Structural Identifiability in Low-Rank Matrix Factorization

  • Epameinondas Fritzilas
  • Yasmin A. Rios-Solis
  • Sven Rahmann
Part of the Lecture Notes in Computer Science book series (LNCS, volume 5092)


In many signal processing and data mining applications, we need to approximate a given matrix Y of “sensor measurements” over several experiments by a low-rank product Y ≈ A·X, where X contains source signals for each experiment, A contains source-sensor mixing coefficients, and both A and X are unknown. We assume that the only a-priori information available is that A must have zeros at certain positions; this constrains the source-sensor network connectivity pattern.

In general, different AX factorizations approximate a given Y equally well, so a fundamental question is how the connectivity restricts the solution space. We present a combinatorial characterization of uniqueness up to diagonal scaling, called structural identifiability of the model, using the concept of structural rank from combinatorial matrix theory.

Next, we define an optimization problem that arises in the need for efficient experimental design: to minimize the number of sensors while maintaining structural identifiability. We prove its NP-hardness and present a mixed integer linear programming framework with two cutting-plane approaches. Finally, we experimentally compare these approaches on simulated instances of various sizes.


Unable to display preview. Download preview PDF.

Unable to display preview. Download preview PDF.


  1. 1.
    Boscolo, R., Sabatti, C., Liao, J., Roychowdhury, V.: A Generalized Framework for Network Component Analysis. IEEE Trans. Comp. Biol. Bioinf. 2, 289–301 (2005)CrossRefGoogle Scholar
  2. 2.
    Brunet, J.P., Tamayo, P., Golub, T.R., Mesirov, J.P.: Metagenes and Molecular Pattern Discovery Using Matrix Factorization. PNAS 101, 4164–4169 (2004)CrossRefGoogle Scholar
  3. 3.
    Garey, M., Johnson, D.: Computers and Intractability: A Guide to the Theory of NP-Completeness. W.H. Freeman and Company, New York (1979)zbMATHGoogle Scholar
  4. 4.
    Hopcroft, J., Karp, R.: An n 5 / 2 Algorithm for Maximum Matchings in Bipartite Graphs. SIAM J. Comput. 2, 225–231 (1973)zbMATHCrossRefMathSciNetGoogle Scholar
  5. 5.
    Lee, D., Seung, H.: Learning the Parts of Objects by Non-negative Matrix Factorization. Nature 401, 788–791 (1999)CrossRefGoogle Scholar
  6. 6.
    Liao, J., Boscolo, R., Yang, Y.L., Tran, L., Sabatti, C., Roychowdhury, V.: Network Component Analysis: Reconstruction of Regulatory Signals in Biological Systems. PNAS 100, 15522–15527 (2003)CrossRefGoogle Scholar
  7. 7.
    Lin, C.J.: Projected Gradient Methods for Non-Negative Matrix Factorization. Neural Computation 19, 2756–2779 (2007)zbMATHCrossRefMathSciNetGoogle Scholar
  8. 8.
    Lovasz, L., Plummer, M.: Matching Theory. North-Holland, Amsterdam (1986)zbMATHGoogle Scholar
  9. 9.
    Murota, K.: Matrices and Matroids for Systems Analysis. Springer, Heidelberg (2000)zbMATHGoogle Scholar
  10. 10.
    Narasimhan, S., Rengaswamy, R., Vadigepalli, R.: Structural Properties of Gene Regulatory Networks: Definitions and Connections. IEEE Trans. Comp. Biol. Bioinf. (accepted, 2007)Google Scholar
  11. 11.
    Shahnaz, F., Berry, M., Pauca, V., Plemmons, R.: Document Clustering Using Nonnegative Matrix Factorization. Inf. Proc. & Manag. 42, 373–386 (2006)zbMATHCrossRefGoogle Scholar
  12. 12.
    Wolsey, L.: Integer Programming. Wiley Interscience, Chichester (1998)zbMATHGoogle Scholar

Copyright information

© Springer-Verlag Berlin Heidelberg 2008

Authors and Affiliations

  • Epameinondas Fritzilas
    • 1
  • Yasmin A. Rios-Solis
    • 1
  • Sven Rahmann
    • 2
  1. 1.Faculty of TechnologyBielefeld UniversityGermany
  2. 2.Computer Science 11Technische Universität DortmundGermany

Personalised recommendations