Search

Eunicella cavolini (Koch, 1887) generally recognized as yellow gorgonian, is a species of colonial soft coral in the family Gorgoniidae. Although it is common in the Mediterranean basin, the evaluation of biological potential of E. cavolini from the Adriatic Sea has not been performed yet. Taking this into account, as a part of our Center of excellence for Marine Bioprospecting of Adriatic Sea BioProCro, we have studied the chemical composition and biological activities of this gorgonian. [1] In this contribution, we will discuss the results of screenings of the antiproliferative, antibacterial, antioxidant, and anti-inflammatory activities of organic extracts and corresponding semipurified fractions. Additionally, we will show data concerning antiproliferative activity against four cancer cell lines and one non-transformed cell line, as well as its anti-inflammatory potential concerning the successfully isolated pregnane type steroid by employing dereplication methods. In final, we will present the synthesis of granulatamide B [2] isolated from the coral species of the Eunicella genus and its 12 derivatives by employing Fürstner’s Fe-catalyzed C–C coupling reaction. [3] The obtained small compound library of these N-acyl tryptamines differing in the number of C-atoms, saturation degree and conjugation of double bonds, were evaluated towards the formation and development of zebrafish Danio rerio embryos. [4]

Spirulina sp. is a widely cultivated cyanobacteria which contains a high amount of bioactive compounds with health-promoting effects and remarkable antioxidant activity due to the abundant presence of polyphenols and carotenoids. The effect of different drying temperatures (room temperature, 40, 50, 60, 80, and 100°C) on the phenolic profile and biological activity of Spirulina sp. was investigated. Samples were dried at various temperatures, extracted and then fractioned. Antioxidant activity of methanolic (F3) and dichloromethane (F4) fractions was evaluated using ABTS, FRAP, ORAC and DPPH assays. To determine the toxicity potential of prepared fractions, zebrafish embryotoxicity test was performed [2]. Phenolic compounds in Spirulina sp. samples were identified using HPLC, while LC/ MS- Q-TOF screening was used to predict the most probable bioactivities, chemical classes and drug- likeness properties of examined compounds. According to antioxidative assays, F3 fraction had higher ABTS and FRAP activity, while F4 fraction had higher DPPH and ORAC activity. The F4 fraction dried at 50°C showed the highest antioxidant activity using DPPH (687.06 +/- 11.41 mg AAE/g fraction), while ABTS and FRAP assays showed high activity for F3 fractions dried at 25, 30 and 100°C. There was no negative impact on the survival of zebrafish embryos and larvae observed at concentrations lower than 0.02 mg/mL. The most abundant polyphenols were epicatechin, vanillic acid, caffeic acid, diosmin, thymol, apigenin, while the presence of 3 alkanolamines and 25 lipid-like molecules was revealed by QTOF analysis. High antioxidant activity was observed at lower temperatures can be related to the presence of carotenoids and chlorophyll. Antioxidant potential of sample dried at 100°C can be attributed to pigment derivatives. The optimal temperature for preservation of polyphenols was 50 °C. A strong correlation between structural class of detected compounds and highly predicted activities (cardiovascular, antiinflammatory, anticancer) has been observed. The results indicate that drying at a lower temperature retains phenols and other components and maintains high antioxidant activity. From the obtained results, it can be concluded that lower drying temperatures better influence the content of phenolic compounds and biological activity of Spirulina sp., when compared to higher temperatures.

Natural products (NP) are enourmous, still largely unexplored resource of molecules with potentially therapeutic properties. More than 90 % of Earth species live in the marine environments, providing vast structural diversity and bioactivity potential of NPs associated with marine organisms. By investigating these organisms new compounds with applications in agriculture, medicine and various biotechnological processes may be discovered. While plants and terrestrial microorganisms have been more explored for new chemical entities so far, the increasing number of new drug candidates and drugs from marine organisms is showing that this resource will be of major interest in the future. Computational approaches offer a scalable, early approach for evaluating of bioactivity potential of large chemical space of natural products. We provide here a large scale, qualitative breakdown of properties of small molecules (less than 800 Da) stemming from different organisms and compare them against current drugs and nutraceuticals. We used COCONUT [1] (natural products) and DrugBank [2] (drugs and nutraceuticals) as sources of compounds, PASS software [3] to provide bioactivity estimates and Classyfire WEB service [4] to structuraly classify compounds. Table 1. depicts major structure-bioactivity profiles of compounds from different origins, with their rough druglikenes assessment. Using more rigorous computational (AI, machine learning based) approaches we are working to provide deeper structural-biaoctivity assessment of the compounds originating in marine organisms.

Machine learning (ML) approaches are extensively used to analyse chemical (synthetical, analytical, physicochemical) and biological (in vitro and in vivo) data and build models for discovery and design of novel small molecular weight molecules with optimized target activities. The basic ML requirement is to have a large and reliable enough dataset which is used for training and testing a predictive model. In the presentation, application of ML approaches on publicly available data sets for biologically active compounds such as drugs and agrochemicals will be demonstrated.

Spirulina sp. is a widely cultivated cyanobacteria which contains a high amount of bioactive compounds with health-promoting effects and remarkable antioxidant activity due to the abundant presence of polyphenols and carotenoids. The effect of different drying temperatures (room temperature, 40, 50, 60, 80, and 100°C) on the phenolic profile and biological activity of Spirulina sp. was investigated. Samples were dried at various temperatures, extracted and then fractioned. Antioxidant activity of methanolic (F3) and dichloromethane (F4) fractions was evaluated using ABTS, FRAP, ORAC and DPPH assays. To determine the toxicity potential of prepared fractions, zebrafish embryotoxicity test was performed. Phenolic compounds in Spirulina sp. samples were identified using HPLC, while LC/ MS-Q-TOF screening was used to predict the most probable bioactivities, chemical classes and drug-likeness properties of examined compounds. According to antioxidative assays, F3 fraction had higher ABTS and FRAP activity, while F4 fraction had higher DPPH and ORAC activity. The F4 fraction dried at 50°C showed the highest antioxidant activity using DPPH (687.06 +/- 11.41 mg AAE/g fraction), while ABTS and FRAP assays showed high activity for F3 fractions dried at 25, 30 and 100°C. There was no negative impact on the survival of zebrafish embryos and larvae observed at concentrations lower than 0.02 mg/mL. The most abundant polyphenols were epicatechin, vanillic acid, caffeic acid, diosmin, thymol, apigenin, while the presence of 3 alkanolamines and 25 lipid-like molecules was revealed by QTOF analysis. High antioxidant activity observed at lower temperatures can be related to the presence of carotenoids and chlorophyll. Antioxidant potential of sample dried at 100°C can be attributed to pigment derivatives. The optimal temperature for preservation of polyphenols was 50°C. A strong correlation between structural class of detected compounds and highly predicted activities (cardiovascular, anti-inflammatory, anticancer) has been observed. The results indicate that drying at a lower temperature retains phenols and other components and maintains high antioxidant activity. It can also be concluded that lower drying temperatures better influence the content of phenolic compounds and biological activity of Spirulina sp., when compared to higher temperatures.

Bioactive potential of brown alga Padina pavonica from the Adriatic Sea was described according to the tests conducted.

Natural products are a great source of novel molecules that can have beneficial effects on human health, not only as medicines, but also as nutrients and cosmetic ingredients. We have investigated compounds from the largest natural product database COCONUT, using data science and machine learning, and compared them with approved drugs from the ChEMBL database [1, 2]. Among the various machine learning techniques, reducing the dimensionality of the multidimensional space for the purpose of visualization enables better understanding of large data sets. In the presented case study, we used advanced exploration and visualization tools to analyze structure-activity relationships (SAR) and reveal preliminary activities for subsets of natural products. References: [1] M. Sorokina, P. Merseburger, K. Rajan, M. A. Yirik, C. Steinbeck, J. Cheminform. 13 (2021) 2. [2] A. Gaulton, L. J. Bellis, A. P. Bento, J. Chambers, M. Davies, A. Hersey, Y. Light, S. McGlinchey, D. Michalovich, B. Al-Lazikani, J. P. Overington, Nucleic Acids Res. 40 (2012) D1100–D1107.

Spirulina sp. is a widely cultivated microalgae that contains numerous bioactive compounds with health-promoting properties. Phenolic profile and biological activity were investigated in regard to different drying temperatures (25, 40, 50, 60, 80, 100°C). Antioxidant activity (ABTS, FRAP, ORAC, DPPH) of methanolic (F3) and dichloromethane (F4) fractions was evaluated. To assess the toxicity of the fractions, zebrafishembryos were used. Phenolic compounds were identified using HPLC, while prediction of most probable bioactivities, chemical classes, and drug-likeness properties were made using LC/MS-Q-TOF. The F4 (50°C) showed the highest altioxidant activity using DPPH, while ABTS and FRAP showed high activity for F3 (25, 30 and 100°C). No negative impact on the survival of zebrafish was observed at concentrations lower than 0.02 mg/mL. Most abundant polyphenols were epicatechin, vanillic and caffeic acid, while the presence of alkanolamines and lipid-like molecules was revealed by LC/MS-Q-TOF. High antioxidant activity was observed at lower temperatures, due to the presence of carotenoids and chlorophylls, while for the sample dried at 100°C this can be attributed to pigment derivatives. Optimal temperature for preservation of polyphenols was 50°C. A strong correlation between structural class of detected compounds and highly predicted activities (cardiovascular, anti- inflammatory, anticancer) has been observed.

There is enormous public interest in the EU, but also the interest of the pharmaceutical industry in the possibility of finding significant factors that would make it possible to distinguish between approved drugs and drugs that have been withdrawn after several years of use. The EU responded to this interest by funding a large AI4EU project, which launched calls for proposals to develop various artificial intelligence (AI)-based computing solutions.[1] Among them was the task to develop a predictive AI solution based on the deep neural network method to distinguish drugs from withdrawn drugs based on their structure.[2] In this presentation, the process of defining the problem and creating a dataset will be elaborated, as well as related issues such as the different status of the same drug in different countries or regions, or the problem of cleaning datasets, as many drugs are not pure compounds but complexes. We have calculated comprehensive sets of molecular structure descriptors for the final sets of approved and withdrawn drugs. The analysis of the importance/usefulness of each structure-based molecular descriptor (including those related to ADMETox properties of drugs) in discriminating between approved and withdrawn drugs will be presented. As expected, the classification accuracy of the multivariate models based on descriptors is not high. Nevertheless, it is comparable to (or sometimes better than) more complex models and significantly higher than the accuracy achieved by random guessing (random model). REFERENCES [1] The AI4EU Call for Solutions Seeking Companies to Solve AI Challenges, https://internationaldataspaces.org/the-ai4eu-call-for-solutions-seeking-companies-to-solve-aichallenges/ (assessed on April 15, 2022) [2] The AI-on-demand platform, https://www.ai4europe.eu/ and https://www.ai4media.eu/projectoverview/ (assessed on April 15, 2022)

Despite decades of intensive search for compounds that modulate the activity of particular protein targets, a large proportion of the human kinome remains as yet undrugged. Effective approaches are therefore required to map the massive space of unexplored compound–kinase interactions for novel and potent activities. Here, we carry out a crowdsourced benchmarking of predictive algorithms for kinase inhibitor potencies across multiple kinase families tested on unpublished bioactivity data. We find the top-performing predictions are based on various models, including kernel learning, gradient boosting and deep learning, and their ensemble leads to a predictive accuracy exceeding that of single-dose kinase activity assays. We design experiments based on the model predictions and identify unexpected activities even for under-studied kinases, thereby accelerating experimental mapping efforts. The open-source prediction algorithms together with the bioactivities between 95 compounds and 295 kinases provide a resource for benchmarking prediction algorithms and for extending the druggable kinome.

Despite decades of intensive search for compounds that modulate the activity of particular protein targets, a large proportion of the human kinome remains as yet undrugged. Effective approaches are therefore required to map the massive space of unexplored compound-kinase interactions for novel and potent activities. Here, we carry out a crowdsourced benchmarking of predictive algorithms for kinase inhibitor potencies across multiple kinase families tested on unpublished bioactivity data. We find the top-performing predictions are based on various models, including kernel learning, gradient boosting and deep learning, and their ensemble leads to a predictive accuracy exceeding that of single-dose kinase activity assays. We design experiments based on the model predictions and identify unexpected activities even for under-studied kinases, thereby accelerating experimental mapping efforts. The open-source prediction algorithms together with the bioactivities between 95 compounds and 295 kinases provide a resource for benchmarking prediction algorithms and for extending the druggable kinome. The IDG-DREAM Challenge carried out crowdsourced benchmarking of predictive algorithms for kinase inhibitor activities on unpublished data. This study provides a resource to compare emerging algorithms and prioritize new kinase activities to accelerate drug discovery and repurposing efforts.

Previous approaches implemented on drug-kinase binding affinity benchmark datasets show poor performance on rigorous test scenarios with unseen small compounds or protein kinase targets - thus limiting their real-world application. We represent methodology which relies on an ensemble approach, XGBoost trained on fingerprint based and protein kinase domain sequence-based similarity features - and graph convolutional networks (GCN) as more advanced representation learning predictive methodology. To assess the uncertainty surrounding model predictions, we defined a structure-based applicability domain with a focus on density of compound space in the training set.

Machine learning (ML) approaches are widely used to analyse and model various types of problems, including those related to food and climate changes. The basic premise is to have a large and reliable dataset which is used for training and testing predictive model. In the lecture, application of ML in the food field will be presented through the results of ML study performed for herbicides. Very extensive and wide use of herbicides leads to increasing of (i) weed resistance and (ii) human health issues. We have applied ML approaches (Random Forest, clustering) for developing predictive models for mode of action and weed selectivity of herbicides grouped in HRAC/WSSA list (https://github.com/mlkr- rbi/Herbicide-Classification). Our ML study points to shortcomings of usage rotation strategy which is based exclusively on HRAC/WSSA classification for reducing weed selectivity. In addition, phytotoxic natural molecules are identified as different in chemical as well as biological space in comparison to synthetic herbicides and thus their usage and development may provide a complementary way to slow down weed resistance.

The cheapest, fastest and efficient way to find active compounds for biological targets, novel biological activities for compounds “in shelves” or to reveal targets for discovered natural products, is to test compounds virtually by using proper in silico approaches and/or models at the beginning. The predictive machine learning (ML) model developed for specific (bio)activity can be applied for prioritizing assays for in vitro testing of compounds and their (re)purposing. The two ML models which may shed light on potential biological activities for compounds, will be presented. They have been developed by using publicly available datasets and are incorporated within virtual modular platforms for detecting potential phytotoxic molecules and kinase inhibitors.

In data streams new classes can appear over time due to changes in the data statistical distribution. Consequently, models can become outdated, which requires the use of incremental learning algorithms capable of detecting and learning the changes over time. However, when a single classification model is used for novelty detection, there is a risk that its bias may not be suitable for new data distributions. A solution could be the combination of several models into an ensemble. Besides, because models can only be updated when labeled data arrives, we propose two unsupervised ensemble approaches: one combining clustering partitions using the same clustering technique; and other using different clustering techniques. We compare the performance of the proposed methods with well known novelty detection algorithms. The methods were tested on datasets commonly used in the novelty detection literature. The experimental results show that proposed ensembles have competitive performance for novelty detection in data streams.

This presentation talks about a novel benchmark for disentangled concept learning, called “BricksND”. The benchmark consists of synthetically generated brick scenes in 1D, 2D and 3D spaces. This benchmark is designed to address the challenges of machine learning, especially disentanglement representations and concept learning.

This paper proposes a new method to measure the relative importance of features in Artificial Neural Networks (ANN) models. Its underlying principle assumes that the more important a feature is, the more the weights, connected to the respective input neuron, will change during the training of the model. To capture this behavior, a running variance of every weight connected to the input layer is measured during training. For that, an adaptation of Welford’s online algorithm for computing the online variance is proposed. When the training is finished, for each input, the variances of the weights are combined with the final weights to obtain the measure of relative importance for each feature. This method was tested with shallow and deep neural network architectures on several well-known classification and regression problems. The results obtained confirm that this approach is making meaningful measurements. Moreover, results showed that the importance scores are highly correlated with the variable importance method from Random Forests (RF).

Additional file 1 Additional figures and tables referenced in the article.

Additional file 2 Review History.

Chatbots have been used in business contexts as a new way of communicating with customers. They use natural language to interact with the customers, whether while offering products and services, or in the support of a specific task. In this context, an important and challenging task is to assess the effectiveness of the machine-to-human interaction, according to business’ goals. Although several analytic tools have been proposed to analyze the user interactions with chatbot systems, to the best of our knowledge they do not consider user-defined criteria, focusing on metrics of engagement and retention of the system as a whole. For this reason, we propose the KnowBots tool, which can be used to discover relevant patterns in the dialogues of chatbots, by considering specific business goals. Given the non-trivial structure of dialogues and the possibly large number of conversational records, we combined sequential pattern mining and subgroup discovery techniques to identify patterns of usage. Moreover, a friendly user-interface was developed to present the results and to allow their detailed analysis. Thus, it may serve as an alternative decision support tool for business or any entity that makes use of this type of interactions with their clients.

Learning from multiple time-series over an unbounded time-frame has received less attention despite the key applications (such as video analysis, home-assisted) generating this data. Inspired by never-ending approaches, this paper presents an algorithm to continuously learn from multiple high-dimensional un-regulated time-series, in a framework based on ensembles which with respect to drift level develops over time in order to reflect the latest concepts. Here, we explicitly look into video surveillance problem as one of the main sources of high-dimensional data in daily life and extensive experiments are conducted on multiple datasets, that demonstrate the advantages of the proposed framework in terms of accuracy and complexity over several baseline approaches.

Rising popularity of social networks allows us to investigate dynamics of social interactions on a scale that would be unimaginable just a couple of decades ago. One particular kind of social interaction is information propagation between users of online social network, which we can divide into that which is endogenous - propagating between users (peers), and that which is exogenous - originating from external information sources like online mass media. In fact, large information cascades in social networks are often driven by exogenous events such as political unrest and natural disasters. In this work we propose a model of social influence which is conceptually similar to the unified model of social influence [2] which was shown to be generalization of many popular influence models, including complex contagion model, independent cascade model and generalized threshold model. At each time step all activated users attempt to activate all of their peers in network with certain probability. Also, there is an external influence which acts equally on all users at a given time step, although it may change over time. Activation times of each individual user are known, and together they form an activation cascade. Our problem is then the following: Given particular form of peer influence along with activation cascade and a network of users, infer parameters of peer and external influence while assuming that parameters of peer influence stay constant throughout the 39 Machine Learning and Data Mining Thursday, 10:40-11:05 period while external influence may change in time. Similar attempts exist in lit- erature, including peer and authority model [1] which, however, requires explicit modeling of authorities responsible for external influence. Also, we try to decou- ple the endogenous and exogenous influence just by using a statistical properties of activation cascades on network without inferring the actual exposure curves. We assume that information propagation in online social networks is mediated by two influences: (i) peer influence which is endogenous to the network, and (ii) external influence which is exogenous to the network. As these two influences are often confounded we are interested in inferring them separately from our observed data - friendship network of users and their activation times, given some functional forms for peer and external influence. We will do this by optimizing a log-likelihood function which gives us a probabilistic description of our problem. The log-likelihood is evaluated in moving time windows and it separates users into those that did or did not activate at a particular time window. The formulation is flexible enuogh to allow for parameters of peer and external influence to vary for each user and each time window. However, we choose to constrain the inference problem in order to better estimate the maximum likelihood solution. First, we choose functional forms for peer and external influence. For simplicity, we will model external influence as a constant probability of activation which acts on all yet nonactivated users at a specific time. For the peer influence models we will choose several models typically encountered in literature: (i) Susceptible- infected model, (ii) Exponential decay model, and (iii) Logistic threshold model. To constrain optimization problem even more we introduce some simplifying assumptions: (i) all users have equal parameters of peer influence, (ii) external influence acts equally to all users at a specific time window, and (iii) parameters of peer influence do not change over time. Because our model gives us probabilities for peer activation as well as for external activation for each user, we can use this information to estimate activation type for each of the users. We can also define a single measure of external responsibility which quantifies to what extent is an activation of each user due to the external influence. To empirically validate our methodology we use two Facebook applications as online political polls one week prior to the upcoming referendum (referen- dum2013.hr dataset, over ten thousand users) and parliamentary elections (sa- bor2015.hr dataset, over six thousand users) in Croatia. To our knowledge these are the largest Facebook datasets obtained in this way as users had to give an explicit consent to participate in the study. We consider an act of registration as an activation event, and infer the magnitude of endogenous and exogenous influence for each Facebook user in the activation cascade. We estimate magni- tude of endogenous and exogenous influence by performing maximum likelihood estimation using various peer influence models. Length of the time window used 40 Thursday, 10:40-11:05 Machine Learning and Data Mining for evaluating log-likelihood is 30 minutes. We compare our method with a base- line method that classifies an activation as external if the user had no previously activated friends. This is a conservative measure that tends to underestimate the true external influence because after majority of users is activated it is extremely unlikely for newly activated users to have no previously activated friends, even if they really are activated by an external influence. We estimate a separate exoge- nous influence parameter at each time window and a single set of parameters for endogenous influence. The results for the exponential decay model indicate that the half-decay of peer influence is approximately 5 hours. This is expected, as it is reasonable to assume that engagement of users in sharing specific information is halved in a timespan of several hours, and probably decays to zero after a day or two. We also expect for endogenous influence to be smooth and slowly varying, while exogenous influence has sharp spikes of activity which are aligned with the times when major online news providers reported on our application.

In this work, we address the task of phenotypic traits prediction using methods for semi- supervised learning. More specifically, we propose to use supervised and semi-supervised classification trees as well as supervised and semi-supervised random forests of classification trees. We consider 114 datasets for different phenotypic traits referring to 997 microbial species. These datasets present a challenge for the existing machine learning methods: they are not labelled/annotated entirely and their distribution is typically imbalanced. We investigate whether approaching the task of phenotype prediction as a semi- supervised learning task can yield improved predictive performance. The result suggest that the semi-supervised methodology considered here is helpful for phenotype prediction for which the amount of labeled data ranges from 20 to 40%. Furthermore, the semi-supervised classification trees exhibit good predictive performance for datasets where the presence of a given trait is not extremely imbalanced (i.e., less than 6%).

In this work we compare different information fusion approaches in the context of large-scale multi-label classification problems, typical today in bio-domains: early fusion, late fusion and hybrid fusion approach. The experiments are performed on two novel large-scale classification datasets for gene function prediction and prokaryotic phenotype prediction. Both datasets are based on descriptors coming from a number of different representations of biological entities. The results reveal that the fusion approaches exploiting complementarity are best suited for difficult annotation problems featured in complex datasets from bio-domains for which individual classifiers perform well only locally.

Research of eukaryotic cell motility had been based on the persistent random walk model before careful analysis of locomotion paths revealed additional features of the cell migration that lead to modifications and extensions of this standard model. The initial aim of our research was to characterize motility of cortexillin double-null cells of D. discoideum within the framework of a generalized Langevin model extended with a memory term. In the course of the analysis of wild-type and mutant cells, we noticed that cells belonging to a single strain are highly heterogeneous with respect to their mode of migration. This prompted us to perform unsupervised classification of all measured cell trajectories according to a set of attributes that quantitatively describe cell locomotion. The motile behavior of cells from two different strains, wild-type (AX2) and the strain lacking cortexillins I and II (CI-/CII-), was monitored by dark-field microscopy. For both strains, we measured the trajectories of randomly migrating vegetative and aggregation-competent cells. Migration of individual cells was characterized by a set of parameters derived from the speed distribution, the velocity autocorrelogram and the velocity periodogram. We determined that each of the four analyzed populations, i.e. AX2 and CI-/CII- cells in vegetative and aggregation-competent stages, can be divided into several homogenous subpopulations specified by the average speed and the low-frequency segment of the periodogram. CI-/CII- cells showed a larger dispersion of the average speed values relative to AX2 in both phases of their life cycle, whereas the aggregation-competent CI-/CII- cells displayed a loss of low-frequency oscillations. Finally, the motility data for each of the identified subpopulations were fitted to the generalized Langevin equation.

Cell motility is of the crucial importance for many physiological and pathological processes in multicellular organisms. The amoeboid protozoan Dictyostelium discoideum is a standard model organism used in cell motility research as well-suited for studying components of the actin cytoskeleton that are functionally conserved throughout the eukaryotic domain. Initial research of cell motility was based on the Brownian motion model describing random movement of small particles caused by thermal fluctuations. Over time, more precise measurements revealed intrinsic patterns in locomotion paths of eukaryotic cells, leading to modifications and extensions of the aforementioned model. The aim of this research is to characterize cell motility of actin cytoskeleton mutants of D. discoideum within the framework of a generalized Langevin model extended with a memory term. Furthermore, cell populations belonging to a single strain will be subdivided into subgroups based on motility characteristics of individual cells. D. discoideum cells grow as separate individual amoeba while in favorable condition, while upon starvation they drastically change their physiological state and interact with each other to form a multicellular slug. We monitored the motile behavior of two different strains of D. discoideum, wild-type strain (AX2), and the strain lacking cortexillins (CI-/CII-), in different stages of their life cycle: vegetative and aggregation-competent. Cortexillins are actin-bundling proteins and have an important role in the actin cytoskeleton organization. Cell movement was monitored by dark-field microscopy at low magnification. Images acquired at regular time intervals were stored and subsequently analyzed with the ParticleTracker plug-in for ImageJ. Results of the image processing are the cell tracks presented as time sequences of cell centroid coordinates. We identified different cell subpopulations by cluster analysis in RapidMiner environment using kmeans clustering algorithm with 14 parameters obtained by trajectory analysis. Parameters characterizing speed distribution were: the mean, the standard deviation, and the skewness and kurtosis of the distribution. Autocorrelograms of speed and velocity were characterized by their variance and the time lag of the first drop of an autocorrelogram below the threshold of significance. Periodograms of speed and velocity were characterized by their contributions in the following intervals: under a minute, between one and six minutes, and between six and 15 minutes. After the feature selection step, further clustering was performed using the two chosen parameters: the average speed and the sum of the periodogram values between time lags of 6 and 15 minutes. The later parameter represents a cell's oscillatory behavior in this range of time intervals. The starved cells of both AX2 and CI-/CII- strains were able to move faster than their vegetative counterparts, with the CI-/CII- cells showing a less strict control of speed in both phases of life cycle. CI-/CII- starved cells also display a loss of oscillatory behavior in the aggregation-competent phase.

Automated annotation of protein function is challenging. As the number of sequenced genomes rapidly grows, the overwhelming majority of protein products can only be annotated computationally. If computational predictions are to be relied upon, it is crucial that the accuracy of these methods be high. Here we report the results from the first large-scale community-based critical assessment of protein function annotation (CAFA) experiment. Fifty-four methods representing the state of the art for protein function prediction were evaluated on a target set of 866 proteins from 11 organisms. Two findings stand out: (i) today's best protein function prediction algorithms substantially outperform widely used first-generation methods, with large gains on all types of targets; and (ii) although the top methods perform well enough to guide experiments, there is considerable need for improvement of currently available tools.

The paper presents results of the analysis of countries that experienced banking crises in the period 1976-2011. The analysed data are World Bank indicators in the periods before the onset of the crises. The work is done as part of the EU FP7 project “Forecasting Financial Crises”. For the analysis we have used the subgroup discovery methodology, which is especially useful when different subtypes of the main concept are anticipated and when different, possibly even contradictory properties may characterize these subtypes. We describe the data preparation process and the Data Mining Server application for subgroup discovery induction. The applied methodology may also be useful for other applications in business and marketing domains. Our main result is the classification of five subsets of countries with banking crises. Three of them may be recognized as financially driven types of banking crises, while significant socio- economic problems are characteristic for the remaining two subgroups.

In this study, we compare the performance of semi- supervised and supervised machine learning methods applied to various problems of modeling quantitative Structure Activity Relationship (QSAR) in sets of chemical compounds. Semi- supervised learning utilizes unlabeled data in addition to labeled data with the goal of building better predictive models than can be learned by using labeled data alone. Typically, labeled QSAR datasets contain tens to hundreds of compounds, while unlabeled data are easily accessible via public databases containing thousands of chemical compounds: this makes QSAR modeling an attractive domain for the application of semi-supervised learning. We tested four different semi-supervised learning algorithms on three different datasets and compared them to five commonly used supervised learning algorithms. While adding unlabeled data does help for certain pairings of dataset and method, semi-supervised learning is not clearly superior to supervised learning across the QSAR classification problems addressed by this study.

Cell motility is of crucial importance for many physiological and pathological processes in multicellular organisms. The amoeboid protozoan Dictyostelium discoideum is a standard model organism used in cell motility research as well-suited for studying components of the actin cytoskeleton that are functionally conserved throughout the eukaryotic domain. Initial research of cell motility was based on the Brownian motion model describing random movement of small particles caused by thermic fluctuations. Over time, more precise measurements revealed intrinsic patterns in locomotion paths of eukaryotic cells, leading to modifications and extensions of the aforementioned model. The aim of this research is to characterize cell motility of different actin cytoskeleton mutants of D. discoideum within the framework of a generalized Langevin model extended with a memory term. Furthermore, we will subdivide cell populations belonging to single strains into subgroups based on motility characteristics of individual cells.

Postoji rastući interes za korištenje manganoporfirina kao terapeutika za bolesti povezane s oksidativnim stresom, pošto je uklanjanje reaktivnih kisikovih i dušikovih vrsta bazirano na redoks reakcijama. Utvrđena je povezanost strukture i aktivnosti za redukcijski potencijal središnjeg manganovog iona i katalitičke konstante brzine dismutacije O2•- manganoporfirina, što je dovelo do sinteze najsnažnijih funkcionalnih mimetika SOD. Promjena SOD aktivnosti metaloporfirina je postignuta podešavanjem redukcijskog potencijala središnjeg kovinskog iona pomoću alkilacije tetrapiridilporfirinskog prstena. Među manganoporfirinima čija je lipofilnost i katalitička konstanta brzine dismutacije O2•- određena, manganov orto-tetrakis(N-n-butilpiridin-2-il)porfirin pokazuje antioksidativnu učinkovitost usporedivu s najefikasnijim analozima. Provedene su spektrofotometrijske pH-titracije (1

Knowing the execution time of a computational model, especially when dealing with large data, is crucial in deciding whether the solution of the problem is attainable in acceptable time. In the case of data mining processes, typically both the time needed for model learning and model application could be of importance. We developed a meta-mining framework for execution time estimation of data mining algorithm built in RapidMiner. Operator execution time estimation is treated as a machine learning problem for which prediction models are built using execution times obtained by running algorithms on a set of predetermined datasets. With appropriate refitting this experimental methodology is applicable to any data mining environment. We present overall framework with modelling results for a subset of RapidMiner operators, and compare non-parametric distance measures based predictions with polynomial function fitting. Finally, integration of these models in the form of standalone RapidMiner extension is demonstrated and issues related to reliability, scalability and applicability for the overall workflow execution time modelling are discussed.

Recommender systems are ubiquitous in today's information overoaded world. They help users to find and select products from a huge number available in various sources. RapidMiner can be used to construct various information filtering workflows using various data mining techniques. However, it does not explicitly support typical recommendation tasks. In this paper we present a RapidMiner Recommender Extension, developed in order to embed some of the state-of-the-art recommendation techniques into RapidMiner. We present the functionality of the extension, along with examples of diverse recommender system implementations. Integration of the extension with RapidAnalytics, which allows seamless construction of production level recommender systems, is demonstrated.

Plasma biochemical profile was examined in cage cultured shade fish Argirosomus regius in order to evaluate physiological status and provide detailed picture of its seasonal changes. Biochemical procedures were employed to study trygliceride (TRIG), cholesterol (CHOL), total protein (TP), glucose (GLU), urea (BUN), inorganic phosphate (PHOS), calcium (Ca), creatine kinase (CK), lactate dehydrogenase (LDH), ammonia (NH3), magnesium (Mg), aspartate aminotransferase (AST), amylase (AMYL), alkaline phosphatase (ALKP), creatinine (CREA), gamma glutamyltransferase (GGT) and total bilirubin (TBIL). The results were evaluated through different statistical analyses. Conventional statistical methods do not provide solutions that meet the goals of the research, so machine learning methodology was used to construct and evaluate classifiers as well as to assess parameter (plasma metabolites and enzymes) importances with respect to the target variable (season). Two machine learning methods, decision tree algorithm and PARF - a new implementation of Random Forest algorithm were applied to the problem of seasonal fish population modelling based on blood samples data collected. These two methods were used in order to explain which combination of parameters best demonstrates the differences between examined seasonal groups through construction and evaluation of classifiers and using PARF approach to the assess the variable importances with respect to the target variable, i.e. season. It is evident that about half of the parameters (UREA, ALKP, NH3, CHOL, Ca, PHOS, GLU and CREA) have significant importance (i.e. exhibit significant differences between seasons) for seasonal discrimination models. From our classifier learning experiments it is evident that using only 2 (UREA, ALKP), out of 8 high importance ranking parameters (UREA, ALKP, NH3, CHOL, Ca, PHOS, GLU, CREA) is generaly enough to obtain high accuracy classsifiers, which suggests rather high correlation between parameters on a seasonal scale. We found that AST, Ca, TP, LDH, BUN, CREA, AMYL, CK, NH3 were increasing with elevation of water temperature (measured in March, May and July) while CHOL and ALP were decreasing at the same time. AST, TP, GLU are sensitive indicators ; a rise of their activity can be recognized during metabolic modification related to nutritive imbalance (occuring in the colder season) resulting in physiological dysfunction or cellular integrity. ALP is a splitting enzyme of terminal phosphatase groups from organic esters found mainly in the hepatocites. High values are indicative of liver, specifically biliary impairment. CREA and BUN levels can indicate other problems besides those intrinsic to the kidney such as prerenal problem. CHOL could be nutritionally related and it is possible that food intake are higher in the summer compared to the winter. Ca levels were lower in the period of cold water this might be due to perturbations in the hormones during that control Ca. Decrease ALP levels associated with conditions linked to the biliary tract. According to One Way Analysis of Variance (Sigma Statistical Software ver. 2.01), TRIG and LDH were not significantly different between seasons, while CHOL, BUN, Ca, NH3, AST, ALKP and CREA demonstrated significance in all sampling points. Three parameters (TP, PHOS, and AMYL) were significant between March-July and May-July ; while GLU, CK and GLU, Mg showed significance for period March-May and March-July, respectively.

Introduction: Hematological and biochemical tests are a well-established tool for health status monitoring in organic farming. Interpretation of test results critically relies on accurate determination of reference intervals. Due to locally variable conditions specific to organic farming, such as breed, nutrition, climate and season, separate sets of reference intervals are often needed for specific situations. In this work, a machine learning approach was used to identify when separate reference intervals are justified. Novel methods based on robust statistics and bootstrap resampling were used to tackle the problem of reference interval calculation from small samples. Materials and methods: Nine flocks of approximately 30 female sheep each, of various (mostly Croatian indigenous) breeds, organically raised at different locations, were analyzed in different seasons for their key hematological and blood biochemical parameters. Machine learning algorithm RELIEF was used to assess the difference between groups and identify key parameters for their differentiation. Reference intervals were calculated using robust statistics and bootstrap resampling. Results: Location, a parameter closely linked to nutrition and environmental conditions, was the most important for differentiation between groups, even more than breed. No single hematological or biochemical parameter could be identified as critical for discrimination between groups. High values of the kappa statistic justified calculation of separate reference intervals for each group (flock), which was successfully done with small reference populations using robust statistics. Conclusions: Separate reference intervals are justified for narrow sets of conditions in organic farming, and can be successfully estimated from small samples using appropriate robust methods.

In this paper we demonstrate how to construct some typical recommender systems in RapidMiner. Our workflow template library currently includes content-based, LSI content-based, user- and item- based collaborative filtering with SVD dimensionality reduction recommender systems. The library is publicly available via myExperiment collaborative repository. We also demonstrate the use of templates on a realistic recommendation problem within digital multimedia repository.

Live lobsters Palinurus elephas (Fabricius, 1787) were collected from a number of sites around the island of Vis, and transported to laboratory exposed to air for considerable time. The aim of the study was to examine the effect of emersion on the plasma biochemical profile which is a useful indicator of physiological status. Lobsters had average body weight of 517, 09 g and carapace length of 113, 20 mm. Haemolymph was withdrawn from ventral abdominal artery. Biochemical procedures were carried out to study metabolites (trygliceride, TRIG ; cholesterol, CHOL ; total proteins, TP ; glucose, GLU ; urea, BUN ; creatinin, CREA) and several enzymes (alkaline phosphatase, ALKP ; aspartate aminotransferase, AST ; gamma glutamyltransferase, GGT ; lactate dehydrogenase, LDH). In addition to classic statistical analyses, six machine-learning methods utilizing ten times ten-fold cross-validation evaluation were used in order to determine the best classification model for the blood biochemistry data. Blood GLU, BUN, ALKP, TP significantly increased while CREA was significantly lower. This approach clarifies the role and importance of physiological processes which show diversity of functional characteristics of various biochemical parameters in case of long term exposure to air.

Usage of Bayesian networks in medical decision support system is in general case twofold: (1) for obtaining probabilities of occurrence of medical events (i.e. possible diagnosis) and (2) for obtaining information gain of actions that can be taken (i.e. diagnostic tests). On the other hand, typical role of ontology is to provide a framework for definition of medical concepts, their structure and relations among them. In medical practice diagnostic tests are commonly comprised of number of measurements or sub-tests – a structure which is straightforwardly described by ontological language. In this paper we are analyzing the information gain of such structured medical diagnostic tests. The purpose of this analysis is to allow finding (1) which structured medical diagnostic test is at the given point the most informative one and (2) which elementary measurements within a given diagnostic test are the most informative ones. Furthermore, we are analyzing some computational issues which arise in the reasoning process.

The Gene Ontology (GO) is a controlled, hierarchically organized vocabulary for describing function of gene products taking part in biological systems. Today's high-throughput experiments measure expression of thousands of genes simultaneously using microarrays or various proteomics approaches. Afterwards, researchers typically focus on genes whose expression differs between e.g. healthy and diseased tissue, with the resulting gene lists being interpreted by statistical testing for over– and under-representation within GO categories [1]. Such a way of summarizing experimental results may prove inadequate in the future. As the high-throughput techniques become cheaper and more accurate, they will dependably detect even slight changes in gene expression. Consequently, the lists of relevant genes will grow in size, and so will the resulting lists of GO categories. Additionally, the interpretation of results is made difficult by high redundancy between individual GO categories. We propose a computational approach that would (a) enable flexible reduction in size for large user-supplied lists of overlapping GO categories, and (b) visualize the remaining GO terms in a two-dimensional space which reflects the terms' semantic interrelations. Our simple clustering-like algorithm relies on previously defined measures of semantic similarity in the GO space [2]. Dimensionality reduction techniques and graph—based visualization approaches will be used to derive informative visualizations. For more information refer to http://revigo.irb.hr. [1] I. Rivals, L.Personnaz, L. Taing, M-C. Potier. Bioinformatics, 23(4) (2007) 401 [2] A. Schlicker, M. Albrecht. Nucl. Acids Res., 36(Database issue) (2007) D434

We propose a clustering-like approach for flexible reduction in size of large user- supplied lists of overlapping Gene Ontology (GO) terms, typically resulting from high- throughput experiments. The remaining GO terms are visualized so as to faithfully reflect the terms' interrelations, relying on measures of semantic similarity in the GO space.

FUMACS code package upgrade to enable modeling of new fuel type (NGF) and gadolinium integral fuel burnable absorber for use in NPP Krško. Verification and validation of modified FUMACS on operating cycles.

Plasma biochemical profile was examined in cage cultured shade fish Argirosomus regius in order to evaluate physiological profile and provide detailed picture of its seasonal changes. Biochemical procedures were employed to study trygliceride (TRIG), cholesterol (CHOL), total protein (TP), glucose (GLU), urea (BUN), inorganic phosphate (PHOS), calcium (Ca), creatine kinase (CK), lactate dehydrogenase (LDH), ammonia (NH3), magnesium (Mg), aspartate aminotransferase (AST), amylase (AMYL), alkaline phosphatase (ALKP), creatinine (CREA), gamma glutamyltransferase (GGT) and total bilirubin (TBIL). The results were evaluated through different statistical analyses. Conventional statistical methods do not provide solutions that meet the goals of the research, so machine learning methodology was used to construct and evaluate classifiers as well as to assess parameter (plasma metabolites and enzymes) importances with respect to the target variable (season). It is evident that about half of the parameters (UREA, ALKP, NH3, CHOL, Ca, PHOS, GLU and CREA) have significant importance (i.e. exhibit significant differences between seasons) for seasonal discrimination models. It is also important to stress that using only a subset of most important parameters is generally enough to obtain high accuracy classifiers, which suggests rather high correlation between these parameters on a seasonal scale.

GORBI is an online database offering the results of computational gene function prediction in prokaryotic genomes. The analysis was done via the method of correlating gene occurrence patterns in selected organisms, termed phylogenetic profiling [1]. A machine learning algorithm based on decision trees for hierarchical multi-label classification [2] was used, and the annotations are represented in using the Gene Ontology vocabulary. The web interface will be used to present the results of our previous analyses in a clear and easily browsable accessible manner, thereby covering two angles in wet lab research: focusing on 1) function (needing prediction for genes having that function in any organism), and 2) organism (needing prediction of any function in this particular organism). The layout of the interface is shown in the figure. An early version of GORBI is freely available from http://gorbi.irb.hr/ 1. Kensche, P.R., et al., Practical and theoretical advances in predicting the function of a protein by its phylogenetic distribution. Journal of the Royal Society Interface, 2008. 5(19): p. 151-170. 2. Vens, C., et al., Decision trees for hierarchical multi-label classification. Machine Learning, 2008. 73(2): p. 185-214.

In this paper, recently proposed method for fast evaluation of reactor core loading pattern based on novel machine learning technique, Support Vector Regression - SVR, is analysed in respect of two SVR specific application issues: the input vector structure and the SVR parameter influence. The results of the experiments suggest that input vector structure is of secondary interest, and that focus of the research should be pointed to the adaptation of the SVR mechanism, namely optimization of the SVR parameters, to the requirements imposed by the application of the SVR method for the evaluation of core loading pattern.

This poster will present an extensive study of correlation of intergene region lengths to Gene Ontology (GO) assignments in prokaryotic genomes. Our intention is to assess the dependence of intergene region length of the genes in a particular GO category on the role of that GO category in the prokaryotic life cycle. Out of 624 prokaryotic complete genome sequences present at the NCBI FTP site (ftp://ftp.ncbi.nih.gov/genbank/genomes/Bacteria) in November 2007, 461 were selected. The selection was made in order to have exactly one representative of each species, and in that way – most importantly – to remove redundancies, and as a side benefit, to be more computationally efficient. Organisms we kept in the analysis are representatives of each species group contained in the original list. Prokaryotic organisms in general have short intergene regions, compared to many eukaryotes. This means that during evolution the prokaryotic cell was under strong selective pressures for growth rate and doubling time, and was particular about having extra DNA devoted to regulatory regions, as it takes approx. 50 milliseconds and two ATP molecules to transcribe one nucleotide. Combined with the extra cost of regulatory proteins, it would be expected that the minimal promoter is the only regulatory mechanism constitutively expressed genes (operons) would need. Basic metabolism and cell maintenance are pathways anticipated to be present in this group. On the other hand, genes that are expressed in specific environmental conditions (for example alternative carbon source use, heat or cold shock, lower pH or higher osmotic pressure) or at specific cell cycle stages (for example DNA replication/recombination initiation and regulation) require stringent expression control and as such need longer regulatory regions preceding them. In light of this expectation we present the results in an intuitive chart that distributes the Gene Ontology categories in a 3D space by using a semantic similarity metric between GO categories (simRel, Schlicker et al. BMC Bioinformatics 2006, 7:302), alongside the enrichment of the GO category in long or short intergene regions. To assess enrichment of particular Gene Ontology categories in genes having long intergene regions, we have calculated intergene region lengths and made a statistical analysis of their distribution over GO. This was done by first making a list of non-coding region lengths present upstream of each gene. Secondly, for a respective organism, a normalization of this value was done by dividing by the average intergene length in a particular genome element (prokaryote’ s chromosome or plasmid). Lastly, the genes were divided into two groups, depending on their intergene region (IGR) length being above average (long IGR, approx. 33% of genes) and below average (short IGR, approx. 67% of genes). This list was analyzed by using a number of Fisher’ s exact tests for correlation between two categorical variables – first, short or long IGR ; and second, member or non-member of a GO category. We used a permutation-based algorithm to correct the statistical significance estimates for multiple testing (Manly, Randomization, Bootstrap and Monte Carlo Methods in Biology, 3rd ed., 2007). The result is a table showing only significantly enriched Gene Ontology categories in long or short IGR genes. The poster will present a visualization of these results in an intuitive and interpretable manner. There are two aspects in our analysis: one is viewing the Gene Ontology categories that are enriched in long IGR genes (we suggest that these are the genes under rigid transcriptional control, in need of long regulatory regions), and the other is focusing on those enriched in short IGR genes (genes we suspect are constitutively expressed, and therefore do not need to be regulated). We have found here that the genes with long IGRs are assigned to GO categories responsible for functions such as citrate transport, carbon utilization, cell morphogenesis, DNA topoisomerase (ATP-hydrolyzing) activity, glutamate synthase activity, light-harvesting complex and excinuclease repair complex. The short IGR gene group contains members whose functions are designated to lipoprotein and fatty acid biosynthetic process, ion transport, protein refolding, cellular respiration, sensory perception of chemical stimulus, cofactor, RNA and lipid binding, tRNA adenylyltransferase activity, 1-phosphofructokinase activity, genes coding for small and large ribosomal subunits, and proton-transporting ATP synthase complex. In themselves these observations are quite intuitive, but, to our knowledge, no systematic examination of prokaryotic intergene regions with regard to GO annotations has been done previously. We present Gene Ontology categories that are either enriched or depleted in long intergene regions, and show a pool of GOs where constitutively expressed genes are found. We also intend to extend the research beyond this poster presentation to include comparisons of IGR lengths between groups of bacteria living in different habitats. It is tempting to hypothesize that in these long intergene regions a motif conserved for a particular GO function is present. This would reduce the prokaryote’ s cost of regulation by having a small pool of proteins that enable the transcription of the whole Gene Ontology category. One known example would be the cAMP receptor protein (CRP) that regulates transcription of over 100 genes in E. coli via a specific 22 bp motif (Brown and Callan, PNAS 2004, 101/8: 2404-9). Most notably – examining the generality level of the GO categories in question – would be worth looking into.

The purpose of this study was the identification and quantification of biochemical parameters over a 1-year cycle and to provide a detailed picture of seasonal changes in plasma metabolites and enzymes. Using the novel methods of machine learning techniques, the authors created and generated for the first time comprehensible classification models for exploring the importance of blood chemistry parameters, strength, mutual interactions or dependencies, and reliability of particular parameters within the seasonal groups.

There is a number of papers in the recent scientific literature that deal with a set of seemingly dissimilar bioinformatic problems. Their connecting motif is that they all have sequences of peptides, or fragments of proteins, as the focus of their interest. Generally, the peptide sequences in question have first been characterized by wet lab experiments. The results have been published in a paper, and possibly stored in a highly specialized database . In order to use the sequences for bioinformatics analyses of the processes that generated them a researcher needs to extract peptides from the database, analyze them and form a dataset. As the last step is time-consuming and arduous there is a general tendency for other scientists to re-use such datasets for trying to arrive to the even better bioinformatic models and, hopefully, understanding of the processes. The peptide datasets we have collected so far can be roughly divided into four categories, with the possibility of adding more categories as the need arises: a) posttranslational modification of proteins (e.g. phosphorylation) b) cleavage by broad-specificity proteases (e.g. proteasome, HIV-I protease) c) determining protein secondary structure d) epitope recognition (e.g. T-cell epitope recognition) As many of these processes are not only of biological but also of a medical importance, there is great interest in their further study and explanation. The peptide datasets can all generally be viewed as sets of amino acid sequences to which a class label has been assigned experimentally. However unconnected their underlying problem may seem at the first glance (e.g. prediction of protein secondary structure, prediction of phosphorylation sites in proteins) they all often serve for construction of classification models by similar supervised machine learning approaches. The final aim of the computational approaches is to develop a model that will best serve for the in silico predictions of events occuring in live cells. Most attempted modelling of these problems has met with the question of numerical representations of amino acid sequences, numerical representation being necessary for many classification algorithms. Various approaches to finding the best representation have been taken and they commonly compare amino acid representations on the same problem using the same classification method. Only rarely have researchers compared optimal amino acid representations between different problems. In the light of similarity between various peptide classification problems one can not help wondering whether there is an optimal amino acid representation that would work best for different problems and that would explain the most prominent amino acid features in shaping the natural processes in question. It is the aim of our future research to address this question in more detail. Despite the striking similarity between the peptide modelling problems, there is not an established flow of information. Quite a few of the field-specialized databases are readily available but need to be extensively pre-processed to result in the modelling-appropriate dataset. Additionaly, due to the frequent database updates and vagueness of descriptions of dataset production it is hard for different scientists to arrive to exactly the same dataset, and even slight variations in the datasets would invalidate a comparison of modelling approaches. Although some datasets are available on request, there is not an established format of dataset exchange, resulting in more time wasted on managing different data formats. Therefore, easing the process of data exchange by standardizing peptide dataset formats is a fundamental requirement for the better research in protein structural biology. We have chosen Extensible Markup Language (XML) for the production of such a data format. XML has been shown to be highly efficient in storing data in an orderly, researcher-comprehensible manner. On the one hand it is robust and on the other hand extensible which assures not only correct data distribution, but also adaptability in the instances in which hindsight fails. We hereby propose an xml format for the peptide sequences datasets which we hope will lead to the more efficient information exchange within this specific area of protein structural biology. In the future, we hope to further stimulate the information flow by building a peptide dataset repository.

Previously, pyrimidine nucleic base derivatives with a sulfonamide pharmacophore have been indicated as potential antitumor agents. We have shown that N-1-sulfonylpyrimidine derivatives have strong antiproliferative activity on human tumor cell lines, where 1-(p-toluenesulfonyl)cytosine (TsC) in specific was shown to have a selective effect with regard to normal cells and was easily synthesized on a large scale. Past experiments using radio-assays of enzyme activity have indicated that TsC induces a general shutdown in the cellular DNA, RNA and protein biosynthesis. In the present work we have used an interdisciplinary approach to further elucidate the compounds’ mechanistic class. Primarily, we have employed an augmented number of cell lines (eleven), of which eight overlapped with the DTP-NCI screening panel, and one non-transformed human fibroblast cell line. This has allowed us to computationally search for compounds with similar activity profiles and/or mechanistic class by integrating our data with the comprehensive DTP-NCI database ; a permutation testing procedure was used to estimate statistical significance of the matches. We have applied supervised data mining methodology (a Random Forest classifier), allowing us to get a prediction of mechanism of action, along with estimates of predictive reliability. When using only a subset of the full DTP-NCI 60 cell line panel, this approach may complement the information obtained from self organizing maps (SOM), a method commonly used in examinations of cytostatic activity profiles. Finally, we have performed cell cycle perturbation and apoptosis analysis of the most sensitive cell line (MCF-7), which has shown marked G1 phase arrest accompanied with the reduction of the number of cells in S phase. As expected, TsC did not alter the cell cycle of normal cells. Our results point to an unusual mechanism of cytostatic action, possibly a combination of nucleic acid antimetabolite activity and a novel molecular mechanism. We hypothesize that the novel mechanism might be similar to the activity of benzothiazoles, previously described as involving the aryl hydrocarbon receptor, activation of CYP1A1 and CYP1B1 genes and a subsequent DNA damage response. Our work has been published in the “ Investigational New Drugs” journal, Volume 26 (April 2008), pages 97-110, available from http://dx.doi.org/10.1007/s10637-007-9084-1