Search

The functional or regulatory processes within the cell are explicitly governed by the expression levels of a subset of its genes. Gene expression time series captures activities of individual genes over time and aids revealing underlying cellular dynamics. An important step in high-throughput gene expression time series experiment is clustering genes based on their temporal expression patterns and is conventionally achieved by unsupervised machine learning techniques. However, most of the clustering techniques either suffer from the short length of gene expression time series or ignore temporal structure of the data. In this work, we propose DeepTrust, a novel deep learning-based framework for gene expression time series clustering which can overcome these issues. DeepTrust initially transforms time series data into images to obtain richer data representations. Afterwards, a deep convolutional clustering algorithm is applied on the constructed images. Analyses on both simulated and biological data sets exhibit the efficiency of this new framework, compared to widely used clustering techniques. We also utilize enrichment analyses to illustrate the biological plausibility of the clusters detected by DeepTrust. Our code and data are available from http://github.com/tanlab/DeepTrust.

The functional or regulatory processes within the cell are explicitly governed by the expression levels of a subset of its genes. Gene expression time series captures activities of individual genes over time and aids revealing underlying cellular dynamics. An important step in high-throughput gene expression time series experiment is clustering genes based on their temporal expression patterns and is conventionally achieved by unsupervised machine learning techniques. However, most of the clustering techniques either suffer from the short length of gene expression time series or ignore temporal structure of the data. In this work, we propose DeepTrust, a novel deep learning-based framework for gene expression time series clustering which can overcome these issues. DeepTrust initially transforms time series data into images to obtain richer data representations. Afterwards, a deep convolutional clustering algorithm is applied on the constructed images. Analyses on both simulated and biological data sets exhibit the efficiency of this new framework, compared to widely used clustering techniques. We also utilize enrichment analyses to illustrate the biological plausibility of the clusters detected by DeepTrust. Our code and data are available from http://github.com/tanlab/DeepTrust.

Repetitive Atrial Activation Patterns (RAAPs) detected in the coronary sinus (CS) during atrial fibrillation (AF) may represent a reference to construct composite maps of coincident local RAAPs elsewhere in the atria, potentially improving the identification of AF drivers. RAAP coincidence may however depend on AF complexity and may be affected by structural remodeling. Using computer simulations, we investigated coincidence and coupling of RAAPs in the CS and other regions in the atria, in the absence and presence of fibrosis. In our models, the CS displayed highly repetitive behavior unaffected by fibrosis. RAAP coincidence and coupling of other regions in the atria with the CS was high in the absence of fibrosis, but significantly decreased in the presence of fibrosis, most notably in the left atrium. In coincident RAAPs in the CS and atrial regions, quantification of the degree of RAAP coupling is required to provide further confirmation of the validity of employing CS electrograms as a reference for composite RAAPs maps.

Localized AF drivers with repetitive activity are candidate ablation targets for patients with persistent atrial fibrillation (AF). High-density mapping electrodes cover only a fraction of the atria but combining sequential recordings could provide a more comprehensive picture of common repetitive atrial conduction characteristics and enable AF driver localization. We developed a novel algorithm to merge overlapping local activation maps into larger composite maps by linking repetitive patterns detected in neighboring locations with similar conduction directions and cycle lengths. Regions exhibiting high curl, divergence and heterogeneity in composite maps were marked as candidate reentry locations and were compared to those estimated through phase singularities and cycle length coverage maps from the individual recordings. The proposed algorithm led to better estimates of the underlying source density (sensitivity: 0.88/0.87/0.79, specificity: 0.85/0.85/0.68 for stable reentry, meandering reentry, and collision, respectively), compared to the maps from individual recordings (sensitivities 0.85/0.70/0.65 and 0.84/0.86/0.51, specificities 0.86/0.70/0.64 and 0.85/0.87/0.50 for phase singularity and CL coverage, respectively).

26th IEEE Signal Processing and Communications Applications Conference (2018 : Izmir; Turkey), Developing personalized cancer treatment procedures requires a prior knowledge on the effects of different drugs on cancer cell lines. While obtaining this information in vitro is a tedious task, the emergence of numerous large-scale datasets facilitates the usage of machine learning algorithms for this purpose. Conventional methods make an effort to reveal the mapping function between a cell line's identifying features called gene expressions and a certain drug's effect on it. In this work, we move away from this philosophy and represent cell lines as images in which inter-feature relations are preserved. Once these images are obtained, the regression problem is solved with the help of a convolutional neural network, a neural network architecture proven to work well with image inputs. A benchmarking with the other models in the literature exhibits the fruitfulness of our novel strategy. © 2018 IEEE., Aselsan,et al.,Huawei,IEEE Signal Processing Society,IEEE Turkey Section,Netas

Rating driving performance is a challenging topic. It attracts professionals from a variety of domains such as automotive industry and insurance companies. A great majority of the previous studies combines multiple measurement modalities such as Controller Are Network (CAN Bus) data, physiological measurements, camera reconrdings and localization estimates from Global Positioning System (GPS). One school of thought attempted to discriminate agressive/non-agressive, attentive/inattentive or drowsy/wakeful drivers through a statistical learning. Other researchers applied a rule-based approach. However, this approaches are inapplicable since labelled data for supervised learning schemes is scarce and rules that are representative for all road conditions are not feasible. Moreover, the abundance of sensor modalities in a personal vehicle is rather costly. In order to handle these problems, in this work, we propose a fully unsupervised driving style scoring mechanism operating on a minimalistic dataset. The proposed model operates similar to conventional anomaly detecton schemes. In this setting, a driving experience is scored in proportion to its congruency to the driving norms which are obtained as the most common driving patters in the training data. As a novelity of our work, these norms are defined considering road type and traffic flow patterns. This is applied via a probabilistic approach where joint probability densities of the variables controlling road type, traffic flow type and driving style are required. Since estimating this probability is mathematically intractable, we follow an alternative approach relaxing the probability estimation through discretization. In this context, each of these variables are clustered by unsupervised learning techniques and the joint probabilities are approximated by the number elements shared between inter-variable clusters. This probability information is stored in a special architecture which we call Co-Clustering Matrix. (CC, Araç sürüş performansının ölçülmesi, özellikle otomotiv ve sigorta sektörlerinde çalışan araştırmacıların ilgisini çeken, oldukça zorlu bir konudur. Bu alandaki geçmiş çalışmaların bir kolu Denetleyici Alanı Veri Yolu Ağı (CAN Bus) ve Küresel Konum Belirleme Sistemi (GPS) çıktıları, fizyolojik veriler, kamera kayıtları ve pek çok diğer veri tipini öznitelik olarak kullanarak, etiketli veri setleri üzerinde agresif/agresif olmayan, dikkatli/dikkatsiz, uykulu/uykusuz gibi davranışsal ayrımları istatistiksel olarak öğrenmeyi amaçlamışlardır. Bir diğer akımda ise, araştırmacılar sürüş davranışlarını kural-bazlı olarak değerlendirmeyi tercih etmişlerdir. Ancak, bu yaklaşımlar etiketli verinin çoğu zaman mevcut olmaması, bütün yol şartlarını temsil edebilecek kuralların öğrenilememesi ve standart bir aracın gerekli bütün sensör modalitelerine sahip olmamasından dolayı kullanışlı değillerdir. Çalışmamızda, bu problemlerin hepsinin üstesinden gelen, minimalistik bir veri üzerinde skorlama yapma kapasitesine sahip, gözetimsiz bir olasılıksal model tasarlanmıştır. Sunulan model, sürücüleri geleneksel anomali tespiti yaklaşımlarıyla değerlendirir. Buna göre, bir sürüş tecrübesinin geçmişte görülen örnekler üzerinden hesaplanan normlara ne kadar uyumlu olduğu, onun ne kadar yüksek skorlanacağını tanımlar. Bu normlar, diğer çalışmalardan farklı olarak, yolun tipine ve trafik akışına bağlı olarak bulunur. Takip edilen olasılıksal yaklaşım, bu sürekli değişkenlerin bileşik olasılık dağılımlarının bilinmesini gerektirmektedir; ancak bu matematiksel olarak oldukça zorlu bir problemdir. Bu işlemi kolaylaştırmak için, değişkenlerden her birini gözetimsiz öğrenme yöntemleri ile ayrıklaştırma yoluna gidilmiştir. Bu sayede, değişkenleri ayrık az sayıda küme ile temsil etmek ve bu kümeler arasındaki paylaşılan eleman sayılarını kullanarak bileşik olasılık dağılımlarını kestirmek mümkün olmuştur. Bileşik dağılım bilgisi, Birlikte Kümelenme Matrisi (BKM) adlı bir yapıda tutulmuştur ve bu ma

Rating driving performance is a challenging topic. It attracts professionals from a variety of domains such as automotive industry and insurance companies. In this work, we propose a fully unsupervised driver scoring framework using a minimalistic dataset which is composed of Global Positioning System (GPS) and Controller Area Network (CAN Bus) data. Based on the natural expectation that good driving patterns should depend on the road type and traffic flow intensity, our framework attempts to assign a probabilistic score in proportion to the occurrence probability of a certain driving style given the road geometry and traffic conditions. Quantization of these random variables through clustering methods and learning of a co occurrence matrix between clusters of distinct variables provide a computationally relaxed way of otherwise intractable joint probability estimations. Utilizing this approach, we report explicitly different scoring results for aggressive and nonaggressive labelled driving experiences. Besides, we provide a rigorous analysis of clustering schemes applied on trajectory, traffic flow and driving style data.

26th IEEE Signal Processing and Communications Applications Conference (2018 : Izmir; Turkey), Developing personalized cancer treatment procedures requires a prior knowledge on the effects of different drugs on cancer cell lines. While obtaining this information in vitro is a tedious task, the emergence of numerous large-scale datasets facilitates the usage of machine learning algorithms for this purpose. Conventional methods make an effort to reveal the mapping function between a cell line's identifying features called gene expressions and a certain drug's effect on it. In this work, we move away from this philosophy and represent cell lines as images in which inter-feature relations are preserved. Once these images are obtained, the regression problem is solved with the help of a convolutional neural network, a neural network architecture proven to work well with image inputs. A benchmarking with the other models in the literature exhibits the fruitfulness of our novel strategy. © 2018 IEEE., Aselsan,et al.,Huawei,IEEE Signal Processing Society,IEEE Turkey Section,Netas

Chemotherapeutic response of cancer cells to a given compound is one of the most fundamental information one requires to design anti-cancer drugs. Recently, considerable amount of drug-induced gene expression data has become publicly available, in addition to cytotoxicity databases. These large sets of data provided an opportunity to apply machine learning methods to predict drug activity. However, due to the complexity of cancer drug mechanisms, none of the existing methods is perfect. In this paper, we propose a novel ensemble learning method to predict drug response. In addition, we attempt to use the drug screen data together with two novel signatures produced from the drug-induced gene expression profiles of cancer cell lines. Finally, we evaluate predictions by in vitro experiments in addition to the tests on data sets. The predictions of the methods, the signatures and the software are available from http://mtan.etu.edu.tr/drug-response-prediction/. © 2018 Elsevier Inc.