Catálogo de publicaciones - libros

Genetic instability represents an important type of biological markers for cancer and many other diseases. Array Comparative Genome Hybridization (aCGH) is a high-throughput cytogenetic technique that can efficiently detect genome-wide genetic instability events such as chromosomal gain, loss, and more complex aneuploidity, collectively known as genome imbalance (GIM). We propose a new statistical method, Genome Imbalance Scanner (GIMscan), for automatically decoding the underlying DNA dosage states from aCGH data. GIMscan captures both the intrinsic (nonrandom) spatial change of genome hybridization intensities, and the prevalent (random) measurement noise during data acquisition; and it simultaneously segments the chromosome and assigns different states to the segmented DNA. We tested the proposed method on both simulated data and real data measured from a colorectal cancer population, and we report competitive or superior performance of GIMscan in comparison with popular extant methods.

Given the substantial computational requirements of stochastic simulation, approximation is essential for efficient analysis of any realistic biochemical system. This paper introduces a new approximation method to reduce the computational cost of stochastic simulations of an enzymatic reaction scheme which in biochemical systems often includes rapidly changing fast reactions with enzyme and enzyme-substrate complex molecules present in very small counts. Our new method removes the substrate dissociation reaction by approximating the passage time of the formation of each enzyme-substrate complex molecule which is destined to a production reaction. This approach skips the firings of unimportant yet expensive reaction events, resulting in a substantial acceleration in the stochastic simulations of enzymatic reactions. Additionally, since all the parameters used in our new approach can be derived by the Michaelis-Menten parameters which can actually be measured from experimental data, applications of this approximation can be practical even without having full knowledge of the underlying enzymatic reaction. Furthermore, since our approach does not require a customized simulation procedure for enzymatic reactions, it allows biochemical systems that include such reactions to still take advantage of standard stochastic simulation tools. Here, we apply this new method to various enzymatic reaction systems, resulting in a speedup of orders of magnitude in temporal behavior analysis without any significant loss in accuracy.

Specificity of MHC binding to short peptide fragments from cellular as well as pathogens’ proteins has been found to correlate with disease outcome and pathogen or cancer evolution. The large variation in MHC class II epitope length has complicated training of predictors for binding affinities compared to MHC class I. In this paper, we treat the relative position of the peptide inside the MHC protein as a hidden variable, and model the ensemble of different binding configurations. The training procedure iterates the predictions with re estimation of the parameters of a binding groove model. We show that the model generalizes to new MHC class II alleles, which were not a part of the training set. To the best of our knowledge, our technique outperforms all previous approaches to MHC II epitope prediction. We demonstrate how our model can be used to explain previously documented associations between MHC II alleles and disease.

The study of mobile element evolution yields valuable insights into the mechanism and history of genome rearrangement, and can help answer questions about our evolutionary history. However, because the mammalian genome contains millions of copies of mobile elements exhibiting a complex evolutionary history, traditional phylogenetic methods are ill-suited to reconstructing their history. New phylogenetic reconstruction algorithms which exploit the unique properties of mobile elements and handle large numbers of repeats are therefore necessary to better understand both mobile elements’ evolution and our own.

We describe a randomized algorithm for phylogenetic reconstruction that scales easily to a million or more elements. We apply our algorithm to human and chimpanzee and L1 elements, and to SINE elements from 61 species, finding 32 new L1, 111 new SINE, and over 1000 new subfamilies. Our results suggest that the history of mobile elements is significantly more complex than we currently understand.

Reticulate networks are a type of phylogenetic network that are used to represent reticulate evolution involving hybridization, horizontal gene transfer or recombination. The simplest form of these networks are galled trees, in which all reticulations are independent of each other. This paper introduces a more general class of reticulate networks, that we call galled networks, in which reticulations are not necessarily independent, but may overlap in a tree-like manner. We prove a Decomposition Theorem for these networks that has important consequences for their computation, and present a fixed-parameter-tractable algorithm for computing such networks from trees or binary sequences. We provide a robust implementation of the algorithm and illustrate its use on two biological datasets, one based on a set of three gene-trees and the other based on a set of binary characters obtained from a restriction site map.

Probabilistic phylogenetic models which relax the site independence evolution assumption often face the problem of infeasible likelihood computations, for example for the task of selecting suitable parameters for the model. We present a new approximation method, applicable for a wide range of probabilistic models, which guarantees to upper bound the true likelihood of data, and apply it to the problem of probabilistic phylogenetic models. The new method is complementary to known variational methods that lower bound the likelihood, and it uses similar methods to optimize the bounds from above and below. We applied our method to aligned DNA sequences of various lengths from human in the region of the CFTR gene and homologous from eight mammals, and found the upper bounds to be appreciably close to the true likelihood whenever it could be computed. When computing the exact likelihood was not feasible, we demonstrated the proximity of the upper and lower variational bounds, implying a tight approximation of the likelihood.

The gene-duplication problem is to infer a species supertree from a collection of gene trees that are confounded by complex histories of gene duplications. This problem is NP-hard and thus requires efficient and effective heuristics. Existing heuristics perform a stepwise search of the tree space, where each step is guided by an exact solution to an instance of a local search problem. We show how this local search problem can be solved efficiently by reusing previously computed information. This improves the running time of the current solution by a factor of , where is the number of species in the resulting supertree solution, and makes the gene-duplication problem more tractable for large-scale phylogenetic analyses. We verify the exceptional performance of our solution in a comparison study using sets of large randomly generated gene trees. Furthermore, we demonstrate the utility of our solution by incorporating large genomic data sets from GenBank into a supertree analysis of plants.

Sequence to structure alignment is an important step in homology modeling of protein structures. Incorporation of features like secondary structure, solvent accessibility, or evolutionary information improve sequence to structure alignment accuracy, but conventional generative estimation techniques for alignment models impose independence assumptions that make these features difficult to include in a principled way. In this paper, we overcome this problem using a Support Vector Machine (SVM) method that provides a well-founded way of estimating complex alignment models with hundred-thousands of parameters. Furthermore, we show that the method can be trained using a variety of loss functions. In a rigorous empirical evaluation, the SVM algorithm outperforms the generative alignment method SSALN, a highly accurate generative alignment model that incorporates structural information. The alignment model learned by the SVM aligns 47% of the residues correctly and aligns over 70% of the residues within a shift of 4 positions.

Machine learning, Pairwise sequence alignment, Protein structure prediction.

It has recently been found that some RNA functions are determined by the actual and not just the RNA’s nucleotide sequence or its native structure. We present new computational tools for simulating and analyzing RNA folding kinetic metrics such as population kinetics, folding rates, and the folding of particular subsequences. Our method first builds an approximate representation (called a map) of the RNA’s folding energy landscape, and then uses specialized analysis techniques to extract folding kinetics from the map. We provide a new sampling strategy called Probabilistic Boltzmann Sampling (PBS) that enables us to approximate the folding landscape with much smaller maps, typically by several orders of magnitude. We also describe a new analysis technique, Map-based Monte Carlo (MMC) simulation, to stochastically extract folding pathways from the map. We demonstrate that our technique can be applied to large RNA (e.g., 200+ nucleotides), where representing the full landscape is infeasible, and that our tools provide results comparable to other simulation methods that work on complete energy landscapes. We present results showing that our approach computes the same relative functional rates as seen in experiments for the relative plasmid replication rates of ColE1 RNAII and its mutants, and for the relative gene expression rates of MS2 phage RNA and its mutants.

Despite considerable efforts, it remains difficult to obtain accurate multiple sequence alignments. By using additional hits from database search of the input sequences, a few strategies have been proposed to significantly improve alignment accuracy, including the construction of profiles from the hits while performing profile alignment, the inclusion of high scoring hits into the input sequences, the use of intermediate sequence search to link distant homologs, and the use of secondary structure information. We develop an algorithm that integrates these strategies to further improve alignment accuracy by modifying the pair-HMM approach in ProbCons to incorporate profiles of intermediate sequences from database search and utilize secondary structure predictions as in SPEM. We test our algorithm on a few sets of benchmark multiple alignments, including BAliBASE, HOMSTRAD, PREFAB and SABmark, and show that it significantly outperforms MAFFT and ProbCons, which are among the best multiple alignment algorithms that do not utilize additional information, and SPEM, which is among the best multiple alignment algorithms that utilize additional hits from database search. The improvement in accuracy over SPEM can be as much as 5 to 10% when aligning divergent sequences. A software program that implements this approach (ISPAlign) is at http://faculty.cs.tamu.edu/shsze/ispalign.