Edward Vitkin and Zohar Yakhini

Department of Computer Science, Technion - Israel Institute of Technology

Background

Efficient and sustainable conversion of biomass

into valuable products is a major challenge for

bioengineering.

The composition of the available feedstock

biomass and the ability of microorganisms to

efficiently ferment it are two most critical factors

influencing the process efficiency.

Even in the two-organism fermentation system, an

analysis of many promising scenarios may require

solution of millions of optimization tasks.

Simulations and computer-assisted optimization

are valuable tools for fermentation processes

designers.

We present BioLEGO, a Microsoft Azure Cloud-

based framework, delivering these heavy

calculations to unskilled users.

Modeling Scalability

Experimental Validation

Feedstock Biomass Fermentation Process Received Product

Class Metabolite Name Biomass A (g/100gDW) Biomass B …

Insoluble fibres (NDF)

Ulvan 23.68 23.68Hemicellulose 20.60 20.60Cellulose 9.13 9.13Lignin (ADL) 1.56 1.56

Monosaccharides

Glucose 17.24 17.24Rhamnose 7.40 7.40Xylose 1.93 1.93Galactose 0.35 0.35Mannose 0.29 0.29Arabinose 0.08 0.08

Amino acids

Aspartic acid 1.09 1.09Threonine 0.53 0.53Serine 0.59 0.59Glutamic acid 1.09 1.09Proline 0.35 0.35Glycine 0.56 0.56Alanine 0.77 0.77Valine 0.77 0.77Methionine 0.20 0.20Cystine 0.16 0.16Isoleucine 0.40 0.40Leucine 0.70 0.70Tyrosine 0.51 0.51Phenylalanine 0.23 0.23Histidine 0.12 0.12Lysine 0.55 0.55Arginine 0.52 0.52

Fatty acids

Myristic 0.19 0.19Palmitic 4.67 4.67Palmitoleic 0.54 0.54Stearic 0.15 0.15Oleic 1.25 1.25Linoleic 0.19 0.19a-Linolenic 0.25 0.25Arachidic 0.09 0.09Eicosenoic 0.12 0.12Behenic 0.33 0.33Docosahexaenoic 0.09 0.09

Compositions of Expected Feedstock Biomasses Potential products

Potential fermentation setups

• Fermentation Efficiency

• Per target product

• Per feedstock

• Per fermentation setup

• Beneficial adjustments

• Of feedstock

• Of fermenting organisms

Product NameEthanolButanolAcetone

Organisms, organism order and growth conditions

E. coli; no O2S. cerevisiae; with O2 S. cerevisiae followed by E.coli; with O2

Flux Balance Analysis (FBA) framework

Modular Approach – Single Module

Modular Approach – Combining Modules

Formulation of Optimization problem

The reaction stoichiometry in a metabolic

model is represented by matrix S, wherein

Sm,r corresponds to stoichiometric

coefficient of metabolite m in the reaction r.

The vector of metabolic fluxes that are

carried by the model reactions, normally

denoted as v, is constrained both by mass-

balance and by maximal/minimal feasible

fluxes vUB and vLB

Module interaction

rules for two-step

serial fermentation

process

Feedstock Media →

Media Org1

Waste Org1 →

Media Org2

Growth Org1 →

Media Org2

Product Org1 →

Total product

Waste Org2 →

Total waste

Growth Org2 →

Total waste

Product Org2 →

Total product

Runtime Statistics

Research Setup Number of Tasks Computational Setup Runtime

Exploratory search for optimal

biomass utilization setups for

distinct types of corn biomasses:

1. Corn Cobs

2. Corn Fiber

3. Corn Stover

• 240 model

configurations

• 480 single

fermentation task

simulations

• 10 X Model Setup Constructors • Standard A1v2 nodes (1 core, 2048MB)• 2 threads

• 10 X Single Fermentation Task Estimators• Standard A1v2 nodes (1 core, 2048MB) • 3 threads

21min

(1) Sensitivity and (2) Gradient

analyses for anaerobic two-step

ethanol production from

Kappaphycus alvarezzi

• 2 X 37 single

fermentation task

simulations

• 5 X Single Fermentation Task Estimators• Standard A1v2 nodes (1 core, 2048MB) • 3 threads 2 X

4min

Analysis of two-step

fermentation with knock-outs in

each organism:

1. S. cerevisiae (2,280 reactions)

2. E. coli (2,914 reactions)

• 6,649,115 single

fermentation task

simulations

• 45 X Single Fermentation Task Estimators• Standard D3v2 nodes (4 core, 14336MB) • 10 threads

LOCAL SERVER• 25 X Task calculation request threads• 10 X Task result processing threads

126hr*

* Stronger

Local Server

can improve

up to 50%

Heatmap of knockout performances

Analyzing 6.6M knockouts in two-step process

Use Case

Simulation Results

Reaction

pairs of

interest

GOrilla analysis (Enriched Functionality in results of S. cerevisiae)