- MsMetrix
Contents
1. MsX MsCompare High Resolution Peak Matching QuickRef2013 2 MsX MsCompare High Resolution Peak Picking QuickRef 2013 3 MsX User Manual MsCompare Univariate amp Multivariate Data Analysis Quickref 20132. 19 88 73 17 06 1 193 10 78 0 0004 88 366 1 100 6 22 43 59 426 2151 4919 30 25 11 272 284 5 o 83 7 3713 1 100 164 25 44 60 1072 0481 74 91 136 1 096 4 49 0 0109 46 162 1 100 2 59 45 61 1260 6312 90 58 14 28 0 884 6 28 0 0033 6 1 241 A o 3 63 46 63 645 7847 47 79 20 39 1 048 2 89 0 0446 23 159 1 100 1 67 47 65 626 3206 56 97 12 78 1 043 5 37 0 0058 24 81 1 100 3 1 48 66 973 5335 91 47 43 74 1 273 2 31 0 082 12 1419 1 100 1 33 49 67 1498 7795 1045 97 0 873 3 88 0 0179 68 152 A 0 2 24 50 68 830 4195 97 61 17 34 0 88 6 45 0 003 6 4 340 A o 3 72 51 69 934 4651 88 73 23 57 1 122 5 88 0 0042 57 420 1 100 3 39 52 70 1423 7698 96 19 257 1151 7 38 0 0018 7 524 1 100 4 26 53 71 1296 6835 104 94 11 11 0 819 14 95 0 0001 10 346 A 0 8 63 54 72 1555 0593 1064 9 77 1 279 7 89 0 0014 12 2 318 1 100 4 55 55 73 1099 5076 857 22 95 1 204 6 6 0 0027 93 548 1 100 3 81 x lt gt Univariate Statistics More than 2 Classes MsCompare contains three types of overviews in situations that you have more than 2 classes in your project One of the tests is the PairWise Ratio Test It will calculate the ratios between all combinations of classes e g for 4 classes A B C and D it will calculate ratios between classes A B A C A D B C B D and C D MsCompare Univariate amp Multivariate Data Analysis Quickref 2013 There is no restriction to the number of classes but the output grows fast The same test is ava
3. L1 B1 and R1 B1 This plot is very useful in situations where you have multiple classes and one reference group e g a group of controls Group Ratio Plot multiple m z values Reference Group B1 E 601 iz 73 11 83 min P 607 miz 73 14 4 min E 608 mz 7315 65 min B 609 mnz 73 17 61 min F151 L11 R161 7 If the Problem is Really Multivariate and Peaks are small If the solution lies in small peak that have no correlation with larger peaks and these small peaks are not unique or up or down regulated than the multivariate techniques will probably fail but univariate methods will fail too In the example below a scatter plot of two small peaks P1 and P2 is shown for a two class situation No single peak is able to discriminate between the classes but together they are very discriminative This is the real multivariate power combining more than a single peak However the two peaks are not correlated with the majority of the large peaks in the data set so they will probably not be detected by PCA or PLS DA at least not in the main principal components MsCompare Univariate amp Multivariate Data Analysis Quickref 2013 PZ How to proceed In these cases use the new Genetic Optimization Algorithms to solve the problem It will search for combinations 2 10 of peaks able to differentiate between the classes For many peaks it will be slow but guaranteed to find the solution Document References 1
4. i indig what s important fast 4 Accelerating Data Analysis in LC MS x MsCompare Univariate and Multivariate Data Analysis Tools A Quick Starting Guide Introduction this quick starting guide teaches how to find significant and relevant peaks discriminating different groups or classes of samples in your project The tutorial will focus on results obtained after Peak Picking and or Peak Matching See the Quick reference guides on Peak Picking and Peak Matching This tutorial assumes you are familiar with the basics of MsCompare 1 Load or Create a Project start MsCompare and load or create a project containing all of your samples If not done before you can create classes or groups by selecting the samples from the Sample Listbox and entering the Class Name next to the Class Label button To sort the samples use the Edit Sample List from the Menu Open the list file bIf and reorder your samples Save and Exit You will have to reload the data to see the effects In MsCompare class colors are directly related to Trace colors and many of the methods are interactive The default coloring order of classes is Blue Red Green Black Magenta Cyan Orange Purple up to 10 classes The order is related to the alphabetical order of you class names To set classes with a specific color the easiest way is to add a number before the Class Name 2 Exploratory Data Analysis before running Peak Picking or Peak Matching y
5. ilable for Fisher Discriminant Scores the PairWise Fisher Test Again the output is a interactive table containing the test result for the different groups You can filter the original table so that only up or down regulated peaks will be left Attention Please don t use long class names to keep the output compact Below and example is given for 4 classes B1 L1 F1 and R1 The peak with number 609 is explored in more detail For Class F1 L1 this peak is up regulated and down regulated when comparing the classes B1 F1 B1 R1 and L1 R1 2 Statistics Overview MsCompare Multi Class Pairwise Ratio Results Median 55 1599 56 600 57 602 58 603 59 604 60 605 61 606 62 607 63 608 64 F09 65 614 66 516 67 619 68 620 69 533 70 638 71 650 72 653 72 RST lt The last Multi Class Overview Statistic is a so called Multi Class Ratio Plot using a Reference Class It will calculate the ratios for selected peaks for all groups against a fixed reference group In the case of 4 classes and class B as the reference class the following ratios will be calculated A B C B and D B You will be able to specify the reference group The output will be an interactive table listing the group ratios and a graph of the group ratios for the selected peaks See the example below Four peaks were selected Class B1 was the Reference Group The plot shows ratios for the selected peaks between classes F1 B1
6. ive group compared to the other group e g for 10 samples in group A a value of 80 means that for a certain peak 8 samples are larger and 2 smaller compared to the other group Fisher Discriminant Score this statistics calculates a value which expresses the difference between the group mean and at the same time takes into account the standard deviation within each group High values gt 5 have clear separation power and not much spread The plot on the right displays part of the ratio graph lower Clicking on a peak number will extract the EIC s top You can filter and sort the full table based on any of the calculated Statistics e g keep all peaks in the table having a ratio value larger than 3 0 The above procedure is applicable to multi group problems too However you then should build data sets containing only two groups from the full data set which is more work All Statistics Overview by selecting this option you can calculate all the E Set Univariate Statistics Thresholds for Markin OJEJ Statistics directly The output will be a table with the calculated statistics for Relative intensity 2 All Peaks Must be Larger empty no use P 0 1 all peaks Up and down regulated peaks will be marked in color Blue means up regulated red means down regulated You will have the option to view all peaks or only peaks that are up or down regulated Optionally i t value Up Down Te
7. l Clustering Clustering and PCA are so called unsupervised techniques they do not use class information to find the solution PLS and ECVA are supervised techniques these explicitly use the class information to find the solution regression MsCompare distinguishes two type of problems related to the setup of the study 2 Classes you can use the supervised technique PLS DA Partial Least Squares Regression or ECVA for problems consisting of two groups Multi Classes use ECVA Extended Canonical Variate Analysis a powerful new technique combining PLS and Linear Discriminant Analysis From the score plot try to find directions that separate the classes Then look at the loading plot in the same directions to find the discriminating peaks Again often the large peaks stick out in the loading plot 6 Univariate Analysis Tools for finding Discriminating Peaks Multivariate techniques in general are variance based which means that the focus is on the large peaks in your data Furthermore it is expected that peaks are highly correlated In many LC MS and GC MS studies the interesting peaks will be very small and the correlation structure with other peaks in your data is missing In these situations almost all multivariate techniques will fail or the interpretation will be very difficult We have seen in many studies that univariate techniques often outperform the multivariate techniques because of the reasons mentioned above MsCompare has powerf
8. ou should have an idea about some specific details of the data A good starting point is always to explore your samples in the MsCompare The MsCompare module has many tools to directly interact with your data Decide if certain artifacts are present check the alignment of your samples visually get a feeling of the peak widths decide at what level peaks are relevant and see if normalization of samples is important etc etc Start with PCA Principal Component Analysis on the TIC or BPC traces to detect outlying samples If you already observe nice group separation you probably have an easy problem in which some of the major peaks are responsible for the differences between the groups ee a 3 Run Peak Picking or Peak Matching see the e oo a a a a a aa tutorials on how to perform Peak Picking or Peak Matching Itis amp s sosom 8 momo moo os assumed that the results one big table containing all peaks for 35 s wo sow m mo so o all samples is present or can be loaded from disk When ee ee ee eee ee clicked in the table the EIC s of the selected peak will be i a Tanaris isa Seats as fa plotted to the lower window You can plot EIC traces or MS eom so mO mO ma we ws me wa m ws nma me as w spectra at any resolution and automatically zoom in on the ei eL e a a e a a a a a a peaks of interest MsCompare has many tools to directly explore the table in a graphical interactive way Table plots a
9. re available from the Menu Table Functions Example on the right a small part of the Table Profile Plot is shown a graphical presentation of all intensities for all samples and all peaks Unique peaks or peaks responsible for group separation can be directly observed from the color and unique behavior Clicking in the plot will extract the EIC s for the selected peak MsCompare Univariate amp Multivariate Data Analysis Quickref 2013 4 Run PCA onthe Table start exploring the table by x10 PCA Score Piat Tabie Data pe 1 93 09 5 po 2 285 t 0 09 23 min z running PCA on the peak list optionally decide on scaling normalization etc PCA is an unsupervised multivariate technique It does not specifically search for groupings The score plot on the right already shows a very nice separation but in more difficult problems this will not be the case You can check the loading plot to see which peaks are responsible but is not a very easy task Often you will only see large peaks sticking out in the loading plot and probably you will check no more than 2 i _ l l principal components Even auto scaling making all peaks z equally important is often not very easy see loading plot on the 2 right oo A pi p oo N o N w an 5 Multivariate Analysis Tools Two Class or Multi Class Problems the Multivariate Tools in MsCompare consist of PCA PLS DA ECVA and Hierarchica
10. st empty n0 use the full table can be filtered on these peaks The Overview Table is 3 interactive clicking an entry will plot the EIC s in MsCompare isin A AM Visto E ENTE Before the table is generated you will have to decide what is a relevant a threshold regarding each of the statistical tests If any of the tests is positive Ful Selectivity Up Down Test empty no use the peak will pop up in the table To use only one type of statistics clear all Ratio Value based on Median Up Down Test empty no use 2 other threshold S Up Regulated Up Test Only empty no use 75 Fisher Discrim Value Up Down Test empty no use Attention the combined results include peaks that pass the test for any of 25 i the individual tests The test color up down is based on the threshold To only view peaks that have e g a Fisher value gt 5 clear all other thresholds D Statistics Overview m Statistics Results Overview Median Class Sample Control ag Peak No miz tR min Avg Int Ratio t value p value Unique Weighted Uniq Selectivity Up Reguiated Fisher Jr JZ 0592 r2 0 2J 07T Tose J I Uso Tee mer T Tov T U aiis 38 514 1402 2258 88 73 14 24 1189 5 714 0 0047 86 296 1 100 3 29 39 52 1124 5192 99 78 23 63 0 901 3 08 0 0369 52 293 A 0 178 40 53 1423 269 96 15 29 05 147 3 69 0 0211 78 599 1 100 2 13 41 54 12741069 82 94 18 47 1 026 3 19 0 0331 13 80 1 100 1 84 42 58 1401 72
11. ul univariate statistics to find your discriminating peaks We make a distinction between 2 class projects and multi class projects Univariate Statistics 2 Classes in MsCompare select from the Menu Biomarker Stats gt Set Selectivity Rules You will have to decide which group is expected to contain the up regulated peaks some statistics use ratios Select the option according to the class setup MsCompare has 7 different statistics for finding discriminating peaks ratio t test p test uniqueness full selectivity up regulated and Fisher Discriminant Score You can create plots for any of the selected Statistics The plots are interactive click on a peak and the EIC s or the Profile plots will be generated Ratio Test will calculate the ratio s between the group means or group medians In one plot you can see both up and down regulated peaks Uniqueness Test calculates a value between 100 and 100 The value 100 means unique and up regulated 100 means unique and down regulated A value of zero means that the group means are equal Full Selectivity Test checks which peaks are larger in one group compared to the other group must be true for all samples MsCompare Univariate amp Multivariate Data Analysis Quickref 2013 PeakNue 584 Extracted on Currerts mz 71 nominal in the group The Full Selectivity has a value of 1 or O Percent Up Regulated Test counts the number of up regulated samples in your act
Download Pdf Manuals
Related Search