JC: Bayesian genome scale modelling identifies thermal determinants of yeast metabolism

Journal Club Presentation for LIF229

1. LIF229 Journ a l Club Present a tion — Rohit Gosw a mi Bayesian genome scale modelling identi f ies thermal determinants of yeast metabolism (Nature Comm. 2021) Authors: G a ng Li, Y a ting Hu, J a n Zrimec, H a o Luo, H a o W a ng, Aleksej Zelezni a k, Boy a ng Ji, Jens Nielsen 1
2. Generic Problem Overview Saccharomyces cerevisiae Temper a ture Constr a ints for Living Org a nisms 2 Temperature (C) Existence 10 30 Optimal Growth Alive 42 DEAD Images from Wikipedia • Small active range • ~12 degrees from optimal • Complex interplay of large molecules at multiple scales • Possibility of transferring mechanisms to orthologs • Like Kluyveromyces marxians
3. Complexity Scale First principle studies require unfeasible amounts of data from experiments including: • Enzymatic data • Protein structures • Kinetic modeling • Rate theory models • Free parameter estimation Generic Unfe a sible Solution 3 Image from Environ. Sci.: Processes Impacts,2017,19,188
4. Without excessive d a t a collection Predicting temperature e ff ects for biological organisms 4
5. Model Temperature Effects Key Point I (KP1) • Genome-scale metabolic models (GEMs) • Enzyme-constrained GEMs (ecGEMs) • Enzyme and Temperature-constrained GEMs (etcGEMs) 5 ecYeast7.6 to etcYeast7.6 • Non growth associated maintenance (NGAM) • Nonzero heat capacity b/w TS and ground
6. Experiment + Flux Balances . . Fin? 6 Key Point II Parameter Estimation Too many non unique parameters, and the experiments have uncertainty estimates as well • Aggravated by ML estimates for • Initial parameters work only below • Metabolic fl ux shift at • Needs more parameters!! • Or more data (unavailable) Topt 30∘C ∼ 36∘C − 38∘C
7. Augmenting Datasets Key Point III • Model process by describing a state space with a high dimensional probability distribution as a generator • Constrain / Update probability distribution with experimental data (using Bayes Theorem via Sequential Monte Carlo) • Simulate “data” by sampling fi nal probability distribution 7
8. Rich a rd P. Feynm a n “Experiment is the sole judge of the validity of any idea” 8
9. Analysing Yeast Growth and Temperature Dependence Results I • Existing studies report NGAM as the rate limiting factor • More likely that multiple factors contribute to the growth limits — Hypothesis 9 • NGAM alone only has a moderate e ff ect • Total growth retardation is due to the join e ff ect of NGAM, , and denaturation — result kcat •From top to bottom, the enzymes showed increased s •Each pixel represents one probability value of an enzyme Native State
10. Explaining Metabolic Shifts Results II • Fixed chemostat culture setting and dilution rate of • Corresponds to a metabolic shift from respiratory to partially fermentive 0.1h−1 10 • ADE3, PGK1, CDC19 are unstable • Ine ffi cient ATP production via the respiratory pathway • Proteome constraint observed — result • Maximal protein amount reached (saturation)
11. Detecting Enzymic Bottlenecks with etcGEM Results III • Growth rate is impaired due to multiple enzymes at — Hypothesis • Prediction via ensemble averaging • 82 enzymes predicted at least once 42∘C 11 • 24 predicted by more than 10% — result • Breakdown by function shown in the inset Simulated maximum speci fi c growth rate by removing the temperature constraints of most rate-limiting enzymes at each step in each Posterior etcGEM at 42 °C
12. The Good Journal Guideline Compliance • Complete dataset provided — Zenodo Repository — 2.4 GB tar file of raw data • Reproducible analysis — GitHub Repository — ~30 fi gures in .ipynb • Review responses — Here — 30 pages detailing 3 rounds of responses • Supplementary with model descriptions — Here — 30 pages on model validation • Open access (CC-BY 4.0) — 12 page article Can get a little overwhelming Summ a ry - I 12
13. Future Directions • The temperature dependent unfolding via the two-stage denaturation is not always valid • E ffi ciency of the Bayesian method can be improved • Some of the statistical parameters are not exactly in good agreement • The values from Tome show a large variability and should be improved • is only • Bayesian methods are normally checked with more posterior probability distribution tests instead of simple cross validation • The distribution may not always converge Topt R2 0.5 Summ a ry - II 13
14. Fin a l Summ a ry Useful Bayesian GEM extension with wide applicability + code 14
15. Questions? Thank You 15