| S.No | QUESTION | ANSWERS |
| 1 |
I went on ahead and looked up the running time of the bipartite matching problem, and there is one fact that has to be considered. It seemed from the groups presentation that we are matching molecules and that the points being matched are already divided into two distinct sets. In this case, the problem IS solvable in polynomial time through graphs algorithms (specifically, O(n + m) where n and m are the number of vertices and edges in the graph, which translates loosely to O(n^3) because m = O(n^2)). The problem would be NP-complete if we didn't have the distinct entities to match and the points were all in one set, initially not partitioned. (Steven Cummings) |
Yeah, the bipartite searching algorithm, which is used in old DOCK, is in the field of Graph calculations; and for the algorithm of new version, DOCK4.0, it is probably "divide and conquer". thank Steve for the important info, and we are working on it and will show the details next Mon. (Ding) |
| 2 |
This question is for group 3 about the binding site. I know X-ray is the major technique to determine the sizes and shapes of the protein binding sites, however the protein must be able to be crystalized before X-ray is applied to it. The question is: if you can't crystalize a protein, are there any other methods to tell the properties of the binding sites? if so, how big a job is it? (Hongying James) |
|
| 3 |
In the paper Group 4 presented, the Rosetta program is less accurate to predict the initial and terminal coding exons because it predicts the initiation or stop codon as splice site. For me, I thought that to predict correctly the initial and stop codon in the initial and terminal coding exons should be very important. Anybody has any ideas about how to improve the accuracy? Liu Shuyu |
No reply!!! |
| 4 |
It is well known that protein domains which perform a similar function often have a similar structure.These domains are grouped together in order to allow a single protein perform a set of functions.Would it be possible to find these domains, and then design a custom protein by just grouping together all the domains which correspond to the desired functions? I'm sure people have already thought of this, but have they had any success? I think that the homologous sequence analysis stuff that we are learning about in class could be applied to the discovering of these domains. And maybe when taking these domains as single entities (that is, ignoring the intradomain interactions) it would not be a very difficult molecular dynamics problem to predict their interactions with each other.They could then be arranged in such a way as to ensure that they do not interfere with each other's functions, and maybe they could even be configured to assist each other or cooperate. What do you guys think, is this feasible? Michael Lawrence |
|
| 5 |
I have a question for group 7.Is NADH not produced in the second pathway,passing through the shunt... As I am a complete naive in the stuff you are dealing with,I would like to know what would be the next step after you resolve the RBC Network. Tulasi |
In pathway 2, denoted p2 in Table 3 in the paper, the net reaction produces NADH, but no ATP like p1.The reason it does not produce any net ATP is that the shunt in p2 skips the ATP production.However, the NADH (in both p1 and p2) is produced in the GAPDH reaction just before the shunt.The reason that p1 has no net NADH production is that the NADH is broken down to NAD+ and H+ in the conversion of pyruvate to lactate at the end of the pathway (p2 stops at pyruvate). Hope that makes sense, Michael Lawrence I read ur doubt. the explaination is as follows: In Rapoport Luebering Shunt instead of going from 1,3 DPG(Diphospho glycerate) directly to 3 PG(Phosphoglycerate)which involves production of an ATP, it passes from 1,3 DPG to 2,3 DPG then to 3 PG, thereby bypassing the normal route and hence no production of ATP at that point.This shunt usually does not occur except when one goes to high altitudes called as oxyhemoglobin modulation. In RBC's (& Skeletal muscle)-> anaerobic glycolysis occurs where in Pyruvate is further broken down to lactate and in this step NADH is converted into NAD+.So the NADH produced in the converion of GA3P(Glyceraldehyde 3 Phsosphate) to 1,3 DPG is utilised in conversion of Pyruvate to Lactate.so "NO" net production of NADH in conversion of Glucose to Lactate. A relatively simple sketch of glycolysis can be seen at: glycol.html. Now coming on to your second question: utility of resolving the RBC network: As you can see the most common pathway involves the conversion of Glucose to lactate in RBC(P1).P2 to P6 represent the extreme pathways which can occur in RBC.This way any reaction occuring in RBC can be limited to 6 pathways and the other 16 reversible reaction can just be excluded. Other examples to prove implementation of these sort of networks is as follows: H. Influenzae: a gram negative pathogen, causes otitis media, acute & chronic respiratory tract infections.It consists of 83 potential substrates,50 products and 461 reactions. Similarly H. Pylori: Commonest cause of Gastric Ulcer.It consists of 583 reactions and 381 metabolites.The number of reactions & metabolites which exist in this reaction are enormously large.We conclude: a) Algorithm cannot be parallelised easily and requires a fast processor with large memory. b)Current calculation of the full pathway structure is infeasible(time & memory requirements are too large). c)We can restrict the output of the matabolic network to smaller subsets. means the philosophy of " Divide & Conquer". Hereby studying the entire network and limiting it to the pathways of importance to us: Pharmaceutical industry to manufacture drugs, biologists to know the rate limiting steps. Note: GAPDH reaction (as mentioned by Lawrence:I have not heard of it). Seth, Raman |
| 6 |
Where does the fuzzy logic apply in the application of today's presentation? I am not
pretty clear with that..
Also I have doubt about this: ei, k = {0,1}, I thought if 0 and 1 indicated the presence
of causality, it would imply that from state i to state k there is a transition which
has the corresponding causality.
It was just the way I understood, I would like to know if what I perceived is correct..
Tulasi
|
We had a very interesting lengthy discussion over this, so the whole of the discussion is at: fuzzy.doc |
| 7 |
I have a question for group 7. You mentioned that the dehydration reaction of glycolysis is mediated by the enzyme dehydratase. How is this enzyme different from dehydrogenase? Stephanie Carns |
|
| 8 |
I remember from when I learned glycolysis and all the surrounding pathways that there were many points along the pathways at which intermidiates could enter into the pathway from other sources or pathways. It seems like that would mess up a method that used mass balance to recreate a pathway. Figure 5 shows a pathway that is much more isolated from outside influence that it would actually be in nature. Does this matter or am I missing something in the paper. Daniel |
it is why they make steady state approximation S x n = 0. Olga |
| 9 |
I just have a quick question for Group 8. In the paper "Creating Metabolic and Regulatory Network Models using Fuzzy Cognitive Maps", in Figure 3 it says that white means the node was on and black means the node was off....what are the gray sections??? Cammack, Kristi M. |
The logistic function used actually did allow for values in between. Because of the choice of c=1,000 only rarely does any input (yi) attain a value other than 0 or 1, but it does happen sometimes, hence some values were 0.5 and gray. Bill |
| 10 |
For today's class, the network structure-enumerating algorithm. Does this method guarantee to get a solution compatiable to the data? Or in other words, is it true that if the algorithm can not get a solution, it means that the data measured in the experiment is wrong? So I am still wondering if the proof of this effectiveness of the method is true. And, it does not has general constraints on how to construct the network, just shows the example of a multi-variate(level) gene in the network. hai |
|
| 11 |
Hello, In the paper "Identifying Gene Regulatory Networks from Experimental Data" in the Activation-Inhibition Scores section, Figure 3 shows an example of a "good candidate activator" and a "good candidate inhibitor". I can follow the Activators graph up until about time step 12. It seems like between time step 12 and 14, the decrease in concentration of the activator actually "causes" the concentration of the activee to increase. Two questions: 1) Is this accounted for when calculating the activation grade they mentioned (where they look at the peaks, leading edges, etc.) In other words, can we just ignore what happens between these two time steps simply because things look nicer for the rest of the graph? and 2) Why in nature would the activee concentration continue to rise even after the concentration of the activator has decreased quite a bit. Should we guess that some of the transcription factors have remained bound to the binding sites? Seth P.S. Theres a similar (but opposite) effect between time steps 13 and 14 of the Inhibitors graph. Hi, Is it that both c1, c2 are dependant on each other(in the graph)??If so in what way? Tulasi |
tulasi, 1)As it is a feedback network (which is obvious), the input for each binding site is always(in the considered network)the output from the previous substance generator.The input for the binding sites b2 & b1 is always the output generated by the substance generators r1 and r2 respectively. For this reason,I think the concentrations for e.g. c1 and c2 are dependent on each other. 2)As the final output is evaluated by considering several combinations of these networks, there might be a situation where the concentrations might be independent. I am not sure whether this a fair answer. Bhavani I think it is fair to say that concentration A depends on concentration B if some change in concentration B crosses over a threshold and triggers a substance generator which affects concentration A. So the dependence of the concentrations is explicit in the model. Does this sound reasonable? Michael |
| 12 |
I just had a quick question for group 9 about the data that was shown at the beginning of the reverse engineering section. This is the graph with the colored circles representing different concentrations at different timepoints. Was that actual microarray data or an example of microarray data? I am kind of confused about where you start from in reverse engineering. The data shown seemed too orderly to come from an actual experiment. Thanks. Daniel |
I am not the presenter or the 1st paper, but the data shown in paper 1 does look too orderly. In paper 2 (that we missed successfuly), they used expression level of gene ORF of Saccharomyces cerevisae (6601 genes total) with 17 observation/measurements taken at 10 minutes interval. The authors in paper 2 did not produce the data, the data was first presented by R. Cho et. al., "A genome-wide transcriptional analysis of the mitotic cell cycle", Molecular Cell, 2:65-73, July 1998. Paper 2 can be viewed as reversed engineering approached from different angle. They first pruned the genes by keeping only those that are active and have discernable peaks in their ORF profiles. Rather than model the network mathematically, paper 2 clusters genes that behave similarly using average linkage (hierarchical & agglomerative) clustering algorithm. Next they use a continuous weight function (I am tempted to call it a fuzzy membership function, because it looks like one) to determine the type of edge that connects two clusters, Cr and Cs. Cr is said to be the activator of Cs, if the strength of "activator" is higher than that for "inhibitor" and vice versa. They determine this for all possible pair of clusters. Essentially creating a directed graph whose edges are labeled A or I and assigned appropriate weight to show how high the influence of A or I given by the regulatory cluster to the regulated cluster. Next, they apply a simulated annealing based optimization algorithm (known to be good at finding "a solution" out of high combinatorial problem) to find the most optimum subgraph based on some constraints. The constraint they first use is quite restrictive, like they only allow each vertex to have two in-edges (one A and one I, which are the ones having the highest edge weight from each type). They also want to determine which vertices that act as regulatory elements. Hence, the vertex is slowly labeled as A, I or N (I guess it means they are the regulated ones). Again vertex labeled as A or I does not receive input edges from other vertices, and vertices labeled as N cannot regulate other vertices (it receives but does not send). The constraints are quite restrictive, but I think they just try to limit their test case to show that the optimization can work for a simplified form of regulatory networks. They show some results for this. Later, (Richa Puri was supposed to present this) they also proposed some theorems that explore the complexity and possible solution search space for maximum gene regulation problem (where the constraints they use previously were relaxed). So, if we had been able to present the second paper, I think we would have a much better view of what could be done with gene regulatory network. Ozy |
| 13 |
Hi, all: We discussed in the class what data the biologist should give to the computer engineer. It seems that the biologist should try each combination of the binding sites and it becomes a NP-hard problem. So my question here is how much experimental data is enough? Since team 9 will talk about how to get the data, could you explain this in your next presentation? Liwen Tu |
We are not going to talk about "how" to get the data. I think there are Biologists in our class who are more qualified to explain this. The second paper is a different approach of "reverse engineering". As a computer engineer, I just look at it as a clustering and graph optimization problem. The use of optimization algorithm with a specific criterion function (hence defining our constraint on the networks produced) allow us to come up with a solution which optimized the given criterion function. Is this the best solution ? No, but this is one possible solution for which the constraints are satisfied to a certain degree. I do not think there are any rules on how many data points needed in order to get a good results. The rules on how many data points needed in order to get a good results. The more the better, and it is always true in both supervised and unsupervised learning problem, perhaps more true in unsupervised cases like this. Ozy |
| 14 |
Hello: Team 9 presented that gene profiling could also be very useful to identify a regulation network.However, the regulatory relations btw genes could be ambiguous in this method. For example, one of questions raised in today's class, as for the peak of gene B follows gene A, there is several possibilites. 1.gene A acivate B 2.gene B inhibit A How could we distinguish this? I know that "genome wide gene profiling" is successfully used to identify genes that are regulated by a certain signal. So we could knockout gene A and do the profiling again. if the expression of B doesn't change. We could say "gene B inhibit A in a negative feedback loop". if the B expression increases, we could say "gene A acivate B". Ding Team #3 |
|
| 15 |
This is a question for all the "network" group: I have noticed that in "network" presentations, concentrations and other parameters (binding constants , disassociation constant, etc) are the factors in creating or identifying a network model, however temperatures are not considered, are we assuming the temperature is alway 37C or room temperature? I think binding constant and disassociation constant are temperature dependant? also would the solvents affect the network? Hongying James |
It's probably safe to assume 37 C unless you're dealing with proteins designed to operate at much higher or lower temperatures. The activities of most proteins occur over a narrow range of conditions, so changing the temperature too much will inhibit protein activity. As far as solvents: since we are talking about proteins, water is really the only solvent you need to consider. Correct protein folding requires a polar solvent and many (if not most) proteins require specific physical interactions with water molecules. Chris Hemme |
| 16 |
I am a little confused with the optimization function: f(g) = -C1(count(A) + count(I)) ... Is this C1 a function or a constant? All the other C's in the paper appear to be constants, but the description of the function says that this term is a penalty for unlabeled vertices. So is C1 a function that subtracts the number of activating and inhibiting vertices from the total number of vertices? Also, they are aiming to maximize this function, but the description of simulated annealing given in class was to minimize an energy function. So in this case would they use gradient ascent instead of gradient descent? Michael Lawrence |
|
| 17 |
The purpose to smooth a function is to make it differentiable, is this correct? How do you smooth a function? Let's say a function: y=x, where x belongs to [0,1] y=-x+2, where x belongs to (1,2] How to do smoothing for this function? This is my research-related question. Would you give me a pointer to the relevant subjects if it takes time to explain. Your help will be greatly appreciated! Mingshu |
One purpose of smoothing is to make it meet certain "regularity" conditions, which relate to conditions on its derivative. In practice, smoothing is mostly done on raw data, which is thought to approximate and underlying function. Since the smoothing is done on the data, rather than the analytical functions (which rarely exist in practice), there is no closed form for the resulting smoothed function, just the new data points from the smoothed function. Smoothing is typically done to de-noise a set of data (density estimation, nonparametric regression) in order to capture the underlying function. There are dozens of smoothing algorithms, but the basis for many is the kernel smother. Basically, a kernel function (typically, a density function, since a kernel K(x) should integrate to 1 ) is used to reweigh the points.The weights are assigned a value from the kernel, and the kernel is passed along the function using a "smoothing window". A larger window results in a smoother function and is less representative of the local behavior than the raw data. A shorter window is less smooth, but more representative. The tradeoff is more bias for less variance, and vice-versa. A simple example of another type of smoother is the moving-average smoother.To use on your example below, sample your function below n times (you pick n) over the range [0,2]. So you have a collection of n points (xn,yn). The new smoothed points will be the average of the nearest k points centered at (x,y).k here would be your window. A kernel smoother would just reassign different weights to the average, so that it is a weighted average. See Silverman (1986) "Density estimation for statistics and data analysis",chp 3, for more details. Wade |
| 18 |
|
|
| 19 |
Hello All, Any pointers to Microarray Roberts?? Are they just capable of Probing and Scanning data or they can also do Clustering of Data? Sarada. |
|
| 20 |
In the rare event that 2 data points can be assigned to each of 2 clusters (in nearest neighbor clustering), how is it determined to which cluster the point should be assigned? Can this effect the results/interpretation of the results of the experiment? Stephanie |
Stephanie, Just to clarify, it cannot happen in nearest neighbor algorithm because it proceeds sequentially. Such a point would be assigned to the first cluster that was within the threshold t. You are probably thinking of k-means. It is conceivable (in k-means) that there could be a situation where a point in a cluster is just as close to the mean of its own cluster as it is to the mean of another cluster. I don't know what would happen in a situation like that, but I think it would be a fairly rare event in a high-dimension data set (d>15) of a decent size (n>100). Even for noncontinuous data, the multivariate mean vector would be "nearly" continuous over a certain hyperspace. Assume a worst case scenario, that this event did happen and assuming the point is always assigned to the opposite cluster, then the cluster assignment would just flip-flop as the algorithm was close to converging (assuming everything held constant). From a statistics point of view, it wouldn't really matter what group this point was assigned to because it is "equally" similar to each group. Clustering is an exploratory tool, so such situations shouldn't be of serious concern. We are not truly testing a hypothesis with clustering in the strict sense of the term "testing". Wade |
| 21 |
How is the intensity of the color measured when microarrays are used? If it is a method similar to spectrophotometry, how is the discrepency between the different wavelengths of red and green light settled or is this not a concern? Stephanie |
|
| 22 |
Hello, I got 2 questions and 1 comment: Question 1: In Eisen paper, last few lines of the first paragraph on page 14865 says "When designing experiments, it may be more valuable to sample a wide variety of conditions than to make repeat observations on identical conditions" What can be conditions? Are those factors able to regulate the expressions of different genes?In figure 2, it provides a an example about some processes, such as high temperature or lower temperature, are they the conditions? Question 2: The conclusion of the paper is something like: based on the similarity in pattern of gene expression, some yeast genes can be clustered in a group which they have similar function (function is already known). Coexpression of genes of known functions with poorly characterized or novel genes may leads to the functions of many genes. In figure 1, at the end of the desdription of figure 1, it says "These clusters (A, B, C, D, E) also contian named genes not involved in these processes and numerous uncharacterized genes. Based on these facts and conclusion, if we analyze some microarray data by clustering and we already know the fuctions of most gene in a cluster, can we infer other new genes with unknown functions have similar functions with those known genes? Comment: In hierarchical clustering, regarding those 3 algorithms, single-link, complete-link, and group average, even the later K-mean methods using single-link, I think selection of algorithms should be based on what knid of data you have. If the data is more evenly scattered within clusters, the average may be better that single-link. Any explanations and comments are welcome! Shuyu |
|
| 23 |
The features of the data are expected to be independent as much as possible. possible. Replicated(or highly correlated) features cause the problem of "curse of dimensionality". For an extreme example, originally we have only two independent feature, then we replicate the second feature by 100 times, we get a data set with 101 dimensions. Now if we treat the dimensions orthorgonally by choosing Euclidean distance metrics, the differentiating ability of the second dimension has been amplified a lot, then the effect of the first dimension disappears. So how to reduce the features? One way is to use covariance matrix of the data to show the shape of the data set in the high-dimensional space(we just compute the eigen values and discard those directions(in the directions of the eigen vectors) associated with very small eigen values). One problem of this approach is that it may not be correct when dealing with the case that the data set is like two parallel line segments and close to each other and each line is a cluster(this approach just save the direction of the line, and discard the direction normal to the line, which is really good to differentiate the two clusters). Another way is to compute the correlations of each pair of features, if the result is too high(but need to be normalized), the two features are probably highly correlated. However, sometimes one feature is the non-linear function of another feature, using this approach may not reflect the relationship of the two. I guess replication of the features is often in microarray data. So this issue is really important. Any other ideas? Hai |
|
| 24 |
DNA chip vs Microarray |
|
| 25 |
|
|
| 26 |
How accurate is the clustering in predicting gene function since it doesn't take into effect post-transcriptional gene regulation? Meyer, Louis John |
Statistically speaking, clustering is not designed for prediction (although you could use it for that) but rather for exploration. So there is no confidence probability you associate with your results. Through cross-validation, (provided you have enough data) you could heuristically determine the error rate and thus get an empirical estimate of how accurate the clustering is for a particular set of data, but these results would not generalize to other data sets. Wade |
| 27 |
To the class, The paper uses term parametric ordering of genes, but hasn't mentioned what are the parameters they are using(though they say it is their alternative approach).Is clustering analysis a non-parametric approach?? Any ideas about these Statements!!! Bhavani |
|
| 28 |
Hi All, In the first reference paper it is said that the microarrays and reverse dot-blot analysis are similar in concept. Then what makes microarrays distinct from the other? Tulasi |
Tulasi, I think it is the number of genes you are working with. In my lab when we are doing dot-blot analysis we are interested in a few particular genes. We make DNA samples of these genes by PCR ( a molecular biology technique that amplifies a specific DNA) and attach it for a membrane. Then, we want to know if these genes are expressed in certain cells upon certain stimulation (UV, growth factors, cytokines). We stimulate cells and isolate total mRNA from these cells and control, unstimulated cells, label it. It can be a radioactive label, or a fluorescent label, or biotin, or an enzyme. If total pool of mRNA contains mRNA of interest, it will hybridize with DNA on the membrane and we will detect it. So, the principal is the same as in microarray analysis. The difference is the number of genes analyzed. Olga |
| 29 |
How can I tell which cluster represent which gene from the microarry figure? Thanks. Liwen Tu |
Notes on clustering by Hai clustering. Just remember clustering isn't magic. garbage in=garbage out. If that data clusters doesn't have a meaningful interpretation , then it really doesn't matter. Clustering is just an intermediate tool. The algorithm and the convergence might work fine, but if the end result doesn't have a useful interpretation , than I would be wary. If you were only concerned only with dimension reduction, then that might be OK. In the case of the uniformly distributed points, clustering would not make much sense. It would be like reading tea leaves! A lot of data CAN be run through a clustering algorithm, but that doesn't mean that it SHOULD be. There are good counterexamples for many algorithms, but most of the time these counterexamples are situations were the algorithm isn't appropriate. That is why they are counterexamples! Wade Davis |
| 30 |
How often does this happen: you assumed that there were k clusters, after you finished classification, you got your k clusters. However, in one of your clusters, there are very distinct two subclusters, and it may be better to classify the two subclusters into two separate clusters so that there are total k+1 clusters. I guess my question is: how do you make assumption on number of clusters and how do you correct yourself? Hongying James |
It depends on what clustering algorithm you use. But my first choice would be the Possibilistic C-means algorithm due to its "mode seeking" characteristics. If you try to find m clusters and there are k naturally existing clusters in your data set: if m>k, you will see that some of clusters in m will be very close to each other if not identical in terms of membership value of their members and their cluster centers. This is an indication that you have over-specify the number of cluster. If m is less than k, you will get m clusters and these are likely to be m highest density cluster out of k cluster in the set, what happen to the other clusters not represented in the m clusters you find ? Most likely, the members of these k-m clusters will receive low membership value. Hence, you can filter them out by applying some thresholding on the membership value to detect data points that should belong to other clusters outside the m you already found. A fuzzy C-means will simply split the data points belonging to the k-m clusters among the m clusters, because the condition that the sum of all membership value should add to 1. Hence, it will be hard to differentiate the data points that should belong to the k-m clusters. Ozy |
| 31 | What is the terminating condition for k-means algorithm if we don't get good clusters from iterations ? Raman Seth |
Here is the discussion on k-means algorithm, the plot.jpg |
| 32 | I've written up a word document in which I try to explain how the decision function works (in dual space, that is, with the kernel function) and how it learns this function by maximizing the distance to the closest examples. The second part I explain with regard to the direct space representation of the decision function, since it is easier to understand. This is an excellent paper to read if you are interested in all the details: http://portal.acm.org/citation.cfm?doid=130385.130401 I've attached the word document which tries to give the explanation in terms that someone without a strong mathematical background can understand. svm.doc Michael |
|
| 33 | To the class, In the microarray analysis we are subjected to get duplicated genes. How are we supposed to deal with these genes? I mean, If we have to calculate the fold changes do we have to average the signal intensities of all the duplicated genes or will that be ok to consider the duplicated gene as an individual gene. In what manner the results vary? bhavani |
With respect to using a microarray for expression analysis, it should not matter if there are multiple copies of a gene. The goal of an expression array is to identify the quantity of mRNA in a sample. The source of that mRNA (i.e. one or multiple copies of a gene) will not be detected by, or will not affect the results of, analysis by a microarray. Alternatively, microarrays have been used to identify multiple copies of genes present in genomic DNA. Aaron |
| 34 | I'm just wondering, do we know how many forms of proteins can a gene express? and which one is the dominant form that have the tendancy to cause cancer? I remembered sometiem ago, it's reported that breast cancer is rare among Asian people because of their large amount of Tofu (a soybean product) consumption, does this indicate that soybean may have the ability to inhibite the production of the cancer_prone protein? Hongying James |
|
| 35 | can the cancer stages such as primary and secondary stage be explained using the dominant proteins/gene theroy? or else is it just based on the extent to which the disease has spread in the body and the damage caused? Venky It's a common knowledge at this point in time that family history is probably considered the most important factor when assessing breast-cancer risk. How far is it possible to use the proteomic approach described in the R. A. Harris et al. paper to establish this theory. Basu |
This is something I have found sometime trying to figure out the reasons causing cancer: The majority of breast cancer is sporadic, occurring in women without a family history of breast cancer. Approximately 15-20% of breast cancer is associated with some family history. In general, a twofold to threefold increase in the risk of breast cancer development has been associated with breast cancer in a mother or sister.(2) Importantly, a woman's risk of breast cancer is strongly related to the number and type of relatives affected as well as the age at which these relatives were diagnosed.(2) Presumably, this familial clustering is a result of multiple, relatively weaker genetic influences, single cancer susceptibility genes with low penetrance and shared environmental risk factors. Only 5-10% of breast cancer is thought to be due to the inheritance of a single, highly penetrant autosomal dominant mutation in a single cancer susceptibility gene such as BRCA1 or BRCA2. You could get some more info at http://www.health.state.ri.us/disease/cancer/canbrca.htm. Tulasi |
| 36> | My idea is there are different proteins in different tissues and the abnormal behaviour of these proteins result in dreadful diseases. So how do we identify which part of the human body is affected by cancer? Do we have to consider the tissues from different parts to identify the cancer effected in one part. Bhavani |
|
| 37 | I had a question for the biologists or anyone who has an answer. It is known that a tumor needs microvasculature (network of blood vessels) to survive. So how can you differentiate between tumor microvasculature and normal microvasculature using the new data coming out of human genome and come up with antigenic differences that can be used to target therapy? Nandini |
I work with vascularization in the placenta. From what I've read, the vascularization process in cancer, the placenta, the ovary and other tissues is very similar at the molecular level. In my opinion, you can't differentiate between tumor microvasculature and normal microvasculature. But that doesn't mean that you can't use drugs of other techniques that inhibit vascularization in an attempt to control cancer growth. Mesa |
| 38 | So if you can't differentiate between tumor and normal microvasculature, wouldn't the drugs meant to control cancer growth, inhibit normal vascularization too? Nandini |
|
| 39 | This question : what's the difference between Northern point and microarray? I didn't get what exactly Northern point is, can any biologist explain this? by Hongying James, went unanswered!! Any ideas!! Tulasi |
|
