The registration is closed.
The registration is closed.
Within the fast-rate growing area of Systems Biology, the last decade has witnessed the increasing development of several computational methodologies, modelling approaches, analysis methods and biotechnological procedures for the holistic investigation of biological systems, to the purpose of gaining a system-level understanding of their complex structure and dynamics.
Mathematical modeling and simulation algorithms nowadays provide solid grounds for quantitative investigations of biological systems, in a synergistic way with traditional experimental research. Despite the possibility offered by various computational methods to achieve an in-depth understanding of cells functioning, typical tasks for model definition, calibration and analysis (e.g., reverse engineering, parameter estimation, sensitivity analysis, etc.) are still computationally challenging. Indeed, these problems require to execute a large number of simulations of the same model, each one corresponding to a different physiological or perturbed system condition. In addition, in the case of large-scale systems, characterized by hundreds or thousands species and reactions, even a single simulation can be unfeasible if executed on conventional computing architectures like Central Processing Units (CPUs).
To overcome these drawbacks, parallel infrastructures can be used to strongly reduce the prohibitive running times of computational methods in Systems Biology, by distributing the workload over multiple independent computing units. In particular, General-Purpose Graphics Processing Units (GPUs) are gaining an increasing attention by the scientific community, since they are pervasive, cheap and extremely efficient parallel multi-core coprocessors, which give access to low-cost, energy-efficient means to achieve tera-scale performances on common workstations.
In this talk, we present the GPU-powered simulators that we designed and implemented to accelerate the simulation and analysis of reaction-based models of biological systems, which can rely either on numerical integration methods or stochastic simulation algorithms. Three models of biological systems characterized by an increasing complexity: the Michaelis-Menten (MM) enzymatic kinetics; a model of gene expression in prokaryotic organisms (PGN); the Ras/cAMP/PKA signaling pathway in the yeast S. cerevisiae, are used as test cases to show the speed-up obtained by GPU-based tools.
Combined with this approach, it is proficient the use of Genetic Algorithms, or Particle Swarm Optimization (PSO) methods, which allow the characterization of the unknown parameters (i.e., binding constants, transcription or translation rates, diffusion rate, etc.) of the system which, except for few special cases where these values can be found in literature, are not available or else ambiguous due to the difficulty in performing accurate measures during in vivo or in vitro experiments.
Nowadays, the amount of heterogeneous biomedical data is increasing more and more thanks to the advancements in imaging acquisition modalities and High-Throughput technologies. This huge information ensemble could overwhelm analytic capabilities concerning both physicians in their decision-making tasks and biologists in investigating complex biological systems.
Quantitative imaging methods provide scientifically and clinically relevant data in prediction, prognostication or response assessment, by also exploiting radiomics approaches. In this regard, Machine Learning and Computational Intelligence can significantly improve traditional Image Processing techniques. Therefore, advanced computational approaches for medical and biological image analysis play a key role in radiology and laboratory applications. However, conventional Machine Learning and Computational techniques must be adapted and tailored to address the unique challenges regarding biomedical images.
In this talk, the challenges and the characteristics of the most recent methods will be introduced and discussed. I will start with some practical applications exploiting classic Image Processing and Pattern Recognition techniques. Afterwards, a recent medical image enhancement method based on Genetic Algorithms will be briefly described. To conclude, the generalization capabilities of Convolutional Neural Networks in medical image segmentation tasks as well as the generation of realistic medical images based on Generative Adversarial Networks will be investigated.
The seminar will be held in INE Auditorium of NOVA IMS, located in Campus de Campolide, Lisbon.
Metro: S. Sebastião (Blue and Red Line); Praça de Espanha (Blue Line)
Carris: 701, 713, 716, 726, 742, 746, 756, 758, 770