Fruitfly Tumors A range of sizes and morphologies observed: Microtumors Ubc9 -  dif -   dl - Ubc9 - - Microtumor Microtumo...
(Chiu et al 2005) :  dUbc9 negatively regulates the Toll-NF-nB pathways in larval hematopoiesis and drosomycin activation ...
Bc allele background FlyBase   GBrowse  modENCODE  GBrowse   Gene DmelBc FB2009_07, released August 10, 2009  General Info...
Comparative Analysis of Area limits 25K to 300K and 300K to 600K in both Genotypes :  Higher Maximum Likelihood mean, vari...
25K-300K Area Size Tumor Log-Normal Distribution in BC-All and Recessive Genotypes (number of micro tumor found or frequen...
PROBLEM STATEMENT Tumor size data from non-random and correlated data. Samples were prepared for 8 days and scored on 9 th...
ASA 10/23/2009 Minneapolis Presentation  Predictive Modeling, Mathematical Simulations and Data Mining: Making Sense Out o...
Analysis of Raw data showing V-shape residual and compensatory response by 25K area limit (R-square =  0.36 VS 0.76 VS 0.8...
LITERATURE REVIEW & METHODS Dynamical Simulation by Taylor’s Power Series like Math equation:  A . Y= x 1  + x 2 +x 3  + x...
B.  Reverse Engineering Prediction by  the equation of  y = 4.251a2 + ln(a2) + 7.243ea-  CF . ( Candida Ferreira. 2003.  w...
Modified Candida Ferreira Method (Equation 3.2): Correction Factor (CF)- Genetic Fitness not as Underestimated: Consistenc...
Reverse Engineering of Polynomial Models of Gene Regulatory Networks (Visual Analytics = Meta Modeling =  what are the ran...
Reverse Engineering of Systems Systems identification  in Engineering: goal is to construct a system with prescribed dynam...
Mathematical Genetics Concepts <ul><li>Average Effects of a Gene: Mean deviation from population mean  of Individuals whic...
Average Effects of Gene Substitution:   І 7.333 І ; very close to equation 3.2 of Candida Ferreira  (frequency of 0= 7.243...
A . “Operon or Tumor Gene Expression occurs in a deterministic way from 25K to 300K area limits, and hence would have high...
Leo Breiman:  Statist. Sci.  Volume 16, Issue 3 (2001), 199-231.    Statistical Modeling: The Two Cultures Abstract There ...
A.  Analysis of size distribution of lwr (-) microtumors from 58 animals Projection >10,000   m 2 ; Estimated volume: 0.5...
<ul><li>RESULTS : Specter and Luke INPUT/OUTPUT Method (Genomics by Stanford University): The frequency of 300K was taken ...
Ekaterina  Vladislavleva- PhD: JUNE 2008  Both measured and simulated data are very often corrupted by noise, and in case ...
VISUAL ANALYTICS:  Meta Modeling : No Plateau Observed! Genetic Fitness keeps increasing-DNA structural similarity is NOT ...
B . COMPENSATORY RESPONSE HYPOTHESES: BRODY et. al.  “Even for Lorentizian like distributions, median of ratios provide di...
B . COMPENSATORY RESPONSE HYPOTHESES… <ul><li>Bc-all/11  /  lwr/11  Ratio   </li></ul><ul><li>Bc-all/8  /  lwr/8  Ratio   ...
COMPENSATORY RESPONSE HYPOTHESES… <ul><li>Bc-all FREQ+11 / lwr FREQ  ratio   </li></ul><ul><li>Simulation: Summary Statist...
Upcoming SlideShare
Loading in …5
×

American Statistical Association October 23 2009 Presentation Part 1

432 views

Published on

large data set is not available for some disease such as Brain Tumor. This and part2 presentation shows how to find &quot;Actionable solution from a difficult cancer dataset

0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total views
432
On SlideShare
0
From Embeds
0
Number of Embeds
25
Actions
Shares
0
Downloads
1
Comments
0
Likes
0
Embeds 0
No embeds

No notes for slide

American Statistical Association October 23 2009 Presentation Part 1

  1. 1. Fruitfly Tumors A range of sizes and morphologies observed: Microtumors Ubc9 - dif - dl - Ubc9 - - Microtumor Microtumor Microtumor Aggregate Cluster Aggregate Small Microtumor Fat Body 419 Projection >10,000  m 2 Estimated volume: 0.5 mm 3 -1 mm 3 932 513
  2. 2. (Chiu et al 2005) : dUbc9 negatively regulates the Toll-NF-nB pathways in larval hematopoiesis and drosomycin activation in Drosophila. Developmental Biology. Genotype  Number of Larvae Ubc9-(transheterozygote) 58 Bc + Ubc9- 55 95% CI Odds Ratio: NS>5% 0.85- 1.25 Ubc9-   Aggregates + Tumors Aggr Tumors Totals 932 513 419 %   55.04% 44.96% Bc Ubc9/+ Ubc9-   Aggregates + Tumors Aggr Tumors Totals 874 262 612 % 29.98% 70.02%
  3. 3. Bc allele background FlyBase GBrowse modENCODE GBrowse Gene DmelBc FB2009_07, released August 10, 2009 General Information Symbol DmelBc Species D. melanogaster Name Black cells Annotation symbol CG5779 Feature type protein_coding_gene FlyBase ID FBgn0000165 Gene Model Status Current Stock availability 68 publicly available Genomic Location Chromosome (arm) 2R Recombination map 2-80.6 Cytogenetic map 54F6-54F6 Sequence location 2R:13,774,718..13,777,477 [-] Genomic Maps The gene Black cells is referred to in FlyBase by the symbol DmelBc (CG5779, FBgn0000165). It is a protein_coding_gene from Drosophila melanogaster. Its sequence location is 2R:13774718..13777477 . It has the cytological map location 54F6 . Its molecular function is described as: monophenol monooxygenase activity; oxygen transporter activity; oxidoreductase activity. It is involved in the biological processes: defense response; melanization defense response; scab formation; response to symbiont; response to wounding; transport. 10 alleles are reported . The phenotypes of these alleles are annotated with: crystal cell; hemocyte; hemolymph; lymph gland; adult; procrystal cell; lamellocyte; posterior lymph gland pair. It has one annotated transcript and one annotated polypeptide . Takehana, A., Katsuyama, T., Yano, T., Oshima, Y., Takada, H., Aigaki, T., Kurata, S. (2002). Overexpression of a pattern-recognition receptor, peptidoglycan-recognition protein-LE, activates imd/relish-mediated antibacterial defense and the prophenoloxidase cascade in Drosophila larvae.  Proc. Natl. Acad. Sci. U.S.A. 99(21): 13705--13710. Ye, Y.H., Chenoweth, S.F., McGraw, E.A. (2009). Effective but costly, evolved mechanisms of defense against a virulent opportunistic pathogen in Drosophila melanogaster.  PLoS Pathog. 5(4): e1000385.
  4. 4. Comparative Analysis of Area limits 25K to 300K and 300K to 600K in both Genotypes : Higher Maximum Likelihood mean, variances and wider confidence interval of 25K-300K shows faster mitosis and cell death than that of 300K-600K. Maximum Likelihood (ML) Estimates of BC-All (BC-lwr) and lwr43-5 All BC-All Mean Tumors Variance Tumors 95% Confidence Interval 25K-300K 4.86 0.85 1.22 to 1.84 300K-600K 1.67 0.02 1.11 to 1.20 lwr43-5 All Mean Tumors Variance Tumors 95% Confidence Interval 25K-300K 4.5 0.97 1.10 to 1.88 300K-600K 1.27 0.02 1.05 to 1.12
  5. 5. 25K-300K Area Size Tumor Log-Normal Distribution in BC-All and Recessive Genotypes (number of micro tumor found or frequency on Y-axis; every 25K scale)
  6. 6. PROBLEM STATEMENT Tumor size data from non-random and correlated data. Samples were prepared for 8 days and scored on 9 th day- cumulative effects on frequencies of BC-All and recessive (lwr-) Area size Units between 25k to 600k size distributions? Effects of new VS experienced PhD student on data collection? 612 VS 419. This difference is not statistically significant (P> 5%). EXPECTED frequency higher at all area size for Semidominant gene in the hypothetical Y-axis. Does not have a pattern to quantify by a Dynamical simulation equations- tried 100’s of published math methods…. Sample size is ONLY 48 rows of Tumor Frequency data!
  7. 7. ASA 10/23/2009 Minneapolis Presentation Predictive Modeling, Mathematical Simulations and Data Mining: Making Sense Out of Really Difficult Cancer Data. Navin K. Sinha, MS (Statistical Genetics), MS (Biometrics) and MBA (Decision Sciences) <ul><li>Bc mutation alters aggregate proportions? </li></ul><ul><li>Bc = Black cells </li></ul><ul><li>Semidominant </li></ul><ul><li>Dead crystal cells </li></ul><ul><li>Visible easily </li></ul>
  8. 8. Analysis of Raw data showing V-shape residual and compensatory response by 25K area limit (R-square = 0.36 VS 0.76 VS 0.86 ). Data Analysis needs Dynamical Simulations, Reverse Engineering Algorithms and Simulated OLS Regression.
  9. 9. LITERATURE REVIEW & METHODS Dynamical Simulation by Taylor’s Power Series like Math equation: A . Y= x 1 + x 2 +x 3 + x 4 . Reference : “Lee Specter and Shawn Luke- Culture Enhances the Evolvability of Cognition. 1996. In Proceedings of the Eighteenth Annual Conference of the Cognitive Science Society. “ According to Specter and Luke, special type of Dynamical Simulation is Symbolic Regression- “ to produce a function, in symbolic form, that fits a provided set of data points. For each element of a set of (x,y) points, the function should map the x value to an appropriate y value. This sort of problem faced by a scientist who has obtained a set of experimental data points and suspects that a simple formula will suffice to explain the data ” . This method is a standard example from Dynamical simulation and used in many different types of biological systems (Koza, J.R. 1992. Genetic Programming: on the programming of computers by means of natural selection. Cambridge, MA, MIT Press).
  10. 10. B. Reverse Engineering Prediction by the equation of y = 4.251a2 + ln(a2) + 7.243ea- CF . ( Candida Ferreira. 2003. www.gene-expression-programming.com/author.asp- equation 3.2 ) Ekaterina Vladislavleva- June 2008- PhD Theses Models to exhibit not only required properties, but also additional convenient properties like compactness, small number of constants, etc. It is important, that generated models are interpretable and transparent, in order to provide additional understanding of the underlying system or process.
  11. 11. Modified Candida Ferreira Method (Equation 3.2): Correction Factor (CF)- Genetic Fitness not as Underestimated: Consistency in Results. <ul><li>Original Frequency Data (Y-axis) Residual Plot </li></ul><ul><li>Residual Plot of Graph of a Function after Matrix Algebra Treatment. </li></ul>
  12. 12. Reverse Engineering of Polynomial Models of Gene Regulatory Networks (Visual Analytics = Meta Modeling = what are the ranges of input variables that cause the response to take certain values, not necessarily optimal? ) Dr. Eduardo Mendoza Mathematics Department Center for NanoScience Ludwig-Maximilians-University Munich, Germany [email_address] eduardom@math.upd.edu.ph Brody et al . October 1, 2002: PNAS : Significance and Statistical Errors in the analysis of DNA microarray data. 99 (20): 12975-12978 ( Even for Lorentizian like distributions, median of ratios provide distributions more Gaussian like ).
  13. 13. Reverse Engineering of Systems Systems identification in Engineering: goal is to construct a system with prescribed dynamical properties In Systems Biology, one is interested in identifying as closely as possible a unique biological system that has been observed experimentally In both cases: sparsity of available measurements will leave the system underdetermined (GIGO- Uninterpretable)
  14. 14. Mathematical Genetics Concepts <ul><li>Average Effects of a Gene: Mean deviation from population mean of Individuals which received that gene from one parent, the gene received from other parent having come at random from the population. </li></ul><ul><li>Average Effects of Gene Substitution: Change one allele (i.e. A2 allele) into another allele (i.e. A1 alleles) at random in the population and observe resulting change in genotypic value. </li></ul><ul><li>Breeding Value: Twice the Average Value of an individual’s offspring, expressed as deviation from population mean. Also known as sum of the average effects of genes. </li></ul>
  15. 15. Average Effects of Gene Substitution: І 7.333 І ; very close to equation 3.2 of Candida Ferreira (frequency of 0= 7.243 x12= 86.916 VS 7.333x12=88.0). <ul><li>Comparison: Lowest to Highest R-sq. is represented by linear, Quadratic and Cubic model Respectively. Very comparable to Original frequencies. </li></ul>
  16. 16. A . “Operon or Tumor Gene Expression occurs in a deterministic way from 25K to 300K area limits, and hence would have high survival probability”. This hypothesis indicates that there are conserved Protein motifs which generates various Brain Tumor sizes in Fruit fly in predetermined frequencies. Thus, micro-tumors counted (frequency) for lower size limits can be predicted by least non-linear mathematical and statistical equations . B . “Log-Normal distribution arose due to compensatory response by lowest size distribution over the next few micro-tumor classes”. If the number of micro-tumors counted for 25K area size is at the expense of next few, then a Log-Normal Distribution can be assured . Log-Normal Distribution explanation
  17. 17. Leo Breiman: Statist. Sci. Volume 16, Issue 3 (2001), 199-231. Statistical Modeling: The Two Cultures Abstract There are two cultures in the use of statistical modeling to reach conclusions from data. One assumes that the data are generated by a given stochastic data model. The other uses algorithmic models and treats the data mechanism as unknown. The statistical community has been committed to the almost exclusive use of data models. This commitment has led to irrelevant theory, questionable conclusions, and has kept statisticians from working on a large range of interesting current problems. Algorithmic modeling, both in theory and practice, has developed rapidly in fields outside statistics. It can be used both on large complex data sets and as a more accurate and informative alternative to data modeling on smaller data sets. If our goal as a field is to use data to solve problems, then we need to move away from exclusive dependence on data models and adopt a more diverse set of tools .
  18. 18. A. Analysis of size distribution of lwr (-) microtumors from 58 animals Projection >10,000  m 2 ; Estimated volume: 0.5 mm 3 -1 mm 3 Taylor series: y = x 1 + x 2 +x 3 + x 4 Area Limit Simulated Frequency 100,000 -01 (1) 200,000 +01 (2) 275,000 -02 (3) MLE:25k-300k Mean=4.5 Tumors Variance=0.97 Tumors CI= 1.10-1.88 Tumors MLE: 300k-600k Mean= 1.27 Tumors Variance= 0.02 Tumors CI= 1.05-1.12 Tumors
  19. 19. <ul><li>RESULTS : Specter and Luke INPUT/OUTPUT Method (Genomics by Stanford University): The frequency of 300K was taken as x 1 value and plugged into the equation. First the whole formula was used (1), then x 4 was dropped (2), 3 was x 1 + x 2 . </li></ul><ul><li>A . Bc-ALL B . Bc-All (corrected) </li></ul><ul><li>Area limit Simul. Freq. Area Size Simul. Freq. </li></ul><ul><li>25K - 97 (1) 25K -18 </li></ul><ul><li>75K + 13 (2) 50K -04 150K + 01 (3) 75K -04 </li></ul><ul><li>175K +01 (3) </li></ul>(1 ) THE PATTERN OF SIZE DISTRIBUTION OF SMALL TUMORS IN BOTH GENOTYPES SUGGESTS THAT MITOSIS IS DRIVING TUMORGENESIS. (2) CELL DEATH CONTRIBUTES TO SHIFTING TUMOR SIZE DISTRIBUTION-AS MORE CELLS DIE FROM COMPETITION, MORE SMALL TUMOR CELLS WERE CREATED TO FILL VACANT SPACE.
  20. 20. Ekaterina Vladislavleva- PhD: JUNE 2008 Both measured and simulated data are very often corrupted by noise, and in case of real measurements can be driven by a combination of both measured and unmeasured input variables, empirical models should not only accurately predict the observed response, but also have some extra generalization capabilities. The same requirement holds for models developed on simulated data. Models to exhibit not only required properties, but also additional convenient properties like compactness, small number of constants, etc. It is important, that generated models are interpretable and transparent, in order to provide additional understanding of the underlying system or process.
  21. 21. VISUAL ANALYTICS: Meta Modeling : No Plateau Observed! Genetic Fitness keeps increasing-DNA structural similarity is NOT Functional Similarity. <ul><li>Original Data </li></ul><ul><li>Reverse Engineering Algorithm </li></ul>
  22. 22. B . COMPENSATORY RESPONSE HYPOTHESES: BRODY et. al. “Even for Lorentizian like distributions, median of ratios provide distributions more Gaussian like” <ul><li>Bc-all Tumor size FREQ/lwr tumor size FREQ Summary Statistics </li></ul><ul><li>Obtain Ratio from all cell sizes and then summary statistics on it. </li></ul><ul><li>Mean = 1.509206 (ratio by lwr freq. of 8 was very similar to it) </li></ul><ul><li>Standard Error = 0.201937 </li></ul><ul><li>Median = 1.513738 Tumors </li></ul><ul><li>Mode #N/A </li></ul><ul><li>Standard Deviation = 0.699531 </li></ul><ul><li>Sample Variance = 0.489343 </li></ul><ul><li>Kurtosis = 0.430923 (ratio by lwr freq. of 11 was very similar to it) </li></ul><ul><li>Skewness = 0.566484 </li></ul><ul><li>Minimum = 0.545455, Maximum = 3.0 </li></ul><ul><li>Count = 12 = Number of Tumor Cell Sizes. </li></ul><ul><li>Confidence Level(95.0%) = 0.444461 </li></ul>
  23. 23. B . COMPENSATORY RESPONSE HYPOTHESES… <ul><li>Bc-all/11 / lwr/11 Ratio </li></ul><ul><li>Bc-all/8 / lwr/8 Ratio </li></ul><ul><li>Mean = 1.509206 </li></ul><ul><li>Standard Error = 0.2019371 </li></ul><ul><li>Median = 1.5137383 </li></ul><ul><li>Mode #N/A </li></ul><ul><li>Standard Deviation = 0.6995308 </li></ul><ul><li>Sample Variance = 0.4893433 </li></ul><ul><li>Kurtosis = 0.4309225 </li></ul><ul><li>Skewness = 0.5664844 </li></ul><ul><li>Minimum 0.5454545 Maximum 3.0 </li></ul><ul><li>Count = 12 </li></ul><ul><li>Confidence Level(95.0%) = 0.4444606 </li></ul><ul><li>Mean = 1.509206 </li></ul><ul><li>Standard Error = 0.2019371 </li></ul><ul><li>Median = 1.5137383 </li></ul><ul><li>Mode #N/A </li></ul><ul><li>Standard Deviation = 0.6995308 </li></ul><ul><li>Sample Variance = 0.4893433 </li></ul><ul><li>Kurtosis = 0.4309225 </li></ul><ul><li>Skewness = 0.5664844 </li></ul><ul><li>Minimum = 0.5454545 Maximum = 3.0 </li></ul><ul><li>Count = 12 </li></ul><ul><li>Confidence Level(95.0%) = 0.4444606 </li></ul>
  24. 24. COMPENSATORY RESPONSE HYPOTHESES… <ul><li>Bc-all FREQ+11 / lwr FREQ ratio </li></ul><ul><li>Simulation: Summary Statistics on Median Data </li></ul><ul><li>Mean = 2.828184 Tumors </li></ul><ul><li>Standard Error = 0.616883 </li></ul><ul><li>Median = 1.742237 </li></ul><ul><li>Mode = 3.0 </li></ul><ul><li>Standard Deviation = 2.136946 </li></ul><ul><li>Sample Variance = 4.56654 </li></ul><ul><li>Kurtosis = 1.466792 </li></ul><ul><li>Skewness = 1.543556 </li></ul><ul><li>Minimum = 0.691358 Maximum = 7.5 </li></ul><ul><li>Count = 12 </li></ul><ul><li>Confidence Level(95.0%) = 1.357751 </li></ul><ul><li>Mean = 5.530856 Tumors </li></ul><ul><li>Standard Error = 2.08372 </li></ul><ul><li>Median = 1.711034 is close to +11 Summary Statistics </li></ul><ul><li>Mode = 1.513738 = Median value prev. </li></ul><ul><li>Standard Deviation = 6.589301 </li></ul><ul><li>Sample Variance = 43.41889 </li></ul><ul><li>Kurtosis = -0.73807 </li></ul><ul><li>Skewness = 1.136088 </li></ul><ul><li>Minimum = 0.703297 Maximum =16.65112 </li></ul><ul><li>Count 10 = # of simulations= 10 medians from various ratio simulations were generated and summery statistics generated here is from those 10 medians. </li></ul><ul><li>Confidence Level(95.0%) = 4.713702 </li></ul>

×