Send to

Choose Destination
See comment in PubMed Commons below
Bioinformatics. 2014 Jan 15;30(2):228-33. doi: 10.1093/bioinformatics/btt671. Epub 2013 Nov 18.

Combining optimization and machine learning techniques for genome-wide prediction of human cell cycle-regulated genes.

Author information

  • 1Dipartimento di Ingegneria Informatica, Automatica e Gestionale, Sapienza Universit√† di Roma, Rome, Italy and Istituto Nazionale dei Tumori Regina Elena, Dipartimento di Oncologia Sperimentale, Laboratorio di Oncogenesi Molecolare, Rome, Italy.



The identification of cell cycle-regulated genes through the cyclicity of messenger RNAs in genome-wide studies is a difficult task due to the presence of internal and external noise in microarray data. Moreover, the analysis is also complicated by the loss of synchrony occurring in cell cycle experiments, which often results in additional background noise.


To overcome these problems, here we propose the LEON (LEarning and OptimizatioN) algorithm, able to characterize the 'cyclicity degree' of a gene expression time profile using a two-step cascade procedure. The first step identifies a potentially cyclic behavior by means of a Support Vector Machine trained with a reliable set of positive and negative examples. The second step selects those genes having peak timing consistency along two cell cycles by means of a non-linear optimization technique using radial basis functions. To prove the effectiveness of our combined approach, we use recently published human fibroblasts cell cycle data and, performing in vivo experiments, we demonstrate that our computational strategy is able not only to confirm well-known cell cycle-regulated genes, but also to predict not yet identified ones.


All scripts for implementation can be obtained on request.

[PubMed - indexed for MEDLINE]
Free full text
PubMed Commons home

PubMed Commons

How to join PubMed Commons

    Supplemental Content

    Full text links

    Icon for HighWire
    Loading ...
    Write to the Help Desk