In this lecture, I would like to survey some of the different types of biological networks that you will face in the field of systems biology and systems pharmacology. Cell signaling pathways are commonly represented as signed mixed graphs. Well, nodes are mostly proteins, but also can be metabolites, lipids, second messengers, or peptides. Interactions designate information flow and can be activation or inhibition. Most of such interactions are direct physical interactions. Interactions are typically enzymatic or binding. There are several pathway databases including the one highlighting here, which is sign signaling when there are others such as [INAUDIBLE] KEGG, ped or biocarta. Typically, biologists are organizing their knowledge about the pathways that they uncover in those diagrams that are called Cell Signaling Pathways. As more information is accumulating about cell signaling pathway, molecular cell biologists are beginning to realize that the signaling pathway view is limited. Since pathways are likely connected to form a large network. For my thesis project I developed from literature a large scale cell signaling network. For this I read many papers, over a thousand papers, describing biochemical regulatory interactions between similar components in a million runs. Extracted from these papers there was cell signaling interactions to form this large scale cell signaling network. You can see in this particular diagram. This network connects lag end receptor interactions to their downstream effectors. And those pathways are terminating the components of different cell of machines such as transcription, translation, secretion, motility and ion channel regulation. If you remember from the first introductory movies that show those various cell signaling pathways. As you can see the system is much more complex. And this is only representing maybe five to 10% of all the types of components and interactions. That are happening in the cell signaling network that controls the cell. So one thing that you can do, you can focus. For example, on one type of interaction. For example, you can just look at protein kinase-substrate networks, and those are directed by bipartite graphs that connect kinases to their substrates through protein phosphorylation. These networks are useful when analyzing data from [INAUDIBLE] will need to be detect the changes in phosphopeptide levels with the kinase cascaded are responsible for those observed changes. And we will discuss this further along the course. What I mentioned before, it's very important to understand that we have incomplete knowledge about the connectivity of cell signaling networks. And this makes it difficult to build dynamic models of those systems. In this particular example, Rica Albert, who was one of those authors of those first seminal paper. Which was the [INAUDIBLE] paper, has come up with a solution. And she introduced into the network pseudo-nodes, that explain some unfilled, unknown links between known components. It is also common to represent transcriptional regulatory networks with the nodes being merged, where the gene and transcription factor are all one node. And the links represent regulatory interactions that include the effect of the transcription factor on the expression and activity of another transcription factor. Then MacArthur, who was a PLoS Tech in my lab, showed in this study which he did before. He came to my lab that both embryonic stem cells and osteoblasts are regulated by circuits made of nested feedback loops. While the components from each circuit inhibits components from the other positive feedback circuit. Using dynamical simulations, he showed that cells can grow from the undifferentiated states to differentiated states smoothly. However, it's very difficult for the cells to go back to the undifferentiated states. The differentiated cells can jump back to the embryonic state. And this is a something important for understanding the process of a IPS reprogramming. This is another example of [INAUDIBLE] group. Here what you see, she created a dynamic model that not only connects gene regulatory networks, cell signaling networks. But also interactions between cells. So all the networks that I've shown so far were created manually by experts. However there are other ways to build networks from the literature. And here I'm showing two examples. A semiautomatic extractions of paper from publications. And this way, the software identified the abstracts that potentially have interactions. And then a user validates and extract the interactions from the abstracts manually. But he's assisted by self during that highlights the names of proteins and potential types of interactions. There are also completely automated methods to extract interactions from the literature using natural language processing. So far I only talked about literature based networks, but you can also build networks experimentally. And one of the first methods that was used to build those type of networks was Y2 hybrid screens. So experimentally, protein-protein interactions can be determined in high throughput using the yeast to Hybrid screen system. In the first large scale, yeast to hybrid screens that identified protein-protein interactions in yeast. The studies showed little overlap and that raised concerns about the quality of the method. Regardless of these concerns, large scale studies for mapping protein interactions in atom model organism, quickly appeared and were published in top journals. This includes mapping of protein interactions in human cells, and comparing the identified interactions from these two hybrid methods. Who have previously identified interactions. Detected through low throughput methods and reported in many publications. And what they found is that there was still very little overlap between the yeast to hybrid method, and the interactions they had identified in the literature. We will discuss more about protein-protein interactions in later parts of the course. So far we've looked at direct physical molecular interactions, whether self signaling interactions gene regulatory interactions or protein-protein interactions. However, there are other types of molecular biological networks that can be created, that do not require direct physical interaction. And here is an example and of an epistasis network from double deletions in yeast. So if you knock down a gene in yeast, you may not get a phenotype and the yeast will grow fine. However, if you knock out two genes, those two genes may cause the yeast to stop growing, and that is an epistasis interaction. You can also infer interactions directly from expression data. This exercise is known as reverse engineering of biological pathways and networks directly from data. In this example above, which is a simple example, time series expression data is used to infer a directive and signed graph based on delayed correlation. Such signed and directed networks can also be created by large scale multiple perturbation data. And this is another example, so reverse engineering the topology of regulatory biological networks can be done through the analysis of a set of perturbations. And here, Karen Sachs, et al reverse engineer. The hierarchy of a cell signaling networks, using multiple perturbations, and a statistical method called Bayesian Networks Inference. So all the networks we've discussed so far, connect molecular components within cells. However, you can also connect the network of a cell to other things. And here is an example of a network that's connect drugs to their molecular targets. These networks are bipartite graph containing two types of nodes. For example, the FDA approved drugs and their directed target protein. Nodes in biological networks can be connected through more abstract types of interactions. For example, genes can be connected based on the disease that mutations in those genes can cause. In this particular example, each node correspond to a distinct disorder, colored based on the disorder class. The size of each node is proportional to the number of genes identified as containing mutations that can cause the corresponding disorder. And the links thickness is proportional to the number of genes shared by the two disorders. And this brings us to the next topic, which is functional association networks. Network representation that connects genes and proteins with more abstract concepts can be used for data integration. Anchoring many experiments with the genes identified by the experiments can be used to find dense regions in a bipartite graph of genes. And experiments to find defined functional models. And this was done by Tanay et al. When they analyzed the East network not only from a protein-protein interaction perspective. But based on data on knockdown as well as gene expression. We will study later more of the concept of how to construct various functional association networks based on various data types. And how those types of networks can be merged to gain knowledge about relationships between proteins and genes. [MUSIC]
