bioRxiv Science⌕ Search
Home/Archive/bioRxiv

bioRxiv · 10.1101/135574

DeepMetabolism: A Deep Learning System To Predict Phenotype From Genome Sequencing

Guo, W.·Xu, Y. E.·Feng, X.
View original source ↗Export metadata ↓

Abstract

Life science is entering a new era of petabyte-level sequencing data. Converting such \"big data\" to biological insights represents a huge challenge for computational analysis. To this end, we developed DeepMetabolism, a biology-guided deep learning system to predict cell phenotypes from transcriptomics data. By integrating unsupervised pre-training with supervised training, DeepMetabolism is able to predict phenotypes with high accuracy (PCC>0.92), high speed (<30 min for >100 GB data using a single GPU), and high robustness (tolerate up to 75% noise). We envision DeepMetabolism to bridge the gap between genotype and phenotype and to serve as a springboard for applications in synthetic biology and precision medicine.

Explore related subjects

systems biologyPreprint

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Guo, W., Xu, Y. E., Feng, X.. 2017-05-09. DeepMetabolism: A Deep Learning System To Predict Phenotype From Genome Sequencing. https://doi.org/10.1101/135574

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related preprints

bioRxiv2026

Enzyme-Constrained Metabolic Modeling Reveals CO2-Dependent Growth Control in Picochlorum celeri

Picochlorum celeri grows 11.6 times faster at elevated CO2 than in air, yet the abundance of its metabolic enzymes is only 17% higher. To explain this, we reconstructed iPiccel954, a genome-scale metabolic model with 3525 reactions, 2973 metabolites and 954 genes, and eciPiccel954, its enzyme-constrained version limited by the measured enzyme abundance in air and at elevated CO2. Calibrating the enzyme-constrained models against the measured proteome showed that a similar protein budget supports very different growth depending on how it is allocated. With urea as the nitrogen source, urea carboxylase was the only step whose relaxation alone raised growth, by 627% in air and 78% at elevated CO2. Bottleneck analysis showed that growth is limited by many enzymes spread across the network in both conditions, that the limiting enzymes differ between the two, and that control over starch metabolism seen in air disappears at elevated CO2. Measured proteomes agreed with these predictions, since the predicted bottleneck enzymes were more abundant than other enzymes. Rapid growth in P. celeri therefore reflects how the protein budget is allocated more than how large it is. Therefore, eciPiccel954 provides a quantitative basis for engineering this alga for biotechnological application.

systems biology↗
bioRxiv2026

Simulating the dynamics of gene regulatory systems at cellular scale

The temporal nature of eukaryotic gene expression is the result of a complex, non-linear, dynamical system of regulatory processes. So far, models of such systems trade realism for tractability by using strict simplifying assumptions. These simplified models are often unfaithful to our biological understanding; however, without such simplifications, analytical solutions become more challenging and these dynamical systems must then be studied through simulation. Here, we explore the question of how this complex web of gene regulatory connections can be faithfully simulated for all tens of thousands of genes that usually define a eukaryotic organism. We present a proof-of-principle model and simulator, GeneRegulatorySystems.jl which is based on a biophysical chemical master equation that overcomes three key restrictions. First, it allows an exact stochastic simulation of fully non-linear gene expression dynamics, including feedback control, to parsimoniously capture cause and effect relationships of gene regulation. Further, it includes a general framework for implementing diverse mechanisms of gene control such as protein mediated degradation, RNAi, and protein complex formation. Finally, it scales realistically to tens of thousands of genes. We achieve this by introducing a domain-specific programming language to aid the design of experimental protocols from small-scale, well-studied, scenarios of gene regulation inspired by dynamical systems biology up to large-scale, single cell omics experiments with interventions. In this way, our framework provides a grammar of in silico experimental design that can capture single-cell gene expression dynamics for entire genomes.

systems biology↗
bioRxiv2026

Mixing, competence, and commitment set the stability of cell identity

Cells maintain their identities in the face of molecular noise, yet can change them rapidly when cued. Here, we treat this tension as a problem in the control of a stochastic system. We represent a cell identity as a set of molecular configurations, or microstates, among which cells undergo stochastic transitions, and measure its stability by the effective rate of commitment to another identity, using a Markov model in which committed cells are removed. Mixing, competence, and commitment provide three channels for tuning this rate. They govern microstate transitions, how commitment is distributed among microstates, and the overall speed of commitment, respectively. Control over the effective commitment rate is shared between commitment and mixing in proportions that always sum to one. Interventions with the same effective commitment rate cannot be told apart by survival alone but, given the microstate transition rates, can be distinguished by the fraction of cells in each microstate or by how quickly cells sorted on permissive microstates commit. When competence varies in time, pulsed commitment signalling commits more cells than sustained signalling of equal dose, by an amount that grows with the depth of that variation, whereas with constant competence pulsing gains nothing in the models we studied. This suggests a possible function for pulsatile signalling, such as that of ERK, which the experiments we propose could test. Finally, reanalysing single-cell imaging of mouse embryonic stem cells leaving naive pluripotency, we estimate that cells take about as long to acquire competence as they then wait to commit, and we specify experiments to identify which channel drives this exit.

systems biology↗