Team:Tongji-Software/Project
PROJECT
What is Alpha Ant?
Already got a brilliant idea in metabolic engineering?
Too much information to search?
Still,
You need Alpha Ant as a guide
Alpha ant is a computational tool for pathway design and reconstruction. With full consideration of metabolic burden and some useful functions, we provide an efficient and powerful pathway design guide.
Why Alpha Ant?
Background
Pathway engineering has been proven indispensable in synthetic biology for its utility in design of microbes for generating value-added products, which is also the ultimate goal of our project. The core idea is to design and reconstruct pathway for proper use, including introducing heterologous metabolic reaction into a host organism, optimizing genetic processes within cells, modeling for yield prediction, flux balance analysis and so on.
However, it’s quite a challenge to design an efficient pathway while balancing the metabolic burden in certain organism. For example, it require sorting through thousands of possible reactions and enzymes wherein all the parameters have different substrate preferences and kinetic features. Also, valid evaluation and simulation of pathway in silico is indispensable. Of course, wet-lab experiments are necessary for pathway validation.
Inspiration
After investigating into metabolic pathway engineering, we realize that there is much work to do in a certain project. Commonly, we have to do a lot of research before we get started with actual experiment, such as database search, paper reading and so on. So that’s why we come up with this idea that help synthetic biologists do previous work in some ways.
Fig.1. Process of traditional pathway design.
In specific, we collect metabolic data from several databases including metabolic reactions, reaction main pairs, enzyme , gene , compound and so on. In this section, we provide all the related information of metabolic pathway. When it comes to ranking criteria, we choose some reliable ones and design novel ones to make our results more convincing. As for pathway search algorithm, we choose DFS (depth first search) because of its great performance in both speed and quality.
Origin of Name: Alpha Ant
Alpha Ant stands for an efficient and convenient tool for pathway engineering. Alpha means ‘origin’. In fact, Alpha Ant is the first software equipped with comprehensive ranking criteria.
Recently, it impressed people by the project “Alpha GO”, which also endowed “Alpha” with intelligence. Alpha Ant means its capacity to find the most efficient metabolic pathway linking two molecules is just like the ant colony’s intelligence of quickly organizing itself to find the most efficient path to a food source once it has been discovered by scouts.
So ants are great signal detector and way finder.
Fig.2.The meaning and origin of the name-Alpha Ant.
Applied Design
Data processing
We acquire metabolic reactions, gene information from KEGG. Standard Gibbs Energy are from MetaCyc and eQuilibrator. Furthermore, we obtain compound information from KEGG , ChEBI & KnowledgeBase. Enzyme data are from BRENDA and KEGG. The small molecular drug information is from DRUGBANK. Besides, we use MayaChemTools to calculate physiochemical properties of compounds.
Since we use so many databases, we came across some problem during data processing. Most challenging thing is to string all these information together because each database has its unique ID and special data format. We tried our best to integrate all these information and we hope our software can be useful to synthetic biologists.
Fig.3. Databases we use to integrate data.
Algorithm
Finding proper metabolic pathway is a typical search problem.
Consequently, we turn a biosynthesis problem into a directed graph search problem. Not only do we need to get all of the solutions that satisfy the constraints, but also need to record the search path.
Fig.4. we turn a biosynthesis problem into a directed graph search problem
We use DFS algorithm here. In theoretical computer science, DFS is typically used to traverse an entire graph, and takes time Θ(|V| + |E|),[2] linear in the size of the graph. The core idea of DFS is simple and elegant, so that it is convenient for us to introduce appropriate pruning algorithms based on the original algorithm. More details can be found in Model.
Fig.5. DFS(depth-first search).Order in which the nodes are visited.
Ranking criteria
In total, we have three ranking criteria, which are thermodynamic feasibility & competition of heterologous reactions, frequency of reaction and toxicity of compound. Rights are given to users to decide different weights of different ranking criteria. Many of you may think that length of pathway should be one of the ranking criteria , however, the fact is that the shortest pathway could be the most unrealistic one. So we decide not to use it.
1. Thermodynamic feasibility & competition of heterologous reactions
As we all known, thermodynamic feasibility of a certain reaction can decide the probability of reaction. In many occasions , the smaller standard Gibbs is, the more probability of reaction is. And so does competition of heterologous reactions. Enzymes, ribosomes and source compounds are possible things that may trigger
We compute the probability of each reaction with △rG through the Boltzmann distribution. According to study of Hiroyuki Kuwahara et.al[3], they derive a mathematical description of the weighting scheme. And in our software, we use this formula to compute and generate a score of each reaction.
(Equation 1)
2. Frequency of reaction in all organism
The idea came from our visit to Key Synthetic Biology Laboratory of Chinese Academy of Sciences.We were inspired by Prof.Yang who is committed to synthetic biology research. After an in-depth discussion with him, we found that living creatures are of great wisdom. They know how to make good use of energy from nature and develop their own core metabolic system. So the more frequency a certain reaction is, the better and much more efficient it is. There is no doubt that core pathways are more frequently used in organisms.
Consequently, we count the frequency of reaction in all orgainsms. During that process, the data distribution fully confirm that this ranking criteria is reasonable and reliable.Details can be found in Model.
3. Toxicity of compound
We use the data from KnowledgeBase to assess potential toxic effects of chemical compounds on certain organism. Then these effects will be taken into account according to the given weight when we calculate the total score.
Additional functions
To improve is to change, to be perfect is to change often.
At the beginning of the beginning, we only developed the most ordinary search function. After actively engaging with researchers, iGEM teams and social communities, we start to know their needs. All we need to do is to try our best to meet their needs. So we add those four functions, which are organism recommendation, atom conservation, flux balance analysis and SMILES comparison.
1. Organism recommendation
Don’t know which chassis cell to use?
We offer organism recommendation function for experimenters. Based on this purpose, we develop a model to scoring each microorganism (details can be found in model section). After ranking all those score, we provide users with common used organisms in iGEM white list. At the same time, we provide links of related information about organism.
2. Atom conservation
Given a chemical reaction, an atom mapping rule defines which atom of a substrate compound is transferred to which atom of a product compound [4]. This is helpful for many applications of system biology, in particular in metabolic pathway engineering. Reducing the loss of atoms from the start compound to the target compound is likely to provide good route candidates for pathway design. Here, we present users with atom conservation rate of different reactions.
3.Flux Balance Analysis
Flux balance analysis (FBA) is a mathematical method for simulating metabolism in genome-scale reconstructions of metabolic networks. It can evaluate the metabolic flux distribution, and is one of the most used modeling approaches for metabolic systems. In comparison to traditional methods of modeling, FBA is less intensive in terms of the input data required for constructing the model. Simulations performed using FBA are computationally inexpensive and can calculate steady-state metabolic fluxes for large models (over 2000 reactions) in a few seconds on modern personal computers.
Users can select one from pathway search result. Since E.coli is the most frequently used host organism, we will analyze the selected pathway and construct a new model based on classic E.coli core model(from biomodel.com). After simulating this model, our software will provide quantitative predictions of metabolic flux patterns by using cobra toolbox which provides insights into the metabolic pathways [5].
4. SMILES comparison
Original thinking about this topic is derived from our visits to WuXi AppTec. Experts of WuXi AppTec proposed an idea to us. They said that sometimes their company got or designed a novel compound which did not exist in current database, and they want to find a possible way to synthesize it. So it came to our mind that what if we could compare the similarity between different compounds and select the most similar compound as a trigger to help us design new synthetic pathway, which can be very useful in small molecular drug discovery and synthesis.
First of all, we convert user’s input SMILES into molecular fingerprints by using RDkit toolbox. Then we compute similarity score between input compound and compound in databases by comparing their fingerprints. At last, output is similarity score and a ranking list.
The best thing is that we can search not only novel compound, but also existing compound in database. So if you get a compound with structure information and you don’t know what it is, you will find its compound ID and name by using SMILES comparison.
Validation
Pathway Search Validation
We validate Alpha Ant’s outcome against published pathways that were engineered into different organisms. We compare search result with Gil and MRE[3] by using reliable pathways in published paper. The results show that Alpha Ant can perform its intended function well. It identifies several pathways that are known to be productive. The results show as follows.
Case study 1:
Pathway for the production of flavonoids from glucose:
Flavonoids comprise a large family of secondary plant metabolic intermediates that exhibit a wide variety of antioxidant and human health-related properties. However, their wide spread use and availability are currently limited by inefficiencies in both their chemical synthesis and extraction from natural plant sources. [6]As a result, significant strides have been made recent years in improving the microbial production of flavonoids. There are four steps of pathway that are known to be productive for the conversion of L-tyrosine to naringenin(C00509), the main flavonoid precursor. [7]
Parameters in Alpha Ant: ECO maxlength(8) weight:Gibbs(0.6)/Toxicity(0.1)/Frequency(0.3)
Fig.6. Comparison after searching from L-tyrosine to naringenin.
We searched for pathways from L-tyrosine to naringenin in Alpha Ant. The results show that this productive pathway has the highest score in our outcomes, which is also same in MRE. But we didn’t find this path in the top 10 results in the Gil which is developed by Korea_U_Seoul.
Case study 2:
Production of 1,2-PD(C02912):
The individual enantiomers (R-1,2-PD and S-1,2-PD) have potential uses as chiral synthons for the production of pharmaceuticals and novel polymers; however, their use is limited due to their high cost. We applied Alpha Ant to search for pathways of biological production of 1,2-PD from glucose. Most of the top eight pathways contain the core part of converting Glycerone phosphate to (R)-Propane-1,2-diol which is published in literature.[8]
Parameters in Alpha Ant: S.cerevisiae maxlength(8) weight:Gibbs(0.5)/Toxicity(0.1)/Frequency(0.4)
Fig.7. Yu et al. Development of a Saccharomyces cerevisiae strain for increasing the accumulation of triacylglycerol as a microbial oil feedstock for biodiesel production using glycerol as a substrate.
Fig.8. Most of the top eight pathways contain the core part of converting Glycerone phosphate to (R)-Propane-1,2-diol which is published in literature.
Fig.9.Top pathways in MRE do not contain the core part of converting Glycerone phosphate to (R)-Propane-1,2-diol
No pathway could be found in Gil. Top pathways in MRE do not contain the core part of converting Glycerone phosphate to (R)-Propane-1,2-diol
Case study 3:
Artemisinin (C20309) is a sesquiterpene lactone endoperoxide extracted from Artemisia annua L with highly effective against multidrug-resistant Plasmodium spp. The semi-synthesis of artemisinin or any derivative from microbially sourced artemisinic acid, its immediate precursor, could be a cost-effective, environmentally friendly, high-quality and reliable source of artemisinin. The study of Jay D. Keasling etc. designed and constructed an engineered artemisinic acid biosynthetic pathway in S. cerevisiae strain EPY224 that is productive. [9]The biochemical pathway leading from farnesyl pyrophosphate (FPP) to artemisinic acid was introduced into S. cerevisiae from A. annua.
Parameters in Alpha Ant: yeast maxlength(8) weight:Gibbs(0.6)/Toxicity(0.1)/Frequency(0.3)
Fig. 10. Comparison after searching from farnesyl pyrophosphate (FPP) to artemisinic acid.
We searched for pathways from L-tyrosine to naringenin in Alpha Ant. This published pathway can be found in Alpha Ant, which demonstrates that Alpha Ant can perform its intended function well. This pathway couldn’t be found in Gil.
Case study 4:
Production of 1,3-propanediol(C02457) Escherichia coli K-12 ER2925:
The monomeric form of 1,3-propanediol (1,3-PD) has gained use in large-volume production of polyester fibers and polyurethanes in recent years. In order to develop an improved and more environmentally favorable process for 1,3-PD production, many researchers engaged to explore methods for 1,3-PD production via the microbial fermentation. Alpha Ant is able to identify the reported efficient pathway from glycerol to 1,3-PD. [10]
Parameters in Alpha Ant: ECO maxlength(8) weight:Gibbs(0.6)/Toxicity(0.1)/Frequency(0.3)
Fig. 11. Alpha Ant is able to identify the reported efficient pathway from glycerol to 1,3-PD.
Alpha Ant is able to identify the reported efficient pathway from glycerol to 1,3-PD. MRE shows the same result. The pathway can not be found in Gil.
SMILES Comparison Validation
In order to evaluate the function of SMILES similarity, we got structure information of several molecule drugs from DrugBank. We searched for the similar compound in Alpha Ant. The results showed that the molecular with same structure always has the highest score 1, which indicate that this function works well.
Here are some examples.
Artemisinin
Fig. 12. Structure information of Artemisinin from DrugBank and KEGG.
SMILES from DrugBank: [H][C@@]1(C)CC[C@@]2([H])[C@@]([H])(C)C(=O)O[C@]3([H])O[C@@]4(C)CC[C@]1([H])[C@@]23OO4
Fig. 13. Search result of Alpha Ant.
You can clear see that top one result of Alpha Ant is same as KEGG ID. Besides, the similarity score is 1, which infers highly similarity.
Morphine
Fig. 14. Structure information of Morphine from DrugBank and KEGG.
SMILES from DrugBank:[H][C@@]12OC3=C(O)C=CC4=C3[C@@]11CCN(C)[C@]([H])(C4)[C@]1([H])C=C[C@@H]2O
Fig. 15. Search result of Alpha Ant.
Result is in descending order of similarity score. Top one search result is Morphine, same as DrugBank structure information.
Acetylsalicylic acid
Fig. 16. Structure information of Morphine from DrugBank and KEGG.
SMILES from DrugBank: CC(=O)OC1=CC=CC=C1C(O)=O
Fig. 17. Search result of Alpha Ant.
Top one search result is Aspirin, same as DrugBank structure information. Also, Compound ID is same as KEGG,which is C01405.
Through validation we make above, it is convincing that our software is functional and reliable. Alpha Ant performs its intended function under safety control, which is our original design. We hope that Alpha Ant could be actually useful to synthetic biologist.
Demonstrate
Alpha Ant can work well under realistic conditions and our system complies all rules and regulations approved by the iGEM Safty Committee. We will show the demonstration of our work below in four parts which are demonstration video, computational performance, safety and statement.
Demonstration Video
Alpha Ant is directly available on the web and can work well under different search engines. You only need to input the url: Alpha Ant and then use our software on the website conveniently. The video below will show you how to use and apply Alpha Ant for pathway design.
Computational Performance
Other than the video, we also test computational performance of Alpha Ant by recording search speed of ordinary search and one-direction search. We randomly selected 50 pairs of source and target compounds for ordinary search and 50 compounds for one-direction search. The results show that it took less than 5s for Alpha Ant to exhaustively explore routes within 8 steps on average for ordinary search and less than 5s for one-direction search within 6 steps. This speed is relatively friendly to the user. The results show below.
Fig.18. Ordinary search speed test. When we search a pathway with different length(4-10), the time consumed differs.
Fig. 19. One-direction search speed test.
Safety
Q1: Who will use your software? What will the user concern about?
Alpha Ant:
1. For synthetic biologist
Pathway engineering: design proper metabolic pathways while taking into account several criteria such as thermodynamic feasibility, material competition of heterologous reactions, atom conservation, toxicity of intermediates.
2. For environmentalist
Find proper Microorganism to make some harmful substances degrade. The purpose is to be helpful to improve the environment.
3. For chemist
Find more reliable and convenient path to synthesis target compound (cell-free synthetic biology)
4. For drug researcher
By using our software, researchers can study the small molecular drug and explore new pathway to synthesize it
To be continued ...
Q2: Are there any safety problem of biobricks in Alpha Ant we need to concern about?
Alpha Ant:
We provide users with enzyme gene information of Biobricks, including sequence information and physiochemical characteristics. The core idea is to design proper and efficient pathway and recommend mainly enzyme gene for users’ need. Consequently, users can get gene information. After that, they can design new biobricks or search existing one by their own.
Q3: What about chassis cell? Since your software can recommend chassis cell for users, have you ever think about safety of your chassis cell?
Alpha Ant:
In our microorganism recommendation system, we only recommend common used organism in iGEM and organisms in iGEM white list(link). We want ensure the biosafety of our project and people who use our software.
Q4: How many compounds and enzyme genes do you have in your software project?
Fig. 20. Data comparison among Alpha Ant, Gil and KEGG.
As data shows above, we have more data than Gil and almost cover all KEGG reaction,compound data. Besides, we have 630275 gene data, same as KEGG while Gil only have 1950 E.coli K12 gene information.
Q5: Do users have any way to access your data, or change data without any limitations?
Alpha Ant:
Our data has a very neat format, mostly dictionary format. They are all stored in excel files. Users can simply search metabolic pathway in the browser and don’t need to download any files to use our software. So it’s quite convenient for them. Generally, users can not change our data in web client. But they can contact us and exchange suggestions. By the way, full data can be downloaded in github(link).
Statement
We provide Alpha Ant as a public service. We do not collect any personally identifiable information (PII) about you when you visit our Web site. Alpha Ant is only for academic research and will be not for any commercial uses. The sources of data we obtained from other databases are as follows.
Copyright:
KEGG:
Minoru Kanehisa, Miho Furumichi, Mao Tanabe, Yoko Sato, Kanae Morishima; KEGG: new perspectives on genomes, pathways, diseases and drugs, Nucleic Acids Research, Volume 45, Issue D1, 4 January 2017, Pages D353–D361
https://doi.org/10.1093/nar/gkw1092
Metacyc:
Ron Caspi, Richard Billington, Carol A Fulcher, Ingrid M Keseler, Anamika Kothari, Markus Krummenacker, Mario Latendresse, Peter E Midford, Quang Ong, Wai Kit Ong, Suzanne Paley, Pallavi Subhraveti, Peter D Karp; The MetaCyc database of metabolic pathways and enzymes, Nucleic Acids Research, Volume 46, Issue D1, 4 January 2018, Pages D633–D639
https://doi.org/10.1093/nar/gkx935
BRENDA:
Sandra Placzek, Ida Schomburg, Antje Chang, Lisa Jeske, Marcus Ulbrich, Jana Tillack, Dietmar Schomburg; BRENDA in 2017: new perspectives and new tools in BRENDA, Nucleic Acids Research, Volume 45, Issue D1, 4 January 2017, Pages D380–D388
https://doi.org/10.1093/nar/gkw952
BRENDA is available at www.brenda-enzymes.org
eQuilibrator:
Flamholz, A., Noor, E., Bar-Even, A., & Milo, R. (2012). eQuilibrator—the biochemical thermodynamics calculator. Nucleic Acids Research, 40(Database issue), D770–D775.
http://doi.org/10.1093/nar/gkr874
ChEBI:
Hastings J, Owen G, Dekker A, Ennis M, Kale N, Muthukrishnan V, Turner S, Swainston N, Mendes P, Steinbeck C. ChEBI in 2016: Improved services and an expanding collection of metabolites. Nucleic Acids Res. 2016 Jan;44(D1) D1214-9. PMID: 26467479; PMCID: PMC4702775.
http://doi:10.1093/nar/gkv1031
Drugbank:
David S. Wishart, Craig Knox, An Chi Guo, Savita Shrivastava, Murtaza Hassanali, Paul Stothard, Zhan Chang, Jennifer Woolsey; DrugBank: a comprehensive resource for in silico drug discovery and exploration, Nucleic Acids Research, Volume 34, Issue suppl_1, 1 January 2006, Pages D668–D672
https://doi.org/10.1093/nar/gkj067
Improvement
In 2015, Team:Korea_U_Seoul had developed a software called Gil, which is also a pathway finding tool. They did an excellent job in iGEM competition. For us, it’s a great honor to make some improvement on their project.
Gil has four ranking criteria which are ATP, NADPH, NADH and CO2 in different pathways for users’ various needs and they analyze thermodynamic feasibility. The idea is great. They concerned a lot about production of ATP, NADPH, NADH and CO2. The more NADPH,NADH,ATP that pathway produce, the more efficient that the pathway is.
We have to admit that the energy related compound production is important. However, it pays attention to yield instead of describing and evaluating metabolic burden in organism. So our project will do both. We use flux balance analysis to maximize yield and use three different criteria to evaluate the possibility and quality of new designed pathway.
We have done a lot work to improve and make advance in metabolic engineering tool.
1. We can optimize search algorithm;
We use depth-first search algorithm. Comparing to other search algorithm, DFS is a classic method and it just fit our need to find all possible pathways. Besides, using DFS is faster than other algorithm due to our huge network. Detailed information can be found in Model.
2. We can add more criteria and give them different weights;
We select thermodynamic feasibility, heterologous competition, compound toxicity , atom conservation and frequency of reactions in various organism as our ranking criteria. We will give a recommend weight, of course, users can give them different weight which will result in different output list.Detailed information can be found in Project.
Fig. 21. Search page of Alpha Ant.
3. We could recommend chassis cell for users;
After investigating the users’ different need, we would like to provide more convenience to them. Characteristics of different chassis cells are of great diversity. The core idea is to recommend microorganism as chassis cell and offer some related information about it. We build a model to scoring them and give them a rank. See Model
Fig. 22. Organism recommendation page in Alpha Ant.
4. As for novel compound or newly designed compound, we could find a regular compound with similar structure and explore possible synthesis pathway.
We add SMILES comparison to our software as one of the additional function. For novel compound, it can function as novel pathway explorer ( recommend several most matched compound and related reaction and enzyme information ) ; for compound in database, it can function as database search. ( Because the similarity of most matched compound is 100%)
Fig. 23. SMILES comparison page in Alpha Ant.
5. FBA(Flux balance analysis)
FBA is another additional function of our software. For now, we only can do FBA in E.coli. Because E.coli is the most frequently used chassis cell and we know it much more than other microorganisms.
Fig. 24. Result page of Alpha Ant.
6. One-direction search
We want to provide more possibilities to everyone concerned. You only need to input one of source compound and target compound. As long as you enter one of them, you will come up with many possible outcomes, some of which may inspire you to think about the problem with new ideas.
Fig. 25. Result page of Alpha Ant.
7. Involve more data
As data shows below, we have more data than Gil and almost cover all KEGG reaction,compound data. Besides, we have 630275 gene data, same as KEGG while Gil only have 1950 E.coli K12 gene information.
Fig. 26. Data comparison among Alpha Ant, Gil and KEGG.
8. Thorough investigation and validation:more reliable
After getting plenty of practical suggestion from professors, bioscientists and company researchers, we improve our project in different aspects to make the results more reliable. In addition, our software has many humanized options. Users can adjust weight that given to different ranking criteria. When users input a certain compound, we will provide corresponding search hint. More information could be found in Human Practices.
Apart from things we mentioned above, we also did comparison in Validation section.
Future Plan
Future plan1: The optimum conditions for each reactions’ enzymes may be different, we will take into accont the positions and optimum conditions of each reaction’s enzyme when ranking pathway to guarantee that the recommended optimal pathway will achieve the highest efficiency under the same condition.
Future plan2:There are other factors should be taken into account when evaluating the pathway, such as primary metabolite or secondary metabolite,growth phase or fermentation period of microorganism. We learned that in the production of industrial compounds, producers tend to use microbial fermentation in an anaerobic environment to reduce the biomass energy consumption and achieve high productity. So we will take the types of metabolite and growth state of microorganism into account when optimising pathway.
Future plan3: We will consider about modyfing existing enzymes to improve efficiency of pathway. Enzymes are often engineered to carry out new functions in the context of synthetic metabolic pathways. Enzyme modification is a challenging and promising solution for improving pathway efficiency. We are planning to read related literature and develop this excitiing funcion afterwards.
Reference:
[1] Manish Sud .MayaChemTools: An Open Source Package for Computational Drug Discovery. Journal of Chemical Information and Modeling , 2016 , 56 (12), 2292-2297.
[2] Thomas H. Cormen, Charles E. Leiserson, Ronald L. Rivest, and Clifford Stein. Introduction to Algorithms, Second Edition. MIT Press and McGraw-Hill, 2001. ISBN 0-262-03293-7. Section 22.3: Depth-first search, pp. 540–549.
[3] Hiroyuki Kuwahara, Meshari Alazmi, Xuefeng Cui and Xin Gao. MRE: a web tool to suggest foreign enzymes for the biosynthesis pathway design with competing endogenous reactions in mind. Nucleic Acids Research, 2016, Vol. 44, Web Server issue W217–W225.
[4] Jeremiah P. Malerich, Mike Travers, and Peter D. Karp. Accurate Atom-Mapping Computation for Biochemical Reactions Mario Latendresse. Journal of Chemical Information and Modeling 2012 52 (11), 2970-2982.
[5] Schellenberger J, Que R, Fleming RMT, Thiele I, Orth JD, Feist AM, Zielinski DC, Bordbar A, Lewis NE, Rahmanian S et al., Quantitative prediction of cellular metabolism with constraint-based models: the COBRA Toolbox v2.0 Nature Protocol, 2011,6(9):1290-307.
[6] Fowler, Z.L., W.W. Gikandi, and M.A.J.A.E.M. Koffas, Increased Malonyl Coenzyme A Biosynthesis by Tuning the Escherichia coli Metabolic Network and Its Application to Flavanone Production. 2009. 75(18): p. 5831-5839.
[7] Santos, C.N., M. Koffas, and G.J.M.E. Stephanopoulos, Optimization of a heterologous pathway for the production of flavonoids from glucose. 2011. 13(4): p. 392-400.
[8] Yu, K.O., et al., Development of a Saccharomyces cerevisiae strain for increasing the accumulation of triacylglycerol as a microbial oil feedstock for biodiesel production using glycerol as a substrate. 2012. 110(1): p. 343-347.
[9] DK, R., et al., Production of the antimalarial drug precursor artemisinic acid in engineered yeast. 2006. 440(7086): p. 940-943.
[10] Tang, X., et al., Microbial conversion of glycerol to 1,3-propanediol by an engineered strain of Escherichia coli. 2009. 75(6): p. 1628.
Fig.1. Process of traditional pathway design.
In specific, we collect metabolic data from several databases including metabolic reactions, reaction main pairs, enzyme , gene , compound and so on. In this section, we provide all the related information of metabolic pathway. When it comes to ranking criteria, we choose some reliable ones and design novel ones to make our results more convincing. As for pathway search algorithm, we choose DFS (depth first search) because of its great performance in both speed and quality.
Origin of Name: Alpha Ant
Alpha Ant stands for an efficient and convenient tool for pathway engineering. Alpha means ‘origin’. In fact, Alpha Ant is the first software equipped with comprehensive ranking criteria.
Recently, it impressed people by the project “Alpha GO”, which also endowed “Alpha” with intelligence. Alpha Ant means its capacity to find the most efficient metabolic pathway linking two molecules is just like the ant colony’s intelligence of quickly organizing itself to find the most efficient path to a food source once it has been discovered by scouts.
So ants are great signal detector and way finder.
Fig.2.The meaning and origin of the name-Alpha Ant.
Applied Design
Data processing
We acquire metabolic reactions, gene information from KEGG. Standard Gibbs Energy are from MetaCyc and eQuilibrator. Furthermore, we obtain compound information from KEGG , ChEBI & KnowledgeBase. Enzyme data are from BRENDA and KEGG. The small molecular drug information is from DRUGBANK. Besides, we use MayaChemTools to calculate physiochemical properties of compounds.
Since we use so many databases, we came across some problem during data processing. Most challenging thing is to string all these information together because each database has its unique ID and special data format. We tried our best to integrate all these information and we hope our software can be useful to synthetic biologists.
Fig.3. Databases we use to integrate data.
Algorithm
Finding proper metabolic pathway is a typical search problem.
Consequently, we turn a biosynthesis problem into a directed graph search problem. Not only do we need to get all of the solutions that satisfy the constraints, but also need to record the search path.
Fig.4. we turn a biosynthesis problem into a directed graph search problem
We use DFS algorithm here. In theoretical computer science, DFS is typically used to traverse an entire graph, and takes time Θ(|V| + |E|),[2] linear in the size of the graph. The core idea of DFS is simple and elegant, so that it is convenient for us to introduce appropriate pruning algorithms based on the original algorithm. More details can be found in Model.
Fig.5. DFS(depth-first search).Order in which the nodes are visited.
Ranking criteria
In total, we have three ranking criteria, which are thermodynamic feasibility & competition of heterologous reactions, frequency of reaction and toxicity of compound. Rights are given to users to decide different weights of different ranking criteria. Many of you may think that length of pathway should be one of the ranking criteria , however, the fact is that the shortest pathway could be the most unrealistic one. So we decide not to use it.
1. Thermodynamic feasibility & competition of heterologous reactions
As we all known, thermodynamic feasibility of a certain reaction can decide the probability of reaction. In many occasions , the smaller standard Gibbs is, the more probability of reaction is. And so does competition of heterologous reactions. Enzymes, ribosomes and source compounds are possible things that may trigger
We compute the probability of each reaction with △rG through the Boltzmann distribution. According to study of Hiroyuki Kuwahara et.al[3], they derive a mathematical description of the weighting scheme. And in our software, we use this formula to compute and generate a score of each reaction.
(Equation 1)
2. Frequency of reaction in all organism
The idea came from our visit to Key Synthetic Biology Laboratory of Chinese Academy of Sciences.We were inspired by Prof.Yang who is committed to synthetic biology research. After an in-depth discussion with him, we found that living creatures are of great wisdom. They know how to make good use of energy from nature and develop their own core metabolic system. So the more frequency a certain reaction is, the better and much more efficient it is. There is no doubt that core pathways are more frequently used in organisms.
Consequently, we count the frequency of reaction in all orgainsms. During that process, the data distribution fully confirm that this ranking criteria is reasonable and reliable.Details can be found in Model.
3. Toxicity of compound
We use the data from KnowledgeBase to assess potential toxic effects of chemical compounds on certain organism. Then these effects will be taken into account according to the given weight when we calculate the total score.
Additional functions
To improve is to change, to be perfect is to change often.
At the beginning of the beginning, we only developed the most ordinary search function. After actively engaging with researchers, iGEM teams and social communities, we start to know their needs. All we need to do is to try our best to meet their needs. So we add those four functions, which are organism recommendation, atom conservation, flux balance analysis and SMILES comparison.
1. Organism recommendation
Don’t know which chassis cell to use?
We offer organism recommendation function for experimenters. Based on this purpose, we develop a model to scoring each microorganism (details can be found in model section). After ranking all those score, we provide users with common used organisms in iGEM white list. At the same time, we provide links of related information about organism.
2. Atom conservation
Given a chemical reaction, an atom mapping rule defines which atom of a substrate compound is transferred to which atom of a product compound [4]. This is helpful for many applications of system biology, in particular in metabolic pathway engineering. Reducing the loss of atoms from the start compound to the target compound is likely to provide good route candidates for pathway design. Here, we present users with atom conservation rate of different reactions.
3.Flux Balance Analysis
Flux balance analysis (FBA) is a mathematical method for simulating metabolism in genome-scale reconstructions of metabolic networks. It can evaluate the metabolic flux distribution, and is one of the most used modeling approaches for metabolic systems. In comparison to traditional methods of modeling, FBA is less intensive in terms of the input data required for constructing the model. Simulations performed using FBA are computationally inexpensive and can calculate steady-state metabolic fluxes for large models (over 2000 reactions) in a few seconds on modern personal computers.
Users can select one from pathway search result. Since E.coli is the most frequently used host organism, we will analyze the selected pathway and construct a new model based on classic E.coli core model(from biomodel.com). After simulating this model, our software will provide quantitative predictions of metabolic flux patterns by using cobra toolbox which provides insights into the metabolic pathways [5].
4. SMILES comparison
Original thinking about this topic is derived from our visits to WuXi AppTec. Experts of WuXi AppTec proposed an idea to us. They said that sometimes their company got or designed a novel compound which did not exist in current database, and they want to find a possible way to synthesize it. So it came to our mind that what if we could compare the similarity between different compounds and select the most similar compound as a trigger to help us design new synthetic pathway, which can be very useful in small molecular drug discovery and synthesis.
First of all, we convert user’s input SMILES into molecular fingerprints by using RDkit toolbox. Then we compute similarity score between input compound and compound in databases by comparing their fingerprints. At last, output is similarity score and a ranking list.
The best thing is that we can search not only novel compound, but also existing compound in database. So if you get a compound with structure information and you don’t know what it is, you will find its compound ID and name by using SMILES comparison.
Validation
Pathway Search Validation
We validate Alpha Ant’s outcome against published pathways that were engineered into different organisms. We compare search result with Gil and MRE[3] by using reliable pathways in published paper. The results show that Alpha Ant can perform its intended function well. It identifies several pathways that are known to be productive. The results show as follows.
Case study 1:
Pathway for the production of flavonoids from glucose:
Flavonoids comprise a large family of secondary plant metabolic intermediates that exhibit a wide variety of antioxidant and human health-related properties. However, their wide spread use and availability are currently limited by inefficiencies in both their chemical synthesis and extraction from natural plant sources. [6]As a result, significant strides have been made recent years in improving the microbial production of flavonoids. There are four steps of pathway that are known to be productive for the conversion of L-tyrosine to naringenin(C00509), the main flavonoid precursor. [7]
Parameters in Alpha Ant: ECO maxlength(8) weight:Gibbs(0.6)/Toxicity(0.1)/Frequency(0.3)
Fig.6. Comparison after searching from L-tyrosine to naringenin.
We searched for pathways from L-tyrosine to naringenin in Alpha Ant. The results show that this productive pathway has the highest score in our outcomes, which is also same in MRE. But we didn’t find this path in the top 10 results in the Gil which is developed by Korea_U_Seoul.
Case study 2:
Production of 1,2-PD(C02912):
The individual enantiomers (R-1,2-PD and S-1,2-PD) have potential uses as chiral synthons for the production of pharmaceuticals and novel polymers; however, their use is limited due to their high cost. We applied Alpha Ant to search for pathways of biological production of 1,2-PD from glucose. Most of the top eight pathways contain the core part of converting Glycerone phosphate to (R)-Propane-1,2-diol which is published in literature.[8]
Parameters in Alpha Ant: S.cerevisiae maxlength(8) weight:Gibbs(0.5)/Toxicity(0.1)/Frequency(0.4)
Fig.7. Yu et al. Development of a Saccharomyces cerevisiae strain for increasing the accumulation of triacylglycerol as a microbial oil feedstock for biodiesel production using glycerol as a substrate.
Fig.8. Most of the top eight pathways contain the core part of converting Glycerone phosphate to (R)-Propane-1,2-diol which is published in literature.
Fig.9.Top pathways in MRE do not contain the core part of converting Glycerone phosphate to (R)-Propane-1,2-diol
No pathway could be found in Gil. Top pathways in MRE do not contain the core part of converting Glycerone phosphate to (R)-Propane-1,2-diol
Case study 3:
Artemisinin (C20309) is a sesquiterpene lactone endoperoxide extracted from Artemisia annua L with highly effective against multidrug-resistant Plasmodium spp. The semi-synthesis of artemisinin or any derivative from microbially sourced artemisinic acid, its immediate precursor, could be a cost-effective, environmentally friendly, high-quality and reliable source of artemisinin. The study of Jay D. Keasling etc. designed and constructed an engineered artemisinic acid biosynthetic pathway in S. cerevisiae strain EPY224 that is productive. [9]The biochemical pathway leading from farnesyl pyrophosphate (FPP) to artemisinic acid was introduced into S. cerevisiae from A. annua.
Parameters in Alpha Ant: yeast maxlength(8) weight:Gibbs(0.6)/Toxicity(0.1)/Frequency(0.3)
Fig. 10. Comparison after searching from farnesyl pyrophosphate (FPP) to artemisinic acid.
We searched for pathways from L-tyrosine to naringenin in Alpha Ant. This published pathway can be found in Alpha Ant, which demonstrates that Alpha Ant can perform its intended function well. This pathway couldn’t be found in Gil.
Case study 4:
Production of 1,3-propanediol(C02457) Escherichia coli K-12 ER2925:
The monomeric form of 1,3-propanediol (1,3-PD) has gained use in large-volume production of polyester fibers and polyurethanes in recent years. In order to develop an improved and more environmentally favorable process for 1,3-PD production, many researchers engaged to explore methods for 1,3-PD production via the microbial fermentation. Alpha Ant is able to identify the reported efficient pathway from glycerol to 1,3-PD. [10]
Parameters in Alpha Ant: ECO maxlength(8) weight:Gibbs(0.6)/Toxicity(0.1)/Frequency(0.3)
Fig. 11. Alpha Ant is able to identify the reported efficient pathway from glycerol to 1,3-PD.
Alpha Ant is able to identify the reported efficient pathway from glycerol to 1,3-PD. MRE shows the same result. The pathway can not be found in Gil.
SMILES Comparison Validation
In order to evaluate the function of SMILES similarity, we got structure information of several molecule drugs from DrugBank. We searched for the similar compound in Alpha Ant. The results showed that the molecular with same structure always has the highest score 1, which indicate that this function works well.
Here are some examples.
Artemisinin
Fig. 12. Structure information of Artemisinin from DrugBank and KEGG.
SMILES from DrugBank: [H][C@@]1(C)CC[C@@]2([H])[C@@]([H])(C)C(=O)O[C@]3([H])O[C@@]4(C)CC[C@]1([H])[C@@]23OO4
Fig. 13. Search result of Alpha Ant.
You can clear see that top one result of Alpha Ant is same as KEGG ID. Besides, the similarity score is 1, which infers highly similarity.
Morphine
Fig. 14. Structure information of Morphine from DrugBank and KEGG.
SMILES from DrugBank:[H][C@@]12OC3=C(O)C=CC4=C3[C@@]11CCN(C)[C@]([H])(C4)[C@]1([H])C=C[C@@H]2O
Fig. 15. Search result of Alpha Ant.
Result is in descending order of similarity score. Top one search result is Morphine, same as DrugBank structure information.
Acetylsalicylic acid
Fig. 16. Structure information of Morphine from DrugBank and KEGG.
SMILES from DrugBank: CC(=O)OC1=CC=CC=C1C(O)=O
Fig. 17. Search result of Alpha Ant.
Top one search result is Aspirin, same as DrugBank structure information. Also, Compound ID is same as KEGG,which is C01405.
Through validation we make above, it is convincing that our software is functional and reliable. Alpha Ant performs its intended function under safety control, which is our original design. We hope that Alpha Ant could be actually useful to synthetic biologist.
Demonstrate
Alpha Ant can work well under realistic conditions and our system complies all rules and regulations approved by the iGEM Safty Committee. We will show the demonstration of our work below in four parts which are demonstration video, computational performance, safety and statement.
Demonstration Video
Alpha Ant is directly available on the web and can work well under different search engines. You only need to input the url: Alpha Ant and then use our software on the website conveniently. The video below will show you how to use and apply Alpha Ant for pathway design.
Fig.18. Ordinary search speed test. When we search a pathway with different length(4-10), the time consumed differs.
Fig. 19. One-direction search speed test.
Safety
Q1: Who will use your software? What will the user concern about?
Alpha Ant:
1. For synthetic biologist
Pathway engineering: design proper metabolic pathways while taking into account several criteria such as thermodynamic feasibility, material competition of heterologous reactions, atom conservation, toxicity of intermediates.
2. For environmentalist
Find proper Microorganism to make some harmful substances degrade. The purpose is to be helpful to improve the environment.
3. For chemist
Find more reliable and convenient path to synthesis target compound (cell-free synthetic biology)
4. For drug researcher
By using our software, researchers can study the small molecular drug and explore new pathway to synthesize it
To be continued ...
Q2: Are there any safety problem of biobricks in Alpha Ant we need to concern about?
Alpha Ant:
We provide users with enzyme gene information of Biobricks, including sequence information and physiochemical characteristics. The core idea is to design proper and efficient pathway and recommend mainly enzyme gene for users’ need. Consequently, users can get gene information. After that, they can design new biobricks or search existing one by their own.
Q3: What about chassis cell? Since your software can recommend chassis cell for users, have you ever think about safety of your chassis cell?
Alpha Ant:
In our microorganism recommendation system, we only recommend common used organism in iGEM and organisms in iGEM white list(link). We want ensure the biosafety of our project and people who use our software.
Q4: How many compounds and enzyme genes do you have in your software project?
Fig. 20. Data comparison among Alpha Ant, Gil and KEGG.
As data shows above, we have more data than Gil and almost cover all KEGG reaction,compound data. Besides, we have 630275 gene data, same as KEGG while Gil only have 1950 E.coli K12 gene information.
Q5: Do users have any way to access your data, or change data without any limitations?
Alpha Ant:
Our data has a very neat format, mostly dictionary format. They are all stored in excel files. Users can simply search metabolic pathway in the browser and don’t need to download any files to use our software. So it’s quite convenient for them. Generally, users can not change our data in web client. But they can contact us and exchange suggestions. By the way, full data can be downloaded in github(link).
Statement
We provide Alpha Ant as a public service. We do not collect any personally identifiable information (PII) about you when you visit our Web site. Alpha Ant is only for academic research and will be not for any commercial uses. The sources of data we obtained from other databases are as follows.
Copyright:
KEGG:
Minoru Kanehisa, Miho Furumichi, Mao Tanabe, Yoko Sato, Kanae Morishima; KEGG: new perspectives on genomes, pathways, diseases and drugs, Nucleic Acids Research, Volume 45, Issue D1, 4 January 2017, Pages D353–D361
https://doi.org/10.1093/nar/gkw1092
Metacyc:
Ron Caspi, Richard Billington, Carol A Fulcher, Ingrid M Keseler, Anamika Kothari, Markus Krummenacker, Mario Latendresse, Peter E Midford, Quang Ong, Wai Kit Ong, Suzanne Paley, Pallavi Subhraveti, Peter D Karp; The MetaCyc database of metabolic pathways and enzymes, Nucleic Acids Research, Volume 46, Issue D1, 4 January 2018, Pages D633–D639
https://doi.org/10.1093/nar/gkx935
BRENDA:
Sandra Placzek, Ida Schomburg, Antje Chang, Lisa Jeske, Marcus Ulbrich, Jana Tillack, Dietmar Schomburg; BRENDA in 2017: new perspectives and new tools in BRENDA, Nucleic Acids Research, Volume 45, Issue D1, 4 January 2017, Pages D380–D388
https://doi.org/10.1093/nar/gkw952
BRENDA is available at www.brenda-enzymes.org
eQuilibrator:
Flamholz, A., Noor, E., Bar-Even, A., & Milo, R. (2012). eQuilibrator—the biochemical thermodynamics calculator. Nucleic Acids Research, 40(Database issue), D770–D775.
http://doi.org/10.1093/nar/gkr874
ChEBI:
Hastings J, Owen G, Dekker A, Ennis M, Kale N, Muthukrishnan V, Turner S, Swainston N, Mendes P, Steinbeck C. ChEBI in 2016: Improved services and an expanding collection of metabolites. Nucleic Acids Res. 2016 Jan;44(D1) D1214-9. PMID: 26467479; PMCID: PMC4702775.
http://doi:10.1093/nar/gkv1031
Drugbank:
David S. Wishart, Craig Knox, An Chi Guo, Savita Shrivastava, Murtaza Hassanali, Paul Stothard, Zhan Chang, Jennifer Woolsey; DrugBank: a comprehensive resource for in silico drug discovery and exploration, Nucleic Acids Research, Volume 34, Issue suppl_1, 1 January 2006, Pages D668–D672
https://doi.org/10.1093/nar/gkj067
Improvement
In 2015, Team:Korea_U_Seoul had developed a software called Gil, which is also a pathway finding tool. They did an excellent job in iGEM competition. For us, it’s a great honor to make some improvement on their project.
Gil has four ranking criteria which are ATP, NADPH, NADH and CO2 in different pathways for users’ various needs and they analyze thermodynamic feasibility. The idea is great. They concerned a lot about production of ATP, NADPH, NADH and CO2. The more NADPH,NADH,ATP that pathway produce, the more efficient that the pathway is.
We have to admit that the energy related compound production is important. However, it pays attention to yield instead of describing and evaluating metabolic burden in organism. So our project will do both. We use flux balance analysis to maximize yield and use three different criteria to evaluate the possibility and quality of new designed pathway.
We have done a lot work to improve and make advance in metabolic engineering tool.
1. We can optimize search algorithm;
We use depth-first search algorithm. Comparing to other search algorithm, DFS is a classic method and it just fit our need to find all possible pathways. Besides, using DFS is faster than other algorithm due to our huge network. Detailed information can be found in Model.
2. We can add more criteria and give them different weights;
We select thermodynamic feasibility, heterologous competition, compound toxicity , atom conservation and frequency of reactions in various organism as our ranking criteria. We will give a recommend weight, of course, users can give them different weight which will result in different output list.Detailed information can be found in Project.
Fig. 21. Search page of Alpha Ant.
3. We could recommend chassis cell for users;
After investigating the users’ different need, we would like to provide more convenience to them. Characteristics of different chassis cells are of great diversity. The core idea is to recommend microorganism as chassis cell and offer some related information about it. We build a model to scoring them and give them a rank. See Model
Fig. 22. Organism recommendation page in Alpha Ant.
4. As for novel compound or newly designed compound, we could find a regular compound with similar structure and explore possible synthesis pathway.
We add SMILES comparison to our software as one of the additional function. For novel compound, it can function as novel pathway explorer ( recommend several most matched compound and related reaction and enzyme information ) ; for compound in database, it can function as database search. ( Because the similarity of most matched compound is 100%)
Fig. 23. SMILES comparison page in Alpha Ant.
5. FBA(Flux balance analysis)
FBA is another additional function of our software. For now, we only can do FBA in E.coli. Because E.coli is the most frequently used chassis cell and we know it much more than other microorganisms.
Fig. 24. Result page of Alpha Ant.
6. One-direction search
We want to provide more possibilities to everyone concerned. You only need to input one of source compound and target compound. As long as you enter one of them, you will come up with many possible outcomes, some of which may inspire you to think about the problem with new ideas.
Fig. 25. Result page of Alpha Ant.
7. Involve more data
As data shows below, we have more data than Gil and almost cover all KEGG reaction,compound data. Besides, we have 630275 gene data, same as KEGG while Gil only have 1950 E.coli K12 gene information.
Fig. 26. Data comparison among Alpha Ant, Gil and KEGG.
8. Thorough investigation and validation:more reliable
After getting plenty of practical suggestion from professors, bioscientists and company researchers, we improve our project in different aspects to make the results more reliable. In addition, our software has many humanized options. Users can adjust weight that given to different ranking criteria. When users input a certain compound, we will provide corresponding search hint. More information could be found in Human Practices.
Apart from things we mentioned above, we also did comparison in Validation section.
Future Plan
Future plan1: The optimum conditions for each reactions’ enzymes may be different, we will take into accont the positions and optimum conditions of each reaction’s enzyme when ranking pathway to guarantee that the recommended optimal pathway will achieve the highest efficiency under the same condition.
Future plan2:There are other factors should be taken into account when evaluating the pathway, such as primary metabolite or secondary metabolite,growth phase or fermentation period of microorganism. We learned that in the production of industrial compounds, producers tend to use microbial fermentation in an anaerobic environment to reduce the biomass energy consumption and achieve high productity. So we will take the types of metabolite and growth state of microorganism into account when optimising pathway.
Future plan3: We will consider about modyfing existing enzymes to improve efficiency of pathway. Enzymes are often engineered to carry out new functions in the context of synthetic metabolic pathways. Enzyme modification is a challenging and promising solution for improving pathway efficiency. We are planning to read related literature and develop this excitiing funcion afterwards.
Reference:
[1] Manish Sud .MayaChemTools: An Open Source Package for Computational Drug Discovery. Journal of Chemical Information and Modeling , 2016 , 56 (12), 2292-2297.
[2] Thomas H. Cormen, Charles E. Leiserson, Ronald L. Rivest, and Clifford Stein. Introduction to Algorithms, Second Edition. MIT Press and McGraw-Hill, 2001. ISBN 0-262-03293-7. Section 22.3: Depth-first search, pp. 540–549.
[3] Hiroyuki Kuwahara, Meshari Alazmi, Xuefeng Cui and Xin Gao. MRE: a web tool to suggest foreign enzymes for the biosynthesis pathway design with competing endogenous reactions in mind. Nucleic Acids Research, 2016, Vol. 44, Web Server issue W217–W225.
[4] Jeremiah P. Malerich, Mike Travers, and Peter D. Karp. Accurate Atom-Mapping Computation for Biochemical Reactions Mario Latendresse. Journal of Chemical Information and Modeling 2012 52 (11), 2970-2982.
[5] Schellenberger J, Que R, Fleming RMT, Thiele I, Orth JD, Feist AM, Zielinski DC, Bordbar A, Lewis NE, Rahmanian S et al., Quantitative prediction of cellular metabolism with constraint-based models: the COBRA Toolbox v2.0 Nature Protocol, 2011,6(9):1290-307.
[6] Fowler, Z.L., W.W. Gikandi, and M.A.J.A.E.M. Koffas, Increased Malonyl Coenzyme A Biosynthesis by Tuning the Escherichia coli Metabolic Network and Its Application to Flavanone Production. 2009. 75(18): p. 5831-5839.
[7] Santos, C.N., M. Koffas, and G.J.M.E. Stephanopoulos, Optimization of a heterologous pathway for the production of flavonoids from glucose. 2011. 13(4): p. 392-400.
[8] Yu, K.O., et al., Development of a Saccharomyces cerevisiae strain for increasing the accumulation of triacylglycerol as a microbial oil feedstock for biodiesel production using glycerol as a substrate. 2012. 110(1): p. 343-347.
[9] DK, R., et al., Production of the antimalarial drug precursor artemisinic acid in engineered yeast. 2006. 440(7086): p. 940-943.
[10] Tang, X., et al., Microbial conversion of glycerol to 1,3-propanediol by an engineered strain of Escherichia coli. 2009. 75(6): p. 1628.
CONTACT
annecao@tongji.edu.cnADDRESS
1239# Siping RoadTongji University,Shanghai ChinaGET IN TOUCH