Sensor-based human activity recognition (HAR) requires to predict the action of a person based on sensor-generated time series data. HAR has attracted major interest in the past few years, thanks to the large number of applications enabled by modern ubiquitous computing devices. While several techniques based on hand-crafted feature engineering have been proposed, the current state-of-the-art is represented by deep learning architectures that automatically obtain high level representations and that use recurrent neural networks (RNNs) to extract temporal dependencies in the input. RNNs have several limitations, in particular in dealing with long-term dependencies. We propose a novel deep learning framework, TrASenD, based on a purely attention-based mechanism, that overcomes the limitations of the state-of-the-art. We show that our proposed attention-based architecture is considerably more powerful than previous approaches, with an average increment, of more than 7% on the F1 score over the previous best performing model. Furthermore, we consider the problem of personalizing HAR deep learning models, which is of great importance in several applications. We propose a simple and effective transfer-learning based strategy to adapt a model to a specific user, providing an average increment of 6% on the F1 score on the predictions for that user. Our extensive experimental evaluation proves the significantly superior capabilities of our proposed framework over the current state-of-the-art and the effectiveness of our user adaptation technique.
Advances in DNA sequencing technologies have allowed the characterization of somatic mutations in a large number of cancer genomes at an unprecedented level of detail, revealing the extreme genetic heterogeneity of cancer at two different levels: inter-tumor, with different patients of the same cancer type presenting different collections of somatic mutations, and intra-tumor, with different clones coexisting within the same tumor. Both inter-tumor and intra-tumor heterogeneity have crucial implications for clinical practices. Here, we review computational methods that use somatic alterations measured through next-generation DNA sequencing technologies for characterizing tumor heterogeneity and its association with clinical variables. We first review computational methods for studying inter-tumor heterogeneity, focusing on methods that attempt to summarize cancer heterogeneity by discovering pathways that are commonly mutated across different patients of the same cancer type. We then review computational methods for characterizing intra-tumor heterogeneity using information from bulk sequencing data or from single cell sequencing data. Finally, we present some of the recent computational methodologies that have been proposed to identify and assess the association between inter- or intra-tumor heterogeneity with clinical variables.
9 December 2023 in ACM Transactions on Knowledge Discovery From Data
“Sim Sala Bim!” —Silvan, https://en.wikipedia.org/wiki/Silvan_(illusionist) Betweenness centrality is a popular centrality measure with applications in several domains and whose exact computation is impractical for modern-sized networks. We present SILVAN , a novel, efficient algorithm to compute, with high probability, accurate estimates of the betweenness centrality of all nodes of a graph and a high-quality approximation of the top- k betweenness centralities. SILVAN follows a progressive sampling approach and builds on novel bounds based on Monte Carlo Empirical Rademacher Averages, a powerful and flexible tool from statistical learning theory. SILVAN relies on a novel estimation scheme providing non-uniform bounds on the deviation of the estimates of the betweenness centrality of all the nodes from their true values and a refined characterisation of the number of samples required to obtain a high-quality approximation. Our extensive experimental evaluation shows that SILVAN extracts high-quality approximations while outperforming, in terms of number of samples and accuracy, the state-of-the-art approximation algorithm with comparable quality guarantees.
Tumors are the result of a somatic evolutionary process leading to substantial intra-tumor heterogeneity. Single-cell and multi-region sequencing enable the detailed characterization of the clonal architecture of tumors and have highlighted its extensive diversity across tumors. While several computational methods have been developed to characterize the clonal composition and the evolutionary history of tumors, the identification of significantly conserved evolutionary trajectories across tumors is still a major challenge. We present a new algorithm, MAximal tumor treeS TRajectOries (MASTRO), to discover significantly conserved evolutionary trajectories in cancer. MASTRO discovers all conserved trajectories in a collection of phylogenetic trees describing the evolution of a cohort of tumors, allowing the discovery of conserved complex relations between alterations. MASTRO assesses the significance of the trajectories using a conditional statistical test that captures the coherence in the order in which alterations are observed in different tumors. We apply MASTRO to data from nonsmall-cell lung cancer bulk sequencing and to acute myeloid leukemia data from single-cell panel sequencing, and find significant evolutionary trajectories recapitulating and extending the results reported in the original studies. MASTRO is available at https://github.com/VandinLab/MASTRO. Supplementary data are available at Bioinformatics online.
The identification of significant patterns, defined as patterns whose frequency significantly deviates from what is expected under a suitable null model of the data, is a key data mining task with application in several areas. We present PROMISE, an algorithm for identifying significant sequential patterns while guaranteeing that the probability that one or more false discoveries are reported in output (i.e., the Family-Wise Error Rate - FWER) is less than a user-defined threshold. PROMISE employs the Westfall-Young method to correct for multiple hypothesis testing, a more powerful method than the commonly used Bonferroni correction. PROMISE crucially hinges on the generation of (random) permuted datasets with features similar to the input dataset, for which we provide two efficient strategies. We also provide a rigorous analysis of one of such strategies, which is based on a properly defined swap operation, proving a rigorous bound on the number of swaps it requires. The results of our experimental evaluation show that PROMISE is an efficient method that allows the discovery of statistically significant sequential patterns from transactional datasets while properly controlling for false discoveries.
9 June 2020 in Data Mining and Knowledge Discovery