The authors assemble a computable approach to link cell-lineage architecture to organ-level bioenergetics. Using a curated 396-node human cell-lineage tree spanning the zygote to terminal somatic identities, they test whether adult organ standard metabolic rate (SMR) is better explained by developmental time (measured as lineage depth) or by terminal fate identity (anatomical compartment identity). The work aims to produce a formal, layered framework that connects single-cell–level annotations to emergent organ energetics and to propose hypotheses about when metabolic characteristics become fixed during development.
The dataset comprises 396 nodes representing human cell identities from the zygote through differentiated terminal cell types. Each terminal identity is annotated with three key features: mitochondrial abundance, dominant fuel pathway, and lineage of origin. These annotations provide the substrate for testing how cell-level metabolic properties aggregate to organ-level energetic rates.
Two competing explanatory variables are assessed for their ability to predict a cell’s assigned metabolic tier: (1) lineage depth, representing developmental time or the number of branches from zygote to terminal identity, and (2) terminal fate identity operationalized as anatomical compartment identity. Analyses show that lineage depth explains essentially none of the variance in metabolic tier (reported r = 0.11, R2 approximately 1.2%). In contrast, compartment identity explains roughly three-quarters of the variance (approximately 75%), indicating that where a cell ends up anatomically is far more informative about its metabolic characteristics than how long it took to reach that state along the lineage tree.
At the organ level, the authors compute the mean mitochondrial volume fraction of an organ’s constituent terminal cell types and compare this aggregate measure to classical literature-derived organ SMR values. Across five canonical reference-man organ groups, the mean mitochondrial fraction tracks organ SMR with a reported correlation of r = 0.90. This strong association supports the idea that terminal-cell mitochondrial content is a useful cellular-level proxy for organ bioenergetic rate.
Based on the empirical findings, the authors outline a five-layer computable framework intended to formalize how cell-lineage architecture gives rise to organ energetics. The five layers are described as:
These layers provide modular stages for computational modeling, analysis, and potential extension with machine-learning tools.
The authors propose the metabolic commitment-horizon model: the hypothesis that a cell’s metabolic tier becomes fixed at a discrete lineage-commitment event and remains constant thereafter, rather than changing continuously as differentiation proceeds. This model implies that metabolic identity is set at particular branching points in the lineage tree (commitment horizons) and that descendants inherit the metabolic tier established at or before that event.
To evaluate predictive models, the authors perform a worked held-out test using 5-fold cross-validation on the 396-node tree. They compare a nearest-commitment-ancestor predictor to two baselines: a global-mean null and a lineage-depth regression. The nearest-commitment-ancestor predictor outperformed the global-mean null (mean absolute error, MAE, 0.752 vs 0.809; p = 0.015). However, it did not significantly outperform the lineage-depth regression (delta MAE = -0.042, 95% CI -0.098 to +0.012, p = 0.175). The authors therefore report descriptive support for the commitment-horizon model but acknowledge that predictive discrimination against a lineage-depth model is not established with these data.
The authors connect their within-species observations to the classical, cross-species rate-of-living hypothesis, which links metabolic rate to lifespan and energetic scaling. They also interpret insulin resistance as an acquired perturbation of the proposed metabolic organ clock, suggesting that altered systemic regulation might modify the execution layer of the framework and thereby change organ-level energetics.
The paper surveys machine-learning and deep-learning tools that could be applied to primary single-cell data to extend and test the proposed framework. The authors close with eight testable predictions and falsification conditions to guide empirical validation. Specific experimental or computational next steps are implied but not elaborated in detail in the source abstract; the precise content of the eight predictions and falsification criteria are not reported in the abstract.
Both authors are employees of Ainnocence Inc., a company developing computational tools for drug discovery. The authors report that no external funding supported this work. The preprint is posted on bioRxiv and the authors retain copyright while granting bioRxiv a license to display the preprint.