Although artificial neural networks simulate a variety of human functions, their internal structures are hard to interpret. In the life sciences, extensive knowledge of cell biology provides an opportunity to design visible neural networks (VNNs) which couple the model’s inner workings to those of real systems. Here we develop DCell, a VNN embedded in the hierarchical structure of 2526 subsystems comprising a eukaryotic cell (http://d-cell.ucsd.edu/). Trained on several million genotypes, DCell simulates cellular growth nearly as accurately as laboratory observations. During simulation, genotypes induce patterns of subsystem activities, enabling in-silico investigations of the molecular mechanisms underlying genotype-phenotype associations. These mechanisms can be validated and many are unexpected; some are governed by Boolean logic. Cumulatively, 80% of the importance for growth prediction is captured by 484 subsystems (21%), reflecting the emergence of a complex phenotype. DCell provides a foundation for decoding the genetics of disease, drug resistance, and synthetic life.

BackgroundGlobal but predictable changes impact the DNA methylome as we age, acting as a type of molecular clock. This clock can be hastened by conditions that decrease lifespan, raising the question of whether it can also be slowed, for example, by conditions that increase lifespan. Mice are particularly appealing organisms for studies of mammalian aging; however, epigenetic clocks have thus far been formulated only in humans.ResultsWe first examined whether mice and humans experience similar patterns of change in the methylome with age. We found moderate conservation of CpG sites for which methylation is altered with age, with both species showing an increase in methylome disorder during aging. Based on this analysis, we formulated an epigenetic-aging model in mice using the liver methylomes of 107 mice from 0.2 to 26.0 months old. To examine whether epigenetic aging signatures are slowed by longevity-promoting interventions, we analyzed 28 additional methylomes from mice subjected to lifespan-extending conditions, including Prop1df/df dwarfism, calorie restriction or dietary rapamycin. We found that mice treated with these lifespan-extending interventions were significantly younger in epigenetic age than their untreated, wild-type age-matched controls.ConclusionsThis study shows that lifespan-extending conditions can slow molecular changes associated with an epigenetic clock in mice livers.Electronic supplementary materialThe online version of this article (doi:10.1186/s13059-017-1186-2) contains supplementary material, which is available to authorized users.

Gene networks are rapidly growing in size and number, raising the question of which networks are most appropriate for particular applications. Here, we evaluate 21 human genome-wide interaction networks for their ability to recover 446 disease gene sets identified through literature curation, gene expression profiling, or genome-wide association studies. While all networks have some ability to recover disease genes, we observe a wide range of performance with STRING, ConsensusPathDB, and GIANT networks having the best performance overall. A general tendency is that performance scales with network size, suggesting that new interaction discovery currently outweighs the detrimental effects of false positives. Correcting for size, we find that the DIP network provides the highest efficiency (value per interaction). Based on these results, we create a parsimonious composite network with both high efficiency and performance. This work provides a benchmark for selection of molecular networks in human disease research.

Although cancer genomes are replete with noncoding mutations, the effects of these mutations remain poorly characterized. Here we perform an integrative analysis of 930 tumor whole genomes and matched transcriptomes, identifying a network of 193 noncoding loci in which mutations disrupt target gene expression. These “somatic eQTLs” (expression Quantitative Trait Loci) are frequently mutated in specific cancer tissues, and the majority can be validated in an independent cohort of 3,382 tumors. Among these, we find that the effects of noncoding mutations on DAAM1, MTG2 and HYI transcription are recapitulated in multiple cancer cell lines, and that increasing DAAM1 expression leads to invasive cell migration. Collectively the noncoding loci converge on a set of core pathways, permitting a classification of tumors into pathway-based subtypes. The somatic eQTL network is disrupted in 88% of tumors, suggesting widespread impact of noncoding mutations in cancer.

A major ambition of artificial intelligence lies in translating patient data to successful therapies. Machine learning models face particular challenges in biomedicine, however, including handling of extreme data heterogeneity and lack of mechanistic insight into predictions. Here, we argue for "visible" approaches that guide model structure with experimental biology.