Visual characteristics are among the most important features for characterizing the phenotype of ... more Visual characteristics are among the most important features for characterizing the phenotype of biological organisms. Color and geometric properties define population phenotype and allow assessing diversity and adaptation to environmental conditions. To analyze geometric properties classical morphometrics relies on biologically relevant landmarks which are manually assigned to digital images. Assigning landmarks is tedious and error prone. Predefined landmarks may in addition miss out on information which is not obvious to the human eye. The machine learning (ML) community has recently proposed new data analysis methods which by uncovering subtle features in images obtain excellent predictive accuracy. Scientific credibility demands however that results are interpretable and hence to mitigate the black-box nature of ML methods. To overcome the black-box nature of ML we apply complementary methods and investigate internal representations with saliency maps to reliably identify locat...
Recent progress in machine learning and deep learning has enabled the implementation of plant and... more Recent progress in machine learning and deep learning has enabled the implementation of plant and crop detection using systematic inspection of the leaf shapes and other morphological characters for identification systems for precision farming. However, the models used for this approach tend to become black-box models, in the sense that it is difficult to trace characters that are the base for the classification. The interpretability is therefore limited and the explanatory factors may not be based on reasonable visible characters. We investigate the explanatory factors of recent machine learning and deep learning models for plant classification tasks. Based on a Daucus carota and a Beta vulgaris image data set, we implement plant classification models and compare those models by their predictive performance as well as explainability. For comparison we implemented a feed forward convolutional neuronal network as a default model. To evaluate the performance, we trained an unsupervise...
Drug resistance poses a major challenge for targeted cancer therapy. To be able to functionally s... more Drug resistance poses a major challenge for targeted cancer therapy. To be able to functionally screen large randomly mutated target gene libraries for drug resistance mutations, we developed a biochemically defined high-throughput assay termed PhosphoFlowSeq. Instead of selecting for proliferation or resistance to apoptosis, PhosphoFlowSeq directly analyzes the enzymatic activities of randomly mutated kinases, thereby reducing the dependency on the signaling network in the host cell. Moreover, simultaneous analysis of expression levels enables compensation for expression-based biases on a single cell level. Using EGFR and its kinase inhibitor erlotinib as a model system, we demonstrate that the clinically most relevant resistance mutation T790M is reproducibly detected at high frequencies after four independent PhosphoFlowSeq selection experiments. Moreover, upon decreasing the selection pressure, also mutations which only confer weak resistance were identified, including T854A and L792H. We expect that PhosphoFlowSeq will be a valuable tool for the prediction and functional screening of drug resistance mutations in kinases.
Faba bean (Vicia faba L.) is an important source of protein but breeding for increased yield stab... more Faba bean (Vicia faba L.) is an important source of protein but breeding for increased yield stability and stress tolerance is hampered by the scarcity of phenotyping information. Because comparisons of cultivars adapted to different agro-climatic zones improve our understanding of stress tolerance mechanisms, the root architecture and morphology of 16 European faba bean cultivars were studied at maturity. Different machine learning (ML) approaches were tested in their usefulness to analyse trait variations between cultivars. A supervised, i.e. hypothesis-driven, ML approach revealed that cultivars from Portugal feature greater and coarser but less frequent lateral roots at the top of the taproot, potentially enhancing water uptake from deeper soil horizons. Unsupervised clustering revealed that trait differences between Northern and Southern cultivars are not predominant but that two cultivar groups, independently from major and minor types, differ largely in overall root system size. Methodological guidelines on how to use powerful machine learning methods such as random forest models for enhancing the phenotypical exploration of plants are given.
We present primary results from the Sequencing Quality Control (SEQC) project, coordinated by the... more We present primary results from the Sequencing Quality Control (SEQC) project, coordinated by the United States Food and Drug Administration. Examining Illumina HiSeq, Life Technologies SOLiD and Roche 454 platforms at multiple laboratory sites using reference RNA samples with built-in controls, we assess RNA sequencing (RNA-seq) performance for junction discovery and differential expression profiling and compare it to microarray and quantitative PCR (qPCR) data using complementary metrics. At all sequencing depths, we discover unannotated exon-exon junctions, with >80% validated by qPCR. We find that measurements of relative expression are accurate and reproducible across sites and platforms if specific filters are used. In contrast, RNAseq and microarrays do not provide accurate absolute measurements, and gene-specific biases are observed, for these and qPCR. Measurement performance depends on the platform and data analysis pipeline, and variation is large for transcript-level profiling. The complete SEQC data sets, comprising >100 billion reads (10Tb), provide unique resources for evaluating RNA-seq analyses for clinical and regulatory settings. Technological advances have made deep RNA sequencing feasible, expanding our view of the transcriptome 1 and promising to permit quantitative profiling with large dynamic range. 2 Recent comparisons of RNA-seq with established technologies for differential expression analysis have found good overall agreement between RNA-seq, qPCR and microarrays. In general, RNA-seq has provided increased detection sensitivity and opened new avenues of research in transcriptome analyses, such as the study of gene fusions, allele-specific expression and novel alternative transcripts. However, it has been shown that RNA-seq data exhibit measurement noise, which is a direct consequence of the random sampling process inherent to the assay. Assessments of RNA-seq have been limited to individual sequencing Users may view, print, copy, and download text and data-mine the content in such documents, for the purposes of academic research, subject always to the full Conditions of use:
Simplicity renders shake flasks ideal for strain selection and substrate optimization in biotechn... more Simplicity renders shake flasks ideal for strain selection and substrate optimization in biotechnology. Uncertainty during initial experiments may, however, cause adverse growth conditions and mislead conclusions. Using growth models for online predictions of future biomass (BM) and the arrival of critical events like low dissolved oxygen (DO) levels or when to harvest is hence important to optimize protocols. Established knowledge that unfavorable metabolites of growing microorganisms interfere with the substrate suggests that growth dynamics and, as a consequence, the growth model parameters may vary in the course of an experiment. Predictive monitoring of shake flask cultures will therefore benefit from estimating growth model parameters in an online and adaptive manner. This paper evaluates a newly developed particle filter (PF) which is specifically tailored to the requirements of biotechnological shake flask experiments. By combining stationary accuracy with fast adaptation to...
Protein complexes, macro-molecular assemblies of two or more proteins, play vital roles in numero... more Protein complexes, macro-molecular assemblies of two or more proteins, play vital roles in numerous cellular activities and collectively determine the cellular state. Despite the availability of a range of methods for analysing protein complexes, systematic analysis of complexes under multiple conditions has remained challenging. Approaches based on biochemical fractionation of intact, native complexes and correlation of protein profiles have shown promise, for instance in the combination of size exclusion chromatography (SEC) with accurate protein quantification by SWATH/DIA-MS. However, most approaches for interpreting co-fractionation datasets to yield complex composition, abundance and rearrangements between samples depend heavily on prior evidence. We introduce PCprophet, a computational framework to identify novel protein complexes from SEC-SWATH-MS data and to characterize their changes across different experimental conditions. We demonstrate accurate prediction of protein co...
The SIESTA project aims at defininga new description of human sleep. The basicidea of the algorit... more The SIESTA project aims at defininga new description of human sleep. The basicidea of the algorithm described below is thatextreme states of the sleeping subject are classifiedwith higher reliability. Hence we use labelsof the extreme states wake, deep sleep as wellas rapid eye movement sleep as well as all otherdata without labels. In order to have some benefitfrom unlabeled data, the classifier has tomodel class conditional densities and class priors. This allows to use Bayes theorem and predictthe a-posteriori probability of class. ...
The SIESTA project aims at defininga new description of human sleep. The sleepanalyzer is inferre... more The SIESTA project aims at defininga new description of human sleep. The sleepanalyzer is inferred by semi-supervised techniques. Using good features is very importantas we can not improve on them in subsequentprocessing stages. Hence we looked at variouspreprocessing techniques and applied featuresubset selection techniques to select the mostpromising ones. This paper presents results obtainedwith two entirely different techniques: Classical feature subset selection based on featureevaluation criteria and ...
This thesis investigates whether Bayesian inference can improve the reliabilityof biomedical diag... more This thesis investigates whether Bayesian inference can improve the reliabilityof biomedical diagnosis. In particular we discuss time series classication asis for example needed for an analysis of all-night sleep EEG recordings. Suchan attempt needs 4 steps that are further analyzed.
Coupled Hidden Markov Models (CHMM) are a natural extension to standard Hidden Markov Models. The... more Coupled Hidden Markov Models (CHMM) are a natural extension to standard Hidden Markov Models. They model interactions not in observation space, but in state space. In this paper we perform parameter estimation in the Bayesian paradigm, using Gibbs sampling, and in the Maximum Likelihood framework, using the expectation maximisation algorithm. The differences between the estimators are demonstrated on simulated data as well as biomedical data. Some recent work has focused on Hidden Markov models (HMM) in order to avoid some of the stationarity assumptions (the chain is still assumed to be ergodic) and model the dynamics explicitly [3]. HMMs handle the dynamics of the time-series through the use of a discrete state-space. These dynamics are taken from the state transition probabilities from one time step to the next. HMMs, however, typically have a single observation model and observation sequence and thus, by default, cannot handle the composite modelling of multiple time-series. This can be overcome through the use of multivariate observation models, such as multivariate autoregressive models [4]. The latter implies that interactions are modelled in the observationspace and not in state-space. This is unsatisfactory in some biomedical applications where the signals have completely different forms and thus spectra/bandwidths, e.g. heart-rate and respiration. Another approach models the interactions in state-space by coupling the statespaces of each variate [5]. This allows interaction analysis to proceed on arbitrary multichannel sequences.
We report about an automatic continuous sleep stager which isbased on probabilistic principles em... more We report about an automatic continuous sleep stager which isbased on probabilistic principles employing Hidden Markov Models(HMM). Our sleep stager oers the advantage of being objective bynot relying on human scorers, having much ner temporal resolution(one second instead of 30 seconds), and being based on solid probabilisticprinciples rather than a predened set of rules (Rechtschaen& Kales). Results obtained for nine
How does one model data with the aid of labels, when the labels themselves are noisy, unreliable ... more How does one model data with the aid of labels, when the labels themselves are noisy, unreliable and have their own dynamics? How does one measure interactions between variables that are so different in their nature that a direct comparison using, say cross-correlations, is meaningless? In this paper these problems are approached using Coupled Hidden Markov Models which are estimated in the Variational Bayesian framework. Signals can be diverse since each chain has its own observation model. Signals can have their own dynamics and may temporally lag or lead one another by allowing linking edges in the network topology to be estimated and chosen according to the most probable posterior model. Integrated feature extraction and modelling is accomplished by providing the Markov models models with linear observations models. We derive Coupled Hidden Markov Models estimators, apply and compare them with sampling based approaches found in the literature.
In this paper we will treat input selection for a radial basis function (RBF) like classifier wit... more In this paper we will treat input selection for a radial basis function (RBF) like classifier within a Bayesian framework. We approximate the a-posteriori distribution over both model coefficients and input subsets by samples drawn with Gibbs updates and reversible jump moves. Using some public datasets, we compare the classification accuracy of the method with a conventional ARD scheme. These datasets are also used to infer the a-posteriori probabilities of different input subsets.
Visual characteristics are among the most important features for characterizing the phenotype of ... more Visual characteristics are among the most important features for characterizing the phenotype of biological organisms. Color and geometric properties define population phenotype and allow assessing diversity and adaptation to environmental conditions. To analyze geometric properties classical morphometrics relies on biologically relevant landmarks which are manually assigned to digital images. Assigning landmarks is tedious and error prone. Predefined landmarks may in addition miss out on information which is not obvious to the human eye. The machine learning (ML) community has recently proposed new data analysis methods which by uncovering subtle features in images obtain excellent predictive accuracy. Scientific credibility demands however that results are interpretable and hence to mitigate the black-box nature of ML methods. To overcome the black-box nature of ML we apply complementary methods and investigate internal representations with saliency maps to reliably identify locat...
Recent progress in machine learning and deep learning has enabled the implementation of plant and... more Recent progress in machine learning and deep learning has enabled the implementation of plant and crop detection using systematic inspection of the leaf shapes and other morphological characters for identification systems for precision farming. However, the models used for this approach tend to become black-box models, in the sense that it is difficult to trace characters that are the base for the classification. The interpretability is therefore limited and the explanatory factors may not be based on reasonable visible characters. We investigate the explanatory factors of recent machine learning and deep learning models for plant classification tasks. Based on a Daucus carota and a Beta vulgaris image data set, we implement plant classification models and compare those models by their predictive performance as well as explainability. For comparison we implemented a feed forward convolutional neuronal network as a default model. To evaluate the performance, we trained an unsupervise...
Drug resistance poses a major challenge for targeted cancer therapy. To be able to functionally s... more Drug resistance poses a major challenge for targeted cancer therapy. To be able to functionally screen large randomly mutated target gene libraries for drug resistance mutations, we developed a biochemically defined high-throughput assay termed PhosphoFlowSeq. Instead of selecting for proliferation or resistance to apoptosis, PhosphoFlowSeq directly analyzes the enzymatic activities of randomly mutated kinases, thereby reducing the dependency on the signaling network in the host cell. Moreover, simultaneous analysis of expression levels enables compensation for expression-based biases on a single cell level. Using EGFR and its kinase inhibitor erlotinib as a model system, we demonstrate that the clinically most relevant resistance mutation T790M is reproducibly detected at high frequencies after four independent PhosphoFlowSeq selection experiments. Moreover, upon decreasing the selection pressure, also mutations which only confer weak resistance were identified, including T854A and L792H. We expect that PhosphoFlowSeq will be a valuable tool for the prediction and functional screening of drug resistance mutations in kinases.
Faba bean (Vicia faba L.) is an important source of protein but breeding for increased yield stab... more Faba bean (Vicia faba L.) is an important source of protein but breeding for increased yield stability and stress tolerance is hampered by the scarcity of phenotyping information. Because comparisons of cultivars adapted to different agro-climatic zones improve our understanding of stress tolerance mechanisms, the root architecture and morphology of 16 European faba bean cultivars were studied at maturity. Different machine learning (ML) approaches were tested in their usefulness to analyse trait variations between cultivars. A supervised, i.e. hypothesis-driven, ML approach revealed that cultivars from Portugal feature greater and coarser but less frequent lateral roots at the top of the taproot, potentially enhancing water uptake from deeper soil horizons. Unsupervised clustering revealed that trait differences between Northern and Southern cultivars are not predominant but that two cultivar groups, independently from major and minor types, differ largely in overall root system size. Methodological guidelines on how to use powerful machine learning methods such as random forest models for enhancing the phenotypical exploration of plants are given.
We present primary results from the Sequencing Quality Control (SEQC) project, coordinated by the... more We present primary results from the Sequencing Quality Control (SEQC) project, coordinated by the United States Food and Drug Administration. Examining Illumina HiSeq, Life Technologies SOLiD and Roche 454 platforms at multiple laboratory sites using reference RNA samples with built-in controls, we assess RNA sequencing (RNA-seq) performance for junction discovery and differential expression profiling and compare it to microarray and quantitative PCR (qPCR) data using complementary metrics. At all sequencing depths, we discover unannotated exon-exon junctions, with >80% validated by qPCR. We find that measurements of relative expression are accurate and reproducible across sites and platforms if specific filters are used. In contrast, RNAseq and microarrays do not provide accurate absolute measurements, and gene-specific biases are observed, for these and qPCR. Measurement performance depends on the platform and data analysis pipeline, and variation is large for transcript-level profiling. The complete SEQC data sets, comprising >100 billion reads (10Tb), provide unique resources for evaluating RNA-seq analyses for clinical and regulatory settings. Technological advances have made deep RNA sequencing feasible, expanding our view of the transcriptome 1 and promising to permit quantitative profiling with large dynamic range. 2 Recent comparisons of RNA-seq with established technologies for differential expression analysis have found good overall agreement between RNA-seq, qPCR and microarrays. In general, RNA-seq has provided increased detection sensitivity and opened new avenues of research in transcriptome analyses, such as the study of gene fusions, allele-specific expression and novel alternative transcripts. However, it has been shown that RNA-seq data exhibit measurement noise, which is a direct consequence of the random sampling process inherent to the assay. Assessments of RNA-seq have been limited to individual sequencing Users may view, print, copy, and download text and data-mine the content in such documents, for the purposes of academic research, subject always to the full Conditions of use:
Simplicity renders shake flasks ideal for strain selection and substrate optimization in biotechn... more Simplicity renders shake flasks ideal for strain selection and substrate optimization in biotechnology. Uncertainty during initial experiments may, however, cause adverse growth conditions and mislead conclusions. Using growth models for online predictions of future biomass (BM) and the arrival of critical events like low dissolved oxygen (DO) levels or when to harvest is hence important to optimize protocols. Established knowledge that unfavorable metabolites of growing microorganisms interfere with the substrate suggests that growth dynamics and, as a consequence, the growth model parameters may vary in the course of an experiment. Predictive monitoring of shake flask cultures will therefore benefit from estimating growth model parameters in an online and adaptive manner. This paper evaluates a newly developed particle filter (PF) which is specifically tailored to the requirements of biotechnological shake flask experiments. By combining stationary accuracy with fast adaptation to...
Protein complexes, macro-molecular assemblies of two or more proteins, play vital roles in numero... more Protein complexes, macro-molecular assemblies of two or more proteins, play vital roles in numerous cellular activities and collectively determine the cellular state. Despite the availability of a range of methods for analysing protein complexes, systematic analysis of complexes under multiple conditions has remained challenging. Approaches based on biochemical fractionation of intact, native complexes and correlation of protein profiles have shown promise, for instance in the combination of size exclusion chromatography (SEC) with accurate protein quantification by SWATH/DIA-MS. However, most approaches for interpreting co-fractionation datasets to yield complex composition, abundance and rearrangements between samples depend heavily on prior evidence. We introduce PCprophet, a computational framework to identify novel protein complexes from SEC-SWATH-MS data and to characterize their changes across different experimental conditions. We demonstrate accurate prediction of protein co...
The SIESTA project aims at defininga new description of human sleep. The basicidea of the algorit... more The SIESTA project aims at defininga new description of human sleep. The basicidea of the algorithm described below is thatextreme states of the sleeping subject are classifiedwith higher reliability. Hence we use labelsof the extreme states wake, deep sleep as wellas rapid eye movement sleep as well as all otherdata without labels. In order to have some benefitfrom unlabeled data, the classifier has tomodel class conditional densities and class priors. This allows to use Bayes theorem and predictthe a-posteriori probability of class. ...
The SIESTA project aims at defininga new description of human sleep. The sleepanalyzer is inferre... more The SIESTA project aims at defininga new description of human sleep. The sleepanalyzer is inferred by semi-supervised techniques. Using good features is very importantas we can not improve on them in subsequentprocessing stages. Hence we looked at variouspreprocessing techniques and applied featuresubset selection techniques to select the mostpromising ones. This paper presents results obtainedwith two entirely different techniques: Classical feature subset selection based on featureevaluation criteria and ...
This thesis investigates whether Bayesian inference can improve the reliabilityof biomedical diag... more This thesis investigates whether Bayesian inference can improve the reliabilityof biomedical diagnosis. In particular we discuss time series classication asis for example needed for an analysis of all-night sleep EEG recordings. Suchan attempt needs 4 steps that are further analyzed.
Coupled Hidden Markov Models (CHMM) are a natural extension to standard Hidden Markov Models. The... more Coupled Hidden Markov Models (CHMM) are a natural extension to standard Hidden Markov Models. They model interactions not in observation space, but in state space. In this paper we perform parameter estimation in the Bayesian paradigm, using Gibbs sampling, and in the Maximum Likelihood framework, using the expectation maximisation algorithm. The differences between the estimators are demonstrated on simulated data as well as biomedical data. Some recent work has focused on Hidden Markov models (HMM) in order to avoid some of the stationarity assumptions (the chain is still assumed to be ergodic) and model the dynamics explicitly [3]. HMMs handle the dynamics of the time-series through the use of a discrete state-space. These dynamics are taken from the state transition probabilities from one time step to the next. HMMs, however, typically have a single observation model and observation sequence and thus, by default, cannot handle the composite modelling of multiple time-series. This can be overcome through the use of multivariate observation models, such as multivariate autoregressive models [4]. The latter implies that interactions are modelled in the observationspace and not in state-space. This is unsatisfactory in some biomedical applications where the signals have completely different forms and thus spectra/bandwidths, e.g. heart-rate and respiration. Another approach models the interactions in state-space by coupling the statespaces of each variate [5]. This allows interaction analysis to proceed on arbitrary multichannel sequences.
We report about an automatic continuous sleep stager which isbased on probabilistic principles em... more We report about an automatic continuous sleep stager which isbased on probabilistic principles employing Hidden Markov Models(HMM). Our sleep stager oers the advantage of being objective bynot relying on human scorers, having much ner temporal resolution(one second instead of 30 seconds), and being based on solid probabilisticprinciples rather than a predened set of rules (Rechtschaen& Kales). Results obtained for nine
How does one model data with the aid of labels, when the labels themselves are noisy, unreliable ... more How does one model data with the aid of labels, when the labels themselves are noisy, unreliable and have their own dynamics? How does one measure interactions between variables that are so different in their nature that a direct comparison using, say cross-correlations, is meaningless? In this paper these problems are approached using Coupled Hidden Markov Models which are estimated in the Variational Bayesian framework. Signals can be diverse since each chain has its own observation model. Signals can have their own dynamics and may temporally lag or lead one another by allowing linking edges in the network topology to be estimated and chosen according to the most probable posterior model. Integrated feature extraction and modelling is accomplished by providing the Markov models models with linear observations models. We derive Coupled Hidden Markov Models estimators, apply and compare them with sampling based approaches found in the literature.
In this paper we will treat input selection for a radial basis function (RBF) like classifier wit... more In this paper we will treat input selection for a radial basis function (RBF) like classifier within a Bayesian framework. We approximate the a-posteriori distribution over both model coefficients and input subsets by samples drawn with Gibbs updates and reversible jump moves. Using some public datasets, we compare the classification accuracy of the method with a conventional ARD scheme. These datasets are also used to infer the a-posteriori probabilities of different input subsets.
Faba bean (Vicia faba L.) is an important source of protein but breeding for increased yield stab... more Faba bean (Vicia faba L.) is an important source of protein but breeding for increased yield stability and stress tolerance is hampered by the scarcity of phenotyping information. Because comparisons of cultivars adapted to different agro-climatic zones improve our understanding of stress tolerance mechanisms, the root architecture and morphology of 16 pan-European faba bean cultivars were studied at maturity. Different machine learning (ML) approaches were tested in their usefulness to analyse trait variations between cultivars. A supervised, i.e. hypothesis-driven, ML approach revealed that cultivars from Portugal feature greater and coarser but less frequent lateral roots at the top of the taproot, potentially enhancing water uptake from deeper soil horizons. Unsupervised clustering revealed that trait differences between Northern and Southern cultivars are not predominant but that two cultivar groups, independently from major and minor types, differ largely in overall root system size. Methodological guidelines on how to use powerful machine learning methods such as random forest models for enhancing the phenotypical exploration of plants are given.
Uploads
Papers by Peter Sykacek