The Missing Variability Problem in Cancer Prediction: Integrating Multilevel Factors

Efforts to predict cancer risk using genomic and environmental data remain limited because a substantial portion of risk variability is unaccounted for, leading to the missing variability problem (MVP). Current models struggle with individual-level discrimination even when incorporating genomic and environmental predictors, suggesting a difficulty in integrating factors distributed across different levels of biological organization. The MVP arises from the challenge of coordinating predictive factors operating at molecular, cellular, tissue, and organismal levels without a unified theoretical framework for multilevel integration.
The article proposes treating the cellular level as an integrative predictive level where heterogeneous factors can be coordinated within a single biological bearer. This involves characterizing cells by stable differences in variability generation across conditions, operationalized through measurable features like molecular noise and tissue response patterns. Downscaling to molecular processes focuses on molecular noise, while upscaling to tissue levels emphasizes tissue organization. The cellular level is proposed as the site where these molecular and tissue-level influences become jointly attributable.
The text suggests that true prediction requires moving beyond simple aggregation of predictors by finding a level where heterogeneous influences become jointly tractable. This level functions by integrating variability sources into observable features, such as cellular morphology, epigenetic states, and responsiveness to tissue cues, allowing for the coordination of disparate predictive factors.

* Efforts to predict cancer risk typically rely on genomic data and external environmental factors.
* Integrating genomic and environmental data accounts for only a limited portion of individual variability in cancer risk.
* Individual-level discrimination remains imperfect even when using polygenic risk scores and environmental predictors.
* The missing variability problem (MVP) is the inability to reliably predict individual cancer risk from genomic and environmental data.
* The MVP concerns the difficulty of integrating predictive factors distributed across different levels of biological organization.
* Efforts to reduce the predictive gap involve downscaling to molecular processes or upscaling to tissue-level organization.
* Downscaling focuses on molecular noise, which is noted as a source of variability.
* Upscaling focuses on tissue-level organization and developmental conditions.
* The cellular level is proposed as an integrative predictive level where heterogeneous factors can be coordinated.
* Cellular dispositions involve stable differences in how cells produce variability across conditions.
* These dispositions are operationalized through molecular noise and patterns of response to tissue-level influences.
* Molecular noise correlates with cellular characteristics like mitochondrial state and chromatin organization.
* Tissue-level parameters, such as extracellular matrix stiffness, show differential effects depending on the cellular state.

The narrative structures how a prediction failure stems not from insufficient data alone but from an inability to structure heterogeneous information across hierarchical biological scales. The shift from viewing cancer risk as determined by external factors or germline inheritance to recognizing variability generation at the cellular level represents a move from reductionist explanations to systems-based integration. This suggests that predictive power is fundamentally constrained by the framework used for organization, rather than just the quantity of information.
The concept of the integrative predictive level shifts focus from causal primacy (which level causes what) to epistemic function (how levels structure prediction). By positioning the cell as this level, the argument challenges the tendency in biology to privilege single scales, particularly molecular events or organismal totals. The recognition that cellular states—defined by dispositions like stemness—can serve as stable bearers for multi-level influences suggests a critical reorientation of what constitutes a valid unit for risk assessment.
The juxtaposition of downscaling (molecular noise) and upscaling (tissue organization) highlights the failure of siloed approaches, reinforcing the need for a framework that coordinates these scales. The finding that cellular dispositions—the coordination of molecular instability and tissue responsiveness—can integrate these disparate factors provides a mechanism for this coordination. This leads to an insight about potential knowledge gaps: existing biological theories often fail because they treat levels as purely causal rather than predictive integrators, leaving the practical problem of risk prediction unsolved until an integrative, context-aware attribution is established at the cellular level.
What constitutes the stable bearer—the cell—is not a conclusion drawn from prior biology but an operational necessity for achieving a functional predictive framework where variability can be tracked across conditions and lineages. The tension between dynamic, emergent processes (like stemness) and static phenotypic expressions (morphology) suggests that dispositional concepts bridge this divide, allowing the system to account for both underlying mechanisms and observable consequences simultaneously.
How do we establish criteria for when a cellular state is sufficiently stable to be treated as an integrative unit for prediction? If molecular noise is understood not as isolated fluctuation but as a correlated feature within the cellular configuration, and tissue signals are registered as phenotype-dependent thresholds, what new metrics of stability must be introduced to bridge the gap between dynamic process and static prediction? And if this integration is achieved by identifying dispositions, does that imply a move toward viewing phenotypic variation as the fundamental unit of biological inquiry, rather than molecular or tissue organization alone?