Dimitris Anastassiou, Ph.D.
Past Research: Dr. Anastassiou’s previous research interests have been in the area of digital signal processing and information theory with emphasis on the digital representation of multimedia signals, with contributions to the international digital television coding standard, MPEG-2. He is the founder and previous Director of Columbia University’s Image and Advanced Television Laboratory. He is now Director of Columbia University’s Genomic Information Systems Laboratory, in the Columbia Genome Center, and his research is exclusively focused on using his expertise in several traditional engineering disciplines to the emerging field of genomics.
Current Research: The emphasis of the new laboratory is in the area of analysis and simulation of complex systems resulting from networks of mutually interacting biomolecular processes and leading to the coordinated timing of multiple gene activation events. Such processes include gene expression and regulatory pathways, signal transduction, metabolic pathways, cell cycles, membrane transport, cell-to-cell interactions, as well as integrative models of multiple processes. For example, the outputs of gene regulatory networks can be seen as “scripts” – series of coordinated and well-timed events of activation or deactivation of genes, resulting in the assembly of complex “protein machines” at the nanoscale, and the resulting execution of essential cellular processes. In a real sense, these processes are an example of “nanotechnology in action,” and their assembly and operation can conceivably be controlled using quantitative analysis and properly modifying of external stimuli and initial conditions in the describing systems. This formulation is in accordance with the vision of “systems biology”, in which molecular biology and medicine are treated quantitatively as information sciences, and investigated according to a “system-based” approach involving information transfer through intracellular molecular interactions.
Networks of biomolecular interactions are typically modeled by a system of a large number of nonlinear differential equations. At sufficiently high temporal resolution, average rates are replaced by discrete biomolecular events. For example, transcription initiation events in a particular gene can be modeled as “spike trains” in a manner reminiscent of the neural spike trains.
Quantitative analysis is prohibitively complex but can be aided by various models familiar to engineering research. For example, pathways involved in cellular signaling and gene regulatory networks can be modeled as neural networks, or as telecommunication networks that can be analyzed with several probabilistic approaches (such as queuing networks) already familiar to the telecommunications network research community in electrical engineering and computer science; or, by hardware circuit simulation using VLSI technology directly implementing such complex systems.
Developmental pathways that give rise to disease such as cancer may sometimes be strongly related with malfunctioning of such networks. By analyzing the networks as systems (as opposed to merely examining the causal effect of individual nodes in isolation of the others), we are augmenting the set of potential drugable targets, conceivably providing new avenues for therapeutic drug design. Such research relies upon the expertise of cancer biologists, and the assistance of supercomputing simulation facilities, as the mathematics involved in such pathways are prohibitively complex.
One way of addressing complexity is by simulating system behavior using parallel supercomputers, in which processors are associated with network nodes. In such network simulations, orderly outputs occasionally emerge in ways that cannot be predicted or understood in advance. Supercomputers offer the possibility of planned experimentation and observation patterns that can be interpreted at a later time. Such simulations will be implemented on forthcoming IBM “Regatta-H” computers that will be installed at the Columbia Genome Center and will be available for other computational needs of researchers in the Center. These are large consolidated machines based on the forthcoming “Power-4” processors, featuring minimal communication latencies among processors, so they are ideally suited for the needs of the laboratory.
D. Anastassiou, “Frequency-Domain analysis of Biomolecular Sequences,” Bioinformatics, vol. 16, no.12, December 2000, pp. 1073-1081.
D. Anastassiou, “Genomic Signal Processing”, IEEE Signal Processing Magazine, magazine theme article, July 2001.
D. Anastassiou, “DSP in Genomics”, Proceedings, IEEE International conference ICASSP 2001, Salt Lake City, Utah, May 2001.
D. Anastassiou, “Digital Signal Processing of Biomolecular Sequences”, Technical Report EE000420-1, April 2000.