Faculty Bibliography

We developed a model of 545 components (nodes) and 1259 interactions representing signaling pathways and cellular machines in the hippocampal CA1 neuron. Using graph theory methods, we analyzed ligand-induced signal flow through the system. Specification of input and output nodes allowed us to identify functional modules. Networking resulted in the emergence of regulatory motifs, such as positive and negative feedback and feedforward loops, that process information. Key regulators of plasticity were highly connected nodes required for the formation of regulatory motifs, indicating the potential importance of such motifs in determining cellular choices between homeostasis and plasticity.

Recent observations show that the single-cell response of p53 to ionizing radiation (IR) is "digital" in that it is the number of oscillations rather than the amplitude of p53 that shows dependence on the radiation dose. We present a model of this phenomenon. In our model, double-strand break (DSB) sites induced by IR interact with a limiting pool of DNA repair proteins, forming DSB-protein complexes at DNA damage foci. The persisting complexes are sensed by ataxia telangiectasia mutated (ATM), a protein kinase that activates p53 once it is phosphorylated by DNA damage. The ATM-sensing module switches on or off the downstream p53 oscillator, consisting of a feedback loop formed by p53 and its negative regulator, Mdm2. In agreement with experiments, our simulations show that by assuming stochasticity in the initial number of DSBs and the DNA repair process, p53 and Mdm2 exhibit a coordinated oscillatory dynamics upon IR stimulation in single cells, with a stochastic number of oscillations whose mean increases with IR dose. The damped oscillations previously observed in cell populations can be explained as the aggregate behavior of single cells.

A major challenge in cancer diagnosis from microarray data is the need for robust, accurate, classification models which are independent of the analysis techniques used and can combine data from different laboratories. We propose such a classification scheme originally developed for phenotype identification from mass spectrometry data. The method uses a robust multivariate gene selection procedure and combines the results of several machine learning tools trained on raw and pattern data to produce an accurate meta-classifier. We illustrate and validate our method by applying it to gene expression datasets: the oligonucleotide HuGeneFL microarray dataset of Shipp et al. (www.genome.wi.mit.du/MPR/lymphoma) and the Hu95Av2 Affymetrix dataset (DallaFavera's laboratory, Columbia University). Our pattern-based meta-classification technique achieves higher predictive accuracies than each of the individual classifiers , is robust against data perturbations and provides subsets of related predictive genes. Our techniques predict that combinations of some genes in the p53 pathway are highly predictive of phenotype. In particular, we find that in 80% of DLBCL cases the mRNA level of at least one of the three genes p53, PLK1 and CDK2 is elevated, while in 80% of FL cases, the mRNA level of at most one of them is elevated.

One of the major challenges in cancer diagnosis from microarray data is to develop robust classification models which are independent of the analysis techniques used and can combine data from different laboratories. We propose a meta-classification scheme which uses a robust multivariate gene selection procedure and integrates the results of several machine learning tools trained on raw and pattern data. We validate our method by applying it to distinguish diffuse large B-cell lymphoma (DLBCL) from follicular lymphoma (FL) on two independent datasets: the HuGeneFL Affmetrixy dataset of Shipp et al. (www. genome.wi.mit.du/MPR /lymphoma) and the Hu95Av2 Affymetrix dataset (DallaFavera's laboratory, Columbia University). Our meta-classification technique achieves higher predictive accuracies than each of the individual classifiers trained on the same dataset and is robust against various data perturbations. We also find that combinations of p53 responsive genes (e.g., p53, PLK1 and CDK2) are highly predictive of the phenotype.

When the genomic integrity of a cell is challenged, its fate is determined in part by signals conveyed by the p53 tumour suppressor protein. It was observed recently that such signals are not simple gradations of p53 concentration, but rather a counter-intuitive limit-cycle behaviour. Based on a careful mathematical interpretation of the experimental body of knowledge, we propose a model for the p53 signalling network and characterise the p53 stability and oscillatory dynamics. In our model, ATM, a protein that senses DNA damage, activates p53 by phosphorylation. In its active state, p53 has a decreased degradation rate and an enhanced transactivation of Mdm2, a gene whose protein product Mdm2 tags p53 for degradation. Thus the p53-Mdm2 system forms a negative feedback loop. However, the feedback in this loop is delayed, as the pool of Mdm2 molecules being induced by p53 at a given time will mark for degradation the pool of p53 molecules at some later time, after the Mdm2 molecules have been transcribed, exported out of the nucleus, translated and transported back into the nucleus. The analysis of our model demonstrates how this time lag combines with the ATM-controlled feedback strength and effective dampening of the negative feedback loop to produce limit-cycle oscillations. The picture that emerges is that ATM, once activated by DNA damage, makes the p53-Mdm2 oscillator undergo a supercritical Hopf bifurcation. This approach yields an improved understanding of the global dynamics and bifurcation structure of our time-delayed, negative feedback model and allows for predictions of the behaviour of the p53 system under different perturbations.

MOTIVATION/BACKGROUND:Despite the growing literature devoted to finding differentially expressed genes in assays probing different tissues types, little attention has been paid to the combinatorial nature of feature selection inherent to large, high-dimensional gene expression datasets. New flexible data analysis approaches capable of searching relevant subgroups of genes and experiments are needed to understand multivariate associations of gene expression patterns with observed phenotypes. RESULTS:We present in detail a deterministic algorithm to discover patterns of multivariate gene associations in gene expression data. The patterns discovered are differential with respect to a control dataset. The algorithm is exhaustive and efficient, reporting all existent patterns that fit a given input parameter set while avoiding enumeration of the entire pattern space. The value of the pattern discovery approach is demonstrated by finding a set of genes that differentiate between two types of lymphoma. Moreover, these genes are found to behave consistently in an independent dataset produced in a different laboratory using different arrays, thus validating the genes selected using our algorithm. We show that the genes deemed significant in terms of their multivariate statistics will be missed using other methods. AVAILABILITY/BACKGROUND:Our set of pattern discovery algorithms including a user interface is distributed as a package called Genes@Work. This package is freely available to non-commercial users and can be downloaded from our website (http://www.research.ibm.com/FunGen).

We have developed a model of cardiac thin filament activation using an Ising model approach from equilibrium statistical physics. This model explicitly represents nearest-neighbor interactions between 26 troponin/tropomyosin units along a one-dimensional array that represents the cardiac thin filament. With transition rates chosen to match experimental data, the results show that the resulting force-pCa (F-pCa) relations are similar to Hill functions with asymmetries, as seen in experimental data. Specifically, Hill plots showing (log(F/(1-F)) vs. log [Ca]) reveal a steeper slope below the half activation point (Ca(50)) compared with above. Parameter variation studies show interplay of parameters that affect the apparent cooperativity and asymmetry in the F-pCa relations. The model also predicts that Ca binding is uncooperative for low [Ca], becomes steeper near Ca(50), and becomes uncooperative again at higher [Ca]. The steepness near Ca(50) mirrors the steep F-pCa as a result of thermodynamic considerations. The model also predicts that the correlation between troponin/tropomyosin units along the one-dimensional array quickly decays at high and low [Ca], but near Ca(50), high correlation occurs across the whole array. This work provides a simple model that can account for the steepness and shape of F-pCa relations that other models fail to reproduce.

B-chronic lymphocytic leukemia (B-CLL) is an adult-onset leukemia characterized by significant accumulation of apoptosis-resistant monoclonal B lymphocytes. In this study, we performed gene expression profiling on B cells obtained from 10 healthy age-matched individuals and CLL B cells from 38 B-CLL patients to identify key genetic differences between CLL and normal B cells. In addition, we leveraged recent independent studies to assess the reproducibility of our molecular B-CLL signature. We used a novel combination of several methods of data analysis including our own software and identified 70 previously unreported genes that differentiate leukemic cells from normal B cells, as well as confirmed recently reported B-CLL specific expression levels of an additional 10 genes. Importantly, many of these genes have previously been linked with other cancers, thus lending further support to their importance as candidate genes leading to B-CLL pathogenesis. We have also validated a subset of these genes using independent methodologies. Moreover, we show that our genes can be used to create a diagnostics signature that performs with perfect sensitivity and specificity in an independent cohort of 21 B-CLL and 20 normal subjects, thus strongly validating the informative nature of our set of genes. Finally, we identified a group of 31 genes that distinguish between low (Rai stage 0) and high (Rai stage 4) risk patients, suggesting that there may also be a gene expression signature that associates with disease progression.

We applied an automatic and unsupervised system to a nearly complete database of mammalian odor receptor genes. The generated motifs and gene classification were subjected to extensive and systematic downstream analysis to obtain biological insights. Two major results from this analysis were: (1) a map of sequence motifs that may correlate with function and (2) the corresponding receptor classes in which members of each class are likely to share specific functions. We have discovered motifs that have been implicated in structural integrity and posttranslational modification, as well as motifs very likely to be directly involved in ligand binding. We further propose a combinatorial molecular hypothesis, based on unique combinations of the observed motifs, that provides a foundation for understanding the generation of a large number of ligand binding sites.

Gene expression array data provide shadows of intricate cellular processes. Learning how to make the most of the information present in expression arrays has become a discipline in itself. In recent years, there has been an explosion of methods that analyze gene expression arrays to produce long lists of genes that express differentially in distinct cellular states. These lists will have to be organized, and the algorithms that produced them combined, if we wish to piece together the rich cellular structures probed by this high-throughput technology. Researchers will have to understand the benefits and limitations of the many existing methods to produce the combination of algorithms that best suits their gene expression experiments.