Research Presentation Abstracts
- Protein Sessions
- Data Sessions
- Machine Learning Sessions
- Disease Sessions
- Gene Sessions
- System Sessions
Monday 21. 10:30- 12:30. Protein session
10.30 – 10.50 Long talk
- Garegin Papoian, University of Maryland, US
In living systems, biomolecular dynamical processes, such as protein folding and also assembly and function of multi-protein machines, take place within the chemically and physically complex environment of a cell. In these chemically and physically heterogeneous environments, proteins experience both generic excluded volume interactions but also transient chemical interactions with the surrounding milieu. In the context of a living cell, these effects are further complicated by the temporal evolution and spatial heterogeneity of the cytosolic composition. In this talk, I will report on our recent discovery of stochastic resonance in protein folding dynamics, where the local fluctuations of temperature or the cytosolic solution around a protein may significantly accelerating protein folding kinetics. Following a separate line of work, I will also elaborate on our efforts to shed light on biomolecular mechanisms behind assembly and dynamics of nucleosomal particles, which are building blocks for higher-order chromatin fibers, mediating compaction and accessibility of genomic DNA in higher organisms.
10.50 – 12.10 Short talks (13+3)
- Luis Rueda, University of Windsor, CA
- Emilio Fenoy, Universidad Nacional de San Martin, AR
- Lucia Chemes, Fundación Instituto Leloir, AR
- Facundo Orts, Universidad Nacional de Mar del Plata, AR
- Franco Simonetti, Fundacion Instituto Leloir, AR
12.10 - 12.30 Method talks (8+2)
- Yasser Ruiz Blanco, Universidad Central "Marta Abreu" de Las Villas, CU
Background, Alignment-free methods for protein analyses have become increasingly relevant approaches due to their uses for protein annotations when no close homologous of known annotation exist, and for the identification of remote evolutionary relations among dissimilar proteins. ProtDCal is a recently introduced software, which leverages a divide-and-conquer strategy that gives rise to a novel alignment-free codification of amino acids sequences and protein structures. The obtained numerical descriptors are meant for patterns recognition studies in protein data. We have validated the low redundancy of the information content between of ProtDCal’s descriptors and implemented in PROFEAT web server. This program is the first tool presented for the wide-scale generation of general-purpose descriptors for 3D structures of proteins; Results, here we aim to describe the applicability of these features in three applications: i) the prediction of N-glycosylation sites and extraction of meaningful sequence motifs using a benchmarking datasets. ii) The classification of proteins according to enzyme/not-enzyme functional annotation. iii) The prediction of protein domain – domain interactions based on its residues sequence. These analyses show the potential impact of the introduced features for bioinformatic modelling of dissimilar proteomic data; Conclusions, ProtDCal (http://bioinf.sce.carleton.ca/PROTDCAL/) is a free and user-friendly tool that may provide a valuable platform for bioinformatic studies which use machine learning approaches to achieve effective annotations of protein functions, prediction of post-translational modifications and rational design of protein/peptides.
- NALVO ALMEIDA, Universidade Federal de Mato Grosso do Sul, BR
Background: Analyses of differential gene expression by RNA-Seq have been crucial for understanding the basis of cell phenotype and to infer molecular profiling in different conditions. Other important aspect involving gene expression is to determine the interactions between their corresponding expressed proteins, since most of molecular processes are mediated in cells by large protein interaction networks. Thus, joining information from expression levels and interaction networks may be a powerful tool to uncover particular physiological contexts and to develop new treatments and therapeutic targets.
Results: We present DEPICTViz - Differential Expression and Protein InteraCTions Visualization Tool, an easy-to-install, easy-to-use standalone graphical web application for visualizing simultaneously differential expression data and relationships among proteins through an interactive MA (mean average) plot. All genes in the MA plot are clickable, with links to databases of protein interactions and gene ontology, like STRING and COG, allowing further GO functional enrichment. Also, the tool provides links to KEGG, ECC, Swiss-Prot and UCSC Genome Browser, when available. A tab-separated file with MA coordinates and annotation information as well as a very simple configuration file with some parameters are suffice to have DEPICTViz working properly.
Conclusions: Tests using DEPICTViz have been done in several types of experimental data, like cancer sequencing, ncRNAs located in Hox clusters related to Huntington's Disease, genes of interest from Pseudomomas corrugata, and also differential expressed genes from the signalgrass Brachiaria decumbens, when exposed to high presence of aluminum. DEPICTViz is available at http://git.facom.ufms.br/bioinfo/depictviz.
Monday 21. 14:30- 16:30. Data session14.30 – 14.50 Tech-talk
14.50 – 15.30 Long talks (15+5)
- Pedro Ballester, INSERM, FR
- Santiago Videla, Laboratorio de Biología de Sistemas Integrativa , Fundación Instituto Leloir – IIBBA – CONICET, AR
15.30 – 16.30 Short talks (10+2)
- Antoine Zambelli, University of California, Berkeley, US
- Ignacio Sanchez, FCEN UBA - IQUIBICEN CONICET, AR
- Georgina Stegmayer, Institute for Signals, Systems and Computational Intelligence, AR
- Oussama Souiai, Institut Pasteur de Tunis, TN
The exponential growth of Omics and more specifically the NGS data has raised up a great technical challenge to experimentalists who are unused to bioinformatics skills and do not dispose of sufficient computing power.
Integrative analysis of data from various sources is needed to provide biological insights into biological systems.
Through, the invaluable financial support provided by H3ABioNet, we settled and managed a 72 core, 512 Gb of RAM and 12T of storage server. To facilitate the access to the server we settled the Pasteur Galaxy server (Pasteur_Galaxy), an open and powerful free web-based platform for integrative analysis of NGS data.
Pasteur_Galaxy is based upon Galaxy, one of the most popular bioinformatics workflow management systems, which is considered as a standard for sharing bioinformatics data, tools and results. As a Galaxy instance, Pasteur_Galaxy aims at providing a large range of bioinformatics tools for the analysis of various types of NGS data. Galaxy supports reproducible computational research by providing an environment for performing and recording bioinformatics analyses.
The Pasteur_Galaxy project has the following main objectives:
- Provide the academic scientific community with an open and sustainable powerful Galaxy instance with a guaranteed availability. The platform offers access to numerous and up‐to‐date tools for data analysis with help and support.
- Provide the possibility for H3ABioNet nodes and more broadly the H3Africa community to share their data and results.
- Propose innovative developments and new tools packaged for Galaxy (available in the Pasteur_Galaxy toolshed).
- Developing of new tools and services for Galaxy (wrappers, toolshed packages).
- Maria Martinez, Universidad de los Andes, CO
Next Generation Sequencing has moved the Big Data phenomenon into the Biological Sciences, turning into a computational challenge the understanding of biological data. In consequence, it is important to create tools that exploit human visual skills in the interpretation of this information. However, transforming genomic data into an image with biological meaning is particularly difficult because the information is not comprised in a single variable but a set of them. The distribution of genomic composition embedded in k-mer frequencies (frequencies of all possible substrings of size k) and k-mer biases (normalized k-mer frequencies) is a suitable approach, since it allow us to obtain a specific signature of different organisms in order to classify them. The main goal of this study was to develop an R function to transform a genomic sequence into a specific 2D image based on k-mer frequencies and biases. The function was developed in order to fragment the genome, reduce the dimensionality of genomic composition measurements and assign a specific color (RGB) to each fragment, transforming it into an image pixel. The function was applied to 20 genomes from two life domains. Sliding windows between 4-8kb (plus 10% of step) were tested to identify the most discriminatory features within and between genomes. Furthermore, the 2D images showed similar color pattern between closely related microorganisms; for each genome (evaluated through similarity of all ribosomal operons) the range of 5-6kb captured most of the biological information. In conclusion, image-based tools can help improve genomic comparisons, exploiting visual capabilities
Tuesday 22. 10.30-12.30. Machine learning session10.30 – 10.50 Tech-Talk
- CITES, Latin American Business Incubator located in Sunchales, Santa Fe
10.50 – 11.50 Long talks (15+5)
- Søren Brunak
- Morten Nielsen, Center for Biological Sequence Analysis, Technical University of Denmark, DK
- Georgina Stegmayer, Institute for Signals, Systems and Computational Intelligence, CONICET, AR
11.50 – 12.30 Short talks (10+3)
- Fiorella Cravero, Planta Piloto de Ingeniería Química,Planta Piloto de Ingeniería Química, AR
Background Given a new compound, identify compounds of the training set that are structurally similar to
this one it is the first step of a possible strategy to define the Applicability Domain (AD) of a model. To
determine this similarity, the data should be grouped. We are interested in fuzzy clustering algorithms whose
novelty resides in allowing an element belonging to more than one group using a degree of membership.
As a next and last step of the process, using a series of statistical tests, the capacity of the predictor should be
evaluated on the new compound. Determine the applicability domain of a model allows establishing the limits
inside which the prediction of a compound will be reliable. The definition of the chemical domain of a predictive
model allows a more practical use of it, as it will prevent spending time with compounds that will not be
applicable. More specifically, in QSAR/QSPR (Quantitative Structure-Activity Relationship) modeling estimate
the level of certainty to predict a new compound based on how similar it is with respect to the compounds used
to build the model, is a crucial step.
Results The results obtained by applying fuzzy clustering techniques for a variety of physicochemical
properties allow evaluating the advantages and limitations of this AD strategy.
Conclusions In this study the essential contribution is to identify the similarity of the compounds by using
fuzzy clustering techniques and to validate this strategy in predicting relevant molecular properties to rational
- Melissa Woghiren, University of Alberta, CA
Traditional biological methods to identify cis-regulatory genomic elements from transcriptional data can be time-consuming. Complete network mapping is often a problem far too complex to solve as a result of the inherent limitations of these methods. Computational approaches can help facilitate the process by using validated data to train models in efficiently discovering novel interactions. Many studies employ Support Vector Machines as the algorithm of choice, however due to the high-dimensionality of this type of data, we found Random Forest algorithms consistently perform better in cis-regulatory element classification. This result carried over multiple human cell lines and against other various algorithms. This important finding could lead to the development of far more powerful tools for the use of gene ontology creation, while making the classification of previously undefined genomic regions more attainable.
- Guillermo Luque y Guzman Saenz, BCEM - Biología Computacional y Ecología Microbiana, CO
We present TAXOFOR, a novel machine learning classifier using Random Forests to assign taxonomy to 16S rDNA gene derived by sequencing amplicons up to genus level, trained with annotated sequences from the GreenGenes database. Taking apart training time, it is faster than several of the de-facto tools with the same purpose in microbial ecology. In order to manage the DNA sequences, they are numerically represented as projections into a 3D space defined by the vertex of a tetrahedron. Afterwards, Discrete Fourier Transform allows to get their Power Spectra and use them as input to both train the classifier and predict their taxonomy. Parseval’s identity theorem ensures that similarity between the numerical representation of two DNA sequences can be gotten from their power spectra. Performance and assertiveness of TAXOFOR against UCLUST, RDP and MOTHUR was assessed while assigning taxonomy to the same set of 16S rDNA sequences. Fourier analysis has shown to be a useful resource in order to get an efficient way to assign 16S rDNA sequences flanked by forward and reverse primers typically used in microbial surveys. By using an ensemble of randomized trees as classifier, we have outperformed in terms of processing time three of the most popular available tools to do the same task. In addition to that, our classifier has proved to have an impressive prediction power with an average precision score near 98%, for the inspected taxa up to genus level, even without being trained with the whole set of sequences in GreenGenes database.
Tuesday 22. 14.30 – 16.30 Disease session
14.30 – 15.30 Long talks 15+5)
- Madeleine Scott, Stanford University, US
- Mariana Recamonde-Mendoza, Universidade Federal do Rio Grande do Sul, BR
- Maria Rabaglino, CONICET, AR
15.30 – 16.30 Short talks (12+3)
- Patricio Yankilevich, Instituto de Investigación en Biomedicina de Buenos Aires, AR
- Martin Abba, Centro de Investigaciones Inmunológicas Básicas y Aplicadas (CINIBA), AR
Background: The Cancer Genome Atlas project (TCGA) has generated genomic, transcriptomic, epigenomic and clinicopathological data among thousand of samples in almost every human tumor sites. Despite the TCGA data and their associated resources are publicly available, the full integration and interpretation of these data requires specialized knowledge and software. Here, we present an R package, which allows biologists to identify novel putative genetics and epigenetics mechanisms, related to cancer development among the huge omics data available at TCGA project and other functional genomics databases. Results: multiOmics consists of an integrated set of functions and pipelines that allow the retrieval, analysis and visualization of different omics data (https://gitlab.com/cancergenomics/multiomics). Briefly, we developed our own functions and reused several ones available in existing R packages (CGDS-R, multiMiR, survival, etc.). After the user provides the data types to be analyzed (e.g.: mRNA and miRNA profiles; mRNA and DNA methylation profiles, etc), multiOmics identify all significant correlation between paired features across the genome, the predicted sequence based interactions and their prognostic value. To illustrate the use of multiOmics, we performed an integrative analysis of miRNA, mRNA expression profiles and follow-up data among 10,000 samples retrieved from the TCGA Pan-Cancer project. The multiOmics pipeline allowed us to identify a group of novel miRNAs involved in the epigenetic modulation of coding RNAs associated with the malignant progression of the most frequent human carcinomas. Conclusions: multiOmics facilitates the integration and mining process of oncogenomics data obtained from publicly available repositories.
- Andreas Schüller, Pontificia Universidad Católica de Chile, CL
Predicting the macromolecular targets of small molecule compounds is important for drug discovery in order to flag off-targets, identify new targets of known drugs (drug repositioning) and to deorphanize ligands without known targets. Repositioning of known drugs has become especially attractive with rising drug development costs. Here, we present a chemical similarity approach to biological target prediction. Chemical similarity searching is a type of ligand-based approach motivated by the frequent observation that structurally similar compounds have similar physicochemical properties and possibly similar biological profiles. Small molecules were represented by Mold2 topological descriptors that were standardized. Target proteins were predicted by a nearest neighbor estimator based on the chemical similarity of query molecules to annotated ligands of biological targets. We validated our approach with a dataset of 391,219 drug-protein interactions between 240,103 ligands and 2,611 targets, derived from ChEMBL. Results were analyzed by receiver operating characteristic (ROC) analysis and 10-fold cross-validation. We obtained an average area under the ROC curve (AUC) of 0.988. We compared our approach with the SwissTargetPrediction method and obtained favorable results. These results indicate that our chemical similarity method is well suited for target prediction. In summary, we present a fast, structure-independent approach for biological target prediction with straightforward application in drug repositioning.
Acknowledgements: FONDECYT Nº 1161798
- Jose Tort, Universidad de la Republica, UY
Fasciolosis is a zoonotic disease affecting more than 300 million cattle and 250 million sheep worldwide, with estimated costs of 3 billion dollars annually. Fasciolosis is also an emerging human disease being relevant in certain areas of the South American altiplano. Triclabendazole (TCBZ) is the drug of choice for treatment, but reports of drug resistance have emerged in different countries both in livestock and more recently in humans.
As a first approach to characterize the resistant isolates we compared the transcriptome of adult worms from a Peruvian TCBZ and Albendazole (ABZ) resistant isolate, an Uruguayan isolate resistant to Albendazole (ABZ) but susceptible to TCBZ and a sensitive isolate susceptible to both drugs. We analyzed the global profiles of gene expression and also focused on putative candidate genes for resistance. We observed interesting variations in expression levels of genes associated with relevant processes in particular with cytoskeleton genes. Although further studies are needed, these data might lead some insights on the putative molecular mechanisms associated with the emergence of drug resistance.
Wednesday 23. 10.30 – 12.30. Genes session.
10.30 – 10.50Tech-Talk
10.50 -12.05 Long talks (20+5mins)
- Mariano Rubiolo, Center for Research and Development in Information System Engineering, AR
- Andrey Grigoriev, Rutgers University, US
- Alejandro Reyes, Universidad de los Andes, CO
- Jaime Moreno, Graduate Research Assistant/Universidad de los Andes, CO
Viruses are the most abundant biological entities on Earth. It is estimated that viral abundance exceeds that of Bacteria and Achaea by 10-fold. Far from being just parasites affecting host fitness, viruses are major players in any microbial ecosystem. In spite of their broad abundance, viruses, in particular bacteriophages, remain largely unknown since only about 20% of the sequences obtained from viral community DNA surveys could be annotated by comparison to public databases. In order to shed some light into this genetic dark matter, we developed a workflow to identify viral orthologous groups and determine those that are significantly associated with viral taxonomy or viral host data in order to propose viral signatures that could be used to annotate and characterize viral community sequencing efforts. The resulting dataset of Virus and Phages Orthologous Groups (ViPhOGs) is composed of 2,315 orthologous groups that are significantly associated with different viral taxonomy levels, identified using machine learning algorithms based on accurate viral classification. This ViPhOGs represent near 14,000 non-redundant viral genomes coding for 442,007 proteins. As a proof of concept, we download 54 metagenomes from five different biomes, apply the same workflow looking for characteristic ViPhOGs for each environment. Characteristic ViPhOGs were used coupled with GenSeed-HMM for targeted assembly of viral genomes. ViPhOGs that were characteristic for a biome and are also a taxonomic signature were used as seeds. These results show that ViPhOGs and the developed workflow could help the scientific community in the quest of annotating viral metagenomes, characterize environments and assembly viral genomes.
12.05 – 12.30 Short talks (10+2mins)
- Luciana Escobar, Universidad Nacional de La Plata, AR
New massive sequencing technologies have revealed the existence of micro-RNA (miRNA) and non-coding RNA (ncRNA), thus adding complexity to the process of regulating the expression of genomes . Distinguish between functional and non-functional transcripts is a difficult exercise that slows the identification of new genes [2,3]. However, in recent years there have been discovered numerous small functional peptides encoded by small ORF (sORF), many of which were originally considered noncoding [4,5]. Since the identification of candidates for functional sORF escapes the current bioinformatics protocols, it is necessary to develop new strategies capable of identifying them.
In this work we analyzed ribosomal profiles of 5'UTR regions of transcripts derived from Drosophila melanogaster genome. Peviously, we worked with data embryos  and detected 129 new peptides that were analyzed in terms of their functionality and phylogenetic conservation.
We are currently working in the identifcation of short peptides arising from potential sORF in samples of S2 cells . We evaluated the quality of alignments using Bowtie and TopHat software; mapping transcripts to a reference genome to discover RNA splice sites de novo.
Short peptide sequences constitutes a significant fraction of uncharacterized gene products encoded by a genome. the Ribosome Profiling technique infer novel peptide products derived from sequences previously considered non-coding . Furthermore, the use of TopHat as a spliced aligner form RNA-sequence, combines the ability to identify novel splice sites with direct mapping to known transcripts, producing sensitive and accurate alignments . These techniques allows us to deepen analysis in the characterization of sORFs.
- Natalia Macchiaroli, Instituto de investigaciones en microbiología y parasitología médica, AR
Background MicroRNAs (miRNAs), a class of small non-coding RNAs, are key regulators of gene expression at post-transcriptional level and play essential roles in biological processes such as development. MiRNAs silence target mRNAs by binding to complementary sequences in the 3’untranslated regions (UTRs). The parasitic helminths of the genus Echinococcus are the causative agents of echinococcosis, a zoonotic neglected disease. In previous works, we performed a comprehensive identification and characterization of Echinococcus miRNAs. However, current knowledge about their targets is limited. Since target prediction algorithms rely on complementarity between 3’UTRs and miRNA sequences, a major challenge is the lack of accurate sequence information of 3’UTR for most species including parasitic helminths. The aim of this study is to define a set of 3’UTRs and to predict miRNA targets in Echinococcus through a bioinformatics approach.
Results Using a pipeline that integrates genomic and transcriptomic data we generated a high-quality 3’UTR data set in Echinococcus. Bioinformatics prediction identified 941 potential miRNA target sites distributed in 724 3’UTRs. Most of them were found to be conserved among Echinococcus species, adding confidence to the predictions obtained. Functional analysis of miRNA targets showed that MAPK and Wnt signalling pathways were among the most represented pathways suggesting miRNA roles in parasite growth and development.
Conclusions Genome-wide identification and characterization of miRNA target genes in Echinococcus will provide valuable information to guide experimental studies in order to understand miRNA functions in parasite biology. MiRNAs involved in essential functions could be considered as novel therapeutic targets for echinococcosis control.
Wednesday 23, 14.30 – 16.30. Systems session.
14.30 – 15.30 Long talks
- Tiago Mendes, Universidade Federal de Viçosa, BR
- Olivia Alexandra Guerrero Gómez, CESUN, CO
- Yesid Cuesta Astroz, Grupo de Genômica e Biologia Computacional, Centro de Pesquisas René Rachou (CpqRR), Fundação Oswaldo Cruz (FIOCRUZ), Belo Horizonte, Brazil, BR
The study of molecular host-parasite interactions is essential to understand parasite infection, local adaptation within the host and can help identify drug targets. We propose an orthology based-method to predict the host-parasite interactome, using known intra-species interactions from other organisms. We transfer an interaction between two proteins within specie as long as it involves proteins that have an orthologous protein in the parasite and in the host. The method uses data from publicly databases such as eggNOG, which provides the orthologous groups for proteins, and STRINGdb to obtain confident protein-protein interactions (PPIs). iPfam and 3DID databases were used to identify possible host-parasite PPIs based on domain interactions. To give an integrative approach and context to the predicted interactions, we filtered the interactions according to the host’s tissues supporting the parasite’s tropism and subcellular localization. For this purpose we used the databases TISSUES and COMPARTMENTS. In the parasite’s side of the interactome we focused in the secretome predicted using different bioinformatics tools. We applied our method to identify host-parasite interactions in 13 parasites which share the same host (Homo sapiens). Our approach predicted tissue-specific networks. Host’s targeted proteins are key to the ability of the parasite to enter the host, subvert host immune defences and cause disease. These proteins were enriched by GO terms that could be used to evaluate the functional relevance of the predicted PPIs. In general, human’s enriched terms included: immune system pathways, cell adhesion and recognition, cellular response, inflammatory response, platelet activation and signal transduction.
15.30 – 16-30 Short talks
- Richa Batra, Helmholtz Zentrum Munich, DE
- Federico Sevlever, Instituto de Fisiología, Biología Molecular y Neurociencias, AR
Background. In a previous work we have presented the existence of a mechanism known as Pre-Equilibrium Sensing and Signaling (PRESS), which allows cells to discriminate between signaling levels that saturate receptors in equilibrium. This mechanism is based on the coupling between a slow and sustained detection module and a fast and transient response module. What results of PRESS is that certain signaling systems respond on a dose-dependent form, even for doses which saturates detection system in equilibrium, expanding the system’s dynamic range.
Results. Previously, we haven’t considered membrane receptor trafficking, that is, internalizing and recycling processes. These processes are rarely considered, assuming that they happen in slower time scales than sensing and transduction of signals. However, PRESS mechanism requires slower sensing processes, what makes scales comparable. It is important, then, to consider these processes together and this is done in the current work.
We have also analyzed cooperativity effects in membrane receptors with two binding sites. It is known that, in equilibrium, is not possible to distinguish between negative cooperativity and independent sites. Recent works show that pre-equilibrium information may allow to do so. We have studied in detail, with a combined theoretical/computational/statistical approach, the dynamics of cooperativity effects in the context of PRESS, in order to find measurable estimators for discrimination between these two microscopic scenarios.
Conclusions. More realistic models of the sensing steps of signaling, represented by membrane receptors with cooperativity and trafficking processes, are capable of working in PRESS mode, suggesting that PRESS mechanism might be widespread.
- Fredy Guevara Agudelo, Universidad Nacional de Colombia, CO
- Arsenio Rodriguez, Fundacion Instituto Estudios Avanzados - IDEA, VE
From the perspective of agroecology, the soil is considered as a living ecosystem, and soil quality cannot be measured directly, although they can be used certain specific and status of the soil as strategic indicators. Microbial communities are considered complex, dynamic assemblies with large phenotypic diversity and a high adaptive capacity. These "CS" are made up of individual agents, whose non-linear localized interactions have consequences to higher levels of organization, and these levels greater feedback and affect individual behaviors and local interactions. There is a huge effort to link the microbial diversity to soil health, although it has been weak and little progress has been made. It is our interest to address this issue through complexity modeling of the soil microbial diversity using systems biology approaching.
With the information from a case study is built a network of microbial structure and the physicochemical parameters of the soil (nodes) and were correlated between them (links). It was obtained some parameters of the reconstructed network (shortest path length, characteristic path length, network heterogeneity, clustering coefficient, betweenness centrality, closeness centrality, centroid value, and eccentricity). And with multivariate tools be has built a complex, multivariate index that allows to estimate the health of the soil.
We constructed a new index of complex nature both for the soil health and microbial diversity. This would allow us to study the health and diversity as emergent properties of the system soil-plant, as well as predict its resulting stability, resilience and sustainability under perturbations induced by agricultural practices.