expression—and how AI-enabled predictive intelligence can help identify the solid tumors most compatible with a therapy’s mechanism of action.
Precision medicine has shaped oncology development for years. Why does clinical attrition remain so high?
Dr. Kittaneh: One persistent assumption is that target expression alone will predict response. In reality, identifying the target is only the starting point. The biological context in which that target exists can be equally important.
A therapy must operate within a complex tumor ecosystem that includes immune activity, stromal biology, vasculature, antigen presentation, metabolism and resistance mechanisms. Development teams may underestimate how much these factors differ across indications and patient populations.
This creates a central challenge in solid tumor development: A tumor may have the right target but lack the biological conditions required for the therapy to work.
See how TME-Match™ evaluates the biological fit between a therapy’s mechanism of action and the tumor microenvironment.
What does that biological mismatch look like in practice?
Dr. Kittaneh: Biological mismatch occurs when the target is present, but other components of the tumor microenvironment are not aligned with the drug’s mechanism of action.
For example, the tumor may have an immunosuppressive microenvironment, limited immune-cell infiltration, stromal barriers or vascular features that prevent the therapy from reaching or affecting its target as intended. Resistance pathways may also compensate for the pathway being inhibited.
This mismatch can remain hidden during early development. A small study may produce a compelling signal in a selected group of participants. As the program expands to larger and more heterogeneous populations, however, the signal may become diluted. The result can be a modest effect size, inconsistent responses or a late-stage failure that appears surprising but is biologically explainable.
Why is target expression alone an incomplete basis for selecting a solid tumor indication?
Dr. Kittaneh: Two tumors can express the same target and still respond very differently because their surrounding biology differs.
This is especially relevant when a target is involved in immunomodulation or antitumor immune responses. Its significance may depend on whether it is expressed by tumor cells, tumor-resident immune cells or immune-cell populations that are recruited differently across tumor types.
A target found in lung, breast and liver cancers, for example, does not necessarily perform the same biological role in each. Expression tells us that the target is present. It does not tell us whether the full tumor ecosystem supports the therapy’s mechanism. This distinction suggests a more useful unit of prediction for solid tumors: not simply the target, but the target in context.
Basket trials can offer speed and breadth. When can they become a scientific liability?
Dr. Kittaneh: Basket trials are valuable when a shared molecular alteration is genuinely associated with a consistent therapeutic effect across tumor types. The risk arises when target presence is treated as proof that the biology is equivalent.
Tumor types sharing a biomarker or mutation may have very different immune, stromal, vascular and suppressive environments. They may also rely on different resistance mechanisms or present different toxicity trade-offs.
A basket design becomes vulnerable when it groups biologically distinct tumors without first examining whether each tumor microenvironment is compatible with the drug’s mechanism of action. In this situation, the breadth of the trial can obscure meaningful differences and dilute an otherwise promising signal.
What can AI-enabled predictive intelligence add to expert literature review and conventional biomarker analysis?
Dr. Kittaneh: Expert review remains essential, but the volume and complexity of tumor biology make it difficult to evaluate every relevant relationship manually.
AI-enabled approaches can assess thousands of biological relationships simultaneously. They can identify patterns across tumor-intrinsic, immune, stromal, vascular and metabolic biology that may not be visible through a literature review or single-biomarker analysis.
The objective is not to replace scientific judgment. It is to augment it by helping teams identify biologically aligned indications, potential resistance patterns and combination opportunities that warrant deeper investigation.
In this setting, AI is most useful as a hypothesis-generation and decision-support tool—not as an autonomous decision-maker.
How does a platform such as TME-Match apply this approach?
Dr. Kittaneh: TME-Match begins with the asset’s biology. Inputs include its mechanism of action, target pathway, expected biological effects, potential resistance mechanisms and any planned combination strategy.
The platform uses RNA-sequencing data from more than 6,900 samples in The Cancer Genome Atlas spanning 21 solid tumor types. It reconstructs immune and non-immune features of the tumor microenvironment, including immune activity, stromal biology, angiogenesis, hypoxia and metabolism. Together, these features provide a distinct microenvironmental profile for each tumor type.
The platform then evaluates how well the asset’s biology aligns with each tumor ecosystem. Outputs can include:
- A biologically informed ranking of potential indications
- TME-Match scores
- The biological drivers behind each ranking
- Potential biomarker and patient-selection hypotheses
- Opportunities for combination strategies, indication expansion or more focused basket trial designs
Expert judgment enters at every stage. Scientists define the initial biological hypothesis, assess the relevance and quality of the inputs, and interpret the findings alongside preclinical results, clinical evidence and the broader development strategy.
What should happen if the model ranks an indication the development team did not expect?
Dr. Kittaneh: An unexpected ranking should be treated as a testable hypothesis, not a definitive conclusion.
The first step is to understand why the tumor type received that ranking. Which immune, stromal, vascular or tumor-intrinsic features drove the result? Do these drivers have a plausible relationship to the therapy’s mechanism? Are they supported by independent preclinical or clinical evidence?
The same principle applies when the platform predicts a poor fit for an indication the team already favors. The result should prompt an investigation rather than an automatic rejection.
This is one of the most useful aspects of predictive intelligence: It can both reinforce an existing strategy and expose assumptions that might otherwise remain unchallenged. The explanation behind the ranking is ultimately more valuable than the ranking alone.
How can teams determine whether an AI model has learned tumor biology rather than reproducing the indications the industry studies most often?
Dr. Kittaneh: Every prediction should be supported by a transparent biological rationale that can be assessed against independent evidence.
TME-Match combines data-driven analysis with an AI-assisted review of scientific and medical literature indexed in PubMed. This includes research from molecular and cellular studies, animal models, translational investigations and clinical observations—not just evidence from indications already prevalent in industry pipelines.
Teams should still challenge the findings. A credible output should explain why an indication may be a strong or poor fit, identify the biological features driving that conclusion, and generate hypotheses that can be evaluated independently.
The standard should not be whether the model produces an interesting ranking. It should be whether the rationale is biologically sound, scientifically interpretable and useful in guiding the next development decision.
What is the broader implication for solid-tumor development?
Dr. Kittaneh: Precision oncology has become highly sophisticated at finding targets. The next opportunity is to become equally sophisticated at determining whether the biology surrounding those targets supports therapeutic response.
This shift—from target identification to target-in-context evaluation—can influence indication selection, biomarker strategy, combination development and trial design. It may also allow teams to identify biological mismatches earlier, before they become costly late-stage failures.
AI-enabled predictive intelligence cannot eliminate uncertainty from oncology development. Used thoughtfully, however, it can help teams ask a more consequential question: not merely where the target is found, but where the therapy has the biological opportunity to succeed.
Connect with our oncology experts to explore indication selection, biomarker strategy and trial design through a target-in-context lens.
Contributors
Muaiad Kittaneh, MD, MBA, FACP
Head of R&D and Scientific Strategy
Related Content
Looking for exclusive insight?
