Oleg Sysoev
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Oleg Sysoev has a total of 53 co-authors. Here are the top co-authors by number of shared publications:
- Oleg Burdakov
Linköping University, Linköping, Sweden
4 shared publications A. Grimvall
Department of Computer and Information Science , Linköping University , Linköping , SE-58183 , Sweden
2 shared publications- Carina Källestål
London School of Hygiene & Tropical Medicine
2 shared publications - Carlos P. Roca
B Department of Chemical Engineering , Universitat Rovira i Virgili , Tarragona , Spain
2 shared publications - Elena Eyre
BioPharmaceuticals R&D, AstraZeneca, Barcelona, Spain
2 shared publications - Elmer Zelaya Blandón
Asociación para el Desarrollo Económico y Sostenible de El Espino (APRODESE), Chinandega, Nicaragua
2 shared publications
Latest Publications
- Open Access
- 17 August 2026
- Published by Springer Nature in Scientific Reports
- Vol. 16 (1)
Anomaly detection as a tool for identifying cases of interest in postmortem toxicology
Intoxication-induced death, both intentional and unintentional, is a global public health concern, contributing substantially to mortality in many regions. Postmortem toxicological interpretation typically relies on reference concentrations of individual substances, which fails to account for combinatorial effects when multiple drugs are present. More comprehensive approaches that analyse multiple substances simultaneously could improve diagnostic accuracy. This study applied computational anomaly detection methods to identify toxicological profiles deviating from known non-intoxication patterns and assessed whether such deviations correspond to fatal intoxications or could serve as a screening tool for cases requiring further review. Postmortem femoral blood drug concentrations covering 191 substances from 16,710 forensic autopsies were analyzed. Cases were divided into a training set of non-intoxications ( n = 8320) and separate validation ( n = 2674) and test sets ( n = 3342) containing both intoxications and non-intoxications. A machine learning–based anomaly detection model was trained and optimized before being applied to the test set. The model distinguished intoxications from normal patterns with 83.9% accuracy. False negatives often involved opioids and ethanol, while false positives were associated with rare substances, numerous detections, or high concentrations. Overall, this approach demonstrates strong potential as a screening tool for identifying toxicological patterns of forensic interest.
- 17 August 2026
- Published by Springer Nature in Scientific Reports
- Vol. 16 (1)
Open AccessAnomaly detection as a tool for identifying cases of interest in postmortem toxicology
Intoxication-induced death, both intentional and unintentional, is a global public health concern, contributing substantially to mortality in many regions. Postmortem toxicological interpretation typically relies on reference concentrations of individual substances, which fails to account for combinatorial effects when multiple drugs are present. More comprehensive approaches that analyse multiple substances simultaneously could improve diagnostic accuracy. This study applied computational anomaly detection methods to identify toxicological profiles deviating from known non-intoxication patterns and assessed whether such deviations correspond to fatal intoxications or could serve as a screening tool for cases requiring further review. Postmortem femoral blood drug concentrations covering 191 substances from 16,710 forensic autopsies were analyzed. Cases were divided into a training set of non-intoxications ( n = 8320) and separate validation ( n = 2674) and test sets ( n = 3342) containing both intoxications and non-intoxications. A machine learning–based anomaly detection model was trained and optimized before being applied to the test set. The model distinguished intoxications from normal patterns with 83.9% accuracy. False negatives often involved opioids and ethanol, while false positives were associated with rare substances, numerous detections, or high concentrations. Overall, this approach demonstrates strong potential as a screening tool for identifying toxicological patterns of forensic interest.
- Open Access
- 26 February 2026
- Published by American Chemical Society (ACS) in Analytical Chemistry
- Vol. 98 (9) , 6589-6597
LCMS-Net: Deep Learning for Raw High Resolution Mass Spectrometry Data Applied to Forensic Cause-of-Death Screening
Current preprocessing workflows for untargeted metabolomics using liquid chromatography-high resolution mass spectrometry (LC-HRMS) are time-consuming and require significant domain knowledge. Furthermore, they lack reproducibility or may fail to detect some metabolites entirely. We introduce LCMS-Net, an end-to-end deep learning model for the analysis of LC-HRMS data, to address these challenges. LCMS-Net mitigates the need for manual data preprocessing by operating directly on the raw LC-HRMS data and explicitly modeling its spatial properties. The effectiveness of this fully automated workflow is shown through two case-studies, cause-of-death (CoD) screening and colon cancer detection. For the cause-of-death screening task, LCMS-Net achieved a 9% improvement in F1-score compared to the previous state-of-the-art model (OPLS-DA). For the colon cancer detection task, LCMS-Net achieved an F1-score improvement of 1.8% compared to the previous state-of-the-art model (DeepMSProfiler). Furthermore, LCMS-Net significantly reduces batch effects that are a common source of bias in metabolomics data analyses. This was shown by using a training and test set from different measurement instruments, where the performance only differed by at most 3% as to using data from the same instrument. Compared to other end-to-end deep learning methods for LC-HRMS data, LCMS-Net is also structurally simpler and does not rely on pretraining, which makes it faster and computationally more efficient.
- 26 February 2026
- Published by American Chemical Society (ACS) in Analytical Chemistry
- Vol. 98 (9) , 6589-6597
Open AccessLCMS-Net: Deep Learning for Raw High Resolution Mass Spectrometry Data Applied to Forensic Cause-of-Death Screening
Current preprocessing workflows for untargeted metabolomics using liquid chromatography-high resolution mass spectrometry (LC-HRMS) are time-consuming and require significant domain knowledge. Furthermore, they lack reproducibility or may fail to detect some metabolites entirely. We introduce LCMS-Net, an end-to-end deep learning model for the analysis of LC-HRMS data, to address these challenges. LCMS-Net mitigates the need for manual data preprocessing by operating directly on the raw LC-HRMS data and explicitly modeling its spatial properties. The effectiveness of this fully automated workflow is shown through two case-studies, cause-of-death (CoD) screening and colon cancer detection. For the cause-of-death screening task, LCMS-Net achieved a 9% improvement in F1-score compared to the previous state-of-the-art model (OPLS-DA). For the colon cancer detection task, LCMS-Net achieved an F1-score improvement of 1.8% compared to the previous state-of-the-art model (DeepMSProfiler). Furthermore, LCMS-Net significantly reduces batch effects that are a common source of bias in metabolomics data analyses. This was shown by using a training and test set from different measurement instruments, where the performance only differed by at most 3% as to using data from the same instrument. Compared to other end-to-end deep learning methods for LC-HRMS data, LCMS-Net is also structurally simpler and does not rely on pretraining, which makes it faster and computationally more efficient.
- Open Access
bigMICE: Multiple Imputation of Big Data
Missing data is a prevalent issue in many applications, including large medical registries such as the Swedish Healthcare Quality Registries, potentially leading to biased or inefficient analyses if not handled properly. Multiple Imputation by Chained Equations (MICE) is a popular and versatile method for handling multivariate missing data but traditional implementations face significant challenges when applied to big data sets due to computational time and memory limitations. To address this, the bigMICE package was developed, adapting the MICE framework to big data using Apache Spark MLLib and Spark ML. Our implementation allows for controlling the maximum memory usage during the execution, enabling processing of very large data sets on a hardware with a limited memory, such as ordinary laptops. The developed package was tested on a large Swedish medical registry to measure memory usage, runtime and dependence of the imputation quality on sample size and on missingness proportion in the data. In conclusion, our method is generally more memory efficient and faster on large data sets compared to a commonly used MICE implementation. We also demonstrate that working with very large datasets can result in high quality imputations even when a variable has a large proportion of missing data. This paper also provides guidelines and recommendations on how to install and use our open source package.
Open AccessbigMICE: Multiple Imputation of Big Data
Missing data is a prevalent issue in many applications, including large medical registries such as the Swedish Healthcare Quality Registries, potentially leading to biased or inefficient analyses if not handled properly. Multiple Imputation by Chained Equations (MICE) is a popular and versatile method for handling multivariate missing data but traditional implementations face significant challenges when applied to big data sets due to computational time and memory limitations. To address this, the bigMICE package was developed, adapting the MICE framework to big data using Apache Spark MLLib and Spark ML. Our implementation allows for controlling the maximum memory usage during the execution, enabling processing of very large data sets on a hardware with a limited memory, such as ordinary laptops. The developed package was tested on a large Swedish medical registry to measure memory usage, runtime and dependence of the imputation quality on sample size and on missingness proportion in the data. In conclusion, our method is generally more memory efficient and faster on large data sets compared to a commonly used MICE implementation. We also demonstrate that working with very large datasets can result in high quality imputations even when a variable has a large proportion of missing data. This paper also provides guidelines and recommendations on how to install and use our open source package.
- Open Access
- 27 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 101-107
ERFA: Combining expression data with large knowledge graphs and regulation flow for MoA discovery
- 27 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 101-107
Open AccessERFA: Combining expression data with large knowledge graphs and regulation flow for MoA discovery
- Open Access
Regulation Flow Analysis discovers molecular mechanisms of action from large knowledge databases
Drug development is a long and expensive process, with only a small fraction of potential drugs being finally approved. The challenge of drug development is rooted in our limited understanding of biological systems and the disease processes that drugs are trying to modulate. We propose a novel method, called Regulation Flow Analysis (RFA), which is based on the principles of biological regulation, causal graphs, and graph flow. RFA is able to generate causal hypotheses of mechanism of action, using large Knowledge Graphs (KG) of molecular regulation. Our numerical experiments demonstrate that the generated hypotheses, in the form of regulation pathways, often summarize mechanisms of drug action in a manner both understandable and actionable. Thus, RFA can greatly improve our understanding of the biological processes underlying health and disease, and therefore substantially facilitate drug development. Our examples illustrate how RFA recovers known mechanisms of action and can be used for target and biomarker discovery and validation.
Open AccessRegulation Flow Analysis discovers molecular mechanisms of action from large knowledge databases
Drug development is a long and expensive process, with only a small fraction of potential drugs being finally approved. The challenge of drug development is rooted in our limited understanding of biological systems and the disease processes that drugs are trying to modulate. We propose a novel method, called Regulation Flow Analysis (RFA), which is based on the principles of biological regulation, causal graphs, and graph flow. RFA is able to generate causal hypotheses of mechanism of action, using large Knowledge Graphs (KG) of molecular regulation. Our numerical experiments demonstrate that the generated hypotheses, in the form of regulation pathways, often summarize mechanisms of drug action in a manner both understandable and actionable. Thus, RFA can greatly improve our understanding of the biological processes underlying health and disease, and therefore substantially facilitate drug development. Our examples illustrate how RFA recovers known mechanisms of action and can be used for target and biomarker discovery and validation.
- 9 December 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 742-746
Cell type identification for single cell RNA data by bulk data reference projection
Single cell RNA sequencing enables high resolution biological data that can be used for refined biological analyses. Many of these analyses require identification of biological cell types in the data. Some existing methods use bulk data with known cell type labels to identify cell types in the single cell data. In this work, we propose a novel approach that is using bulk data as well. Compared to the existing approaches mapping single cell data into the space of the bulk data, our method does the opposite: it maps the bulk observations to the space of the denoised single cell observations. Our numerical results illustrate that our approach generally leads to more accurate annotations, especially when other methods struggle to find a clear separation between the cell types. In particular, our method was able to separate between CD4+ and CD8+T-cells that other approaches could not achieve by using given bulk data.
- 9 December 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 742-746
Cell type identification for single cell RNA data by bulk data reference projection
Single cell RNA sequencing enables high resolution biological data that can be used for refined biological analyses. Many of these analyses require identification of biological cell types in the data. Some existing methods use bulk data with known cell type labels to identify cell types in the single cell data. In this work, we propose a novel approach that is using bulk data as well. Compared to the existing approaches mapping single cell data into the space of the bulk data, our method does the opposite: it maps the bulk observations to the space of the denoised single cell observations. Our numerical results illustrate that our approach generally leads to more accurate annotations, especially when other methods struggle to find a clear separation between the cell types. In particular, our method was able to separate between CD4+ and CD8+T-cells that other approaches could not achieve by using given bulk data.
- Open Access
- 29 October 2019
- Published by Springer Nature in International Journal for Equity in Health
- Vol. 18 (1) , 1-12
Predicting poverty. Data mining approaches to the health and demographic surveillance system in Cuatro Santos, Nicaragua
Background: In order to further identify the needed interventions for continued poverty reduction in our study area Cuatro Santos, northern Nicaragua, we aimed to elucidate what predicts poverty, measured by the Unsatisfied Basic Need index. This analysis was done by using decision tree methodology applied to the Cuatro Santos health and demographic surveillance databases. Methods: Using variables derived from the health and demographic surveillance update 2014, transferring individual data to the household level we used the decision tree framework Conditional Inference trees to predict the outcome “poverty” defined as two to four unsatisfied basic needs using the Unsatisfied Basic Need Index. We further validated the trees by applying Conditional random forest analyses in order to assess and rank the importance of predictors about their ability to explain the variation of the outcome “poverty.” The majority of the Cuatro Santos households provided information and the included variables measured housing conditions, assets, and demographic experiences since the last update (5 yrs), earlier participation in interventions and food security during the last 4 weeks. Results: Poverty was rare in households that have some assets and someone in the household that has a higher education than primary school. For these households participating in the intervention that installed piped water with water meter was most important, but also when excluding this variable, the resulting tree showed the same results. When assets were not taken into consideration, the importance of education was pronounced as a predictor for welfare. The results were further strengthened by the validation using Conditional random forest modeling showing the same variables being important as predicting the outcome in the CI tree analysis. As assets can be a result, rather than a predictor of more affluence our results in summary point specifically to the importance of education and participation in the water installation intervention as predictors for more affluence. Conclusion: Predictors of poverty are useful for directing interventions and in the Cuatro Santos area education seems most important to prioritize. Hopefully, the lessons learned can continue to develop the Cuatro Santos communities as well as development in similar poor rural settings around the world.
- 29 October 2019
- Published by Springer Nature in International Journal for Equity in Health
- Vol. 18 (1) , 1-12
Open AccessPredicting poverty. Data mining approaches to the health and demographic surveillance system in Cuatro Santos, Nicaragua
Background: In order to further identify the needed interventions for continued poverty reduction in our study area Cuatro Santos, northern Nicaragua, we aimed to elucidate what predicts poverty, measured by the Unsatisfied Basic Need index. This analysis was done by using decision tree methodology applied to the Cuatro Santos health and demographic surveillance databases. Methods: Using variables derived from the health and demographic surveillance update 2014, transferring individual data to the household level we used the decision tree framework Conditional Inference trees to predict the outcome “poverty” defined as two to four unsatisfied basic needs using the Unsatisfied Basic Need Index. We further validated the trees by applying Conditional random forest analyses in order to assess and rank the importance of predictors about their ability to explain the variation of the outcome “poverty.” The majority of the Cuatro Santos households provided information and the included variables measured housing conditions, assets, and demographic experiences since the last update (5 yrs), earlier participation in interventions and food security during the last 4 weeks. Results: Poverty was rare in households that have some assets and someone in the household that has a higher education than primary school. For these households participating in the intervention that installed piped water with water meter was most important, but also when excluding this variable, the resulting tree showed the same results. When assets were not taken into consideration, the importance of education was pronounced as a predictor for welfare. The results were further strengthened by the validation using Conditional random forest modeling showing the same variables being important as predicting the outcome in the CI tree analysis. As assets can be a result, rather than a predictor of more affluence our results in summary point specifically to the importance of education and participation in the water installation intervention as predictors for more affluence. Conclusion: Predictors of poverty are useful for directing interventions and in the Cuatro Santos area education seems most important to prioritize. Hopefully, the lessons learned can continue to develop the Cuatro Santos communities as well as development in similar poor rural settings around the world.
- Open Access
- 27 June 2019
- Published by Wiley in Statistics in Medicine
- Vol. 38 (22) , 4436-4452
PSICA: Decision trees for probabilistic subgroup identification with categorical treatments
Personalized medicine aims at identifying best treatments for a patient with given characteristics. It has been shown in the literature that these methods can lead to great improvements in medicine compared to traditional methods prescribing the same treatment to all patients. Subgroup identification is a branch of personalized medicine, which aims at finding subgroups of the patients with similar characteristics for which some of the investigated treatments have a better effect than the other treatments. A number of approaches based on decision trees have been proposed to identify such subgroups, but most of them focus on two‐arm trials (control/treatment) while a few methods consider quantitative treatments (defined by the dose). However, no subgroup identification method exists that can predict the best treatments in a scenario with a categorical set of treatments. We propose a novel method for subgroup identification in categorical treatment scenarios. This method outputs a decision tree showing the probabilities of a given treatment being the best for a given group of patients as well as labels showing the possible best treatments. The method is implemented in an R package psica available on CRAN. In addition to a simulation study, we present an analysis of a community‐based nutrition intervention trial that justifies the validity of our method.
- 27 June 2019
- Published by Wiley in Statistics in Medicine
- Vol. 38 (22) , 4436-4452
Open AccessPSICA: Decision trees for probabilistic subgroup identification with categorical treatments
Personalized medicine aims at identifying best treatments for a patient with given characteristics. It has been shown in the literature that these methods can lead to great improvements in medicine compared to traditional methods prescribing the same treatment to all patients. Subgroup identification is a branch of personalized medicine, which aims at finding subgroups of the patients with similar characteristics for which some of the investigated treatments have a better effect than the other treatments. A number of approaches based on decision trees have been proposed to identify such subgroups, but most of them focus on two‐arm trials (control/treatment) while a few methods consider quantitative treatments (defined by the dose). However, no subgroup identification method exists that can predict the best treatments in a scenario with a categorical set of treatments. We propose a novel method for subgroup identification in categorical treatment scenarios. This method outputs a decision tree showing the probabilities of a given treatment being the best for a given group of patients as well as labels showing the possible best treatments. The method is implemented in an R package psica available on CRAN. In addition to a simulation study, we present an analysis of a community‐based nutrition intervention trial that justifies the validity of our method.
- 1 April 2019
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-5
Inter-Frequency Radio Signal Quality Prediction for Handover, Evaluated in 3GPP LTE
Radio resource management in cellular networks is typically based on device measurements reported to the serving base station. Frequent measuring of signal quality on available frequencies would allow for highly reliable networks and optimal connection at all times. However, these measurements are associated with costs, such as dedicated device time for performing measurements when the device will be unavailable for communication. To reduce the costs, we consider predictions of inter-frequency radio quality measurements that are useful to assess potential inter-frequency handover decisions. In this contribution, we have considered measurements from a live 3GPP LTE network. We demonstrate that straightforward applications of the most commonly used machine learning models are unable to provide high accuracy predictions. Instead, we propose a novel approach with a duo-threshold for high accuracy decision recommendations. Our approach leads to class specific prediction accuracies as high as 92% and 95%, still drastically reducing the need for inter-frequency measurements.
- 1 April 2019
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-5
Inter-Frequency Radio Signal Quality Prediction for Handover, Evaluated in 3GPP LTE
Radio resource management in cellular networks is typically based on device measurements reported to the serving base station. Frequent measuring of signal quality on available frequencies would allow for highly reliable networks and optimal connection at all times. However, these measurements are associated with costs, such as dedicated device time for performing measurements when the device will be unavailable for communication. To reduce the costs, we consider predictions of inter-frequency radio quality measurements that are useful to assess potential inter-frequency handover decisions. In this contribution, we have considered measurements from a live 3GPP LTE network. We demonstrate that straightforward applications of the most commonly used machine learning models are unable to provide high accuracy predictions. Instead, we propose a novel approach with a duo-threshold for high accuracy decision recommendations. Our approach leads to class specific prediction accuracies as high as 92% and 95%, still drastically reducing the need for inter-frequency measurements.
- Open Access
Assessing the multiple dimensions of poverty. Data mining approaches to the 2004-14 Health and Demographic Surveillance System in Cuatro Santos, Nicaragua
We aimed to describe multiple dimensions of poverty according to the capability approach theory by applying data mining approaches to the Cuatro Santos Health and Demographic Surveillance databases, Nicaragua. Four municipalities in northern Nicaragua constitute the Cuatro Santos area, with 25,893 inhabitants in 5,966 households (2014). A local process analyzing poverty-related problems and prioritizing suggested actions, was initiated 1997 based on a community action plan 2002-2015. Priority interventions were school breakfasts, environmental protection, water and sanitation, preventive healthcare, home gardening, micro credits, technical training, stipends for university education, and the use of Internet. In 2004, a survey of basic health and demographic information was performed in the whole population followed by surveillance updates in 2007, 2009, and 2014 linking households and individuals by unique identifiers. Information included the house (floor, walls) and services (water, sanitation, electricity) as well as demographic data (birth, deaths, migration). Data on participation in interventions, on food security, household assets, and women’s self-rated health were collected in 2014. A K-means algorithm were used to cluster the household data (54 variables). The poverty ranking of household clusters using the unsatisfied basic needs index (UBN) variables changed when including variables describing basic capabilities. The households in the fairly rich cluster, having assets as motorbikes and computers were described as modern. Those in the fairly poor cluster, having different degrees of food insecurity were labeled vulnerable. Poor and poorest clusters of households were traditional e.g. in using horses for transport. Results displayed a society transforming from traditional to a modern, where the forerunners were not the richest but educated, had more working members of household, fewer children and were food secure. Those lagging were the poor, traditional and food insecure. The approach and results may be useful for an improved understanding of poverty and to direct local interventions.
Open AccessAssessing the multiple dimensions of poverty. Data mining approaches to the 2004-14 Health and Demographic Surveillance System in Cuatro Santos, Nicaragua
We aimed to describe multiple dimensions of poverty according to the capability approach theory by applying data mining approaches to the Cuatro Santos Health and Demographic Surveillance databases, Nicaragua. Four municipalities in northern Nicaragua constitute the Cuatro Santos area, with 25,893 inhabitants in 5,966 households (2014). A local process analyzing poverty-related problems and prioritizing suggested actions, was initiated 1997 based on a community action plan 2002-2015. Priority interventions were school breakfasts, environmental protection, water and sanitation, preventive healthcare, home gardening, micro credits, technical training, stipends for university education, and the use of Internet. In 2004, a survey of basic health and demographic information was performed in the whole population followed by surveillance updates in 2007, 2009, and 2014 linking households and individuals by unique identifiers. Information included the house (floor, walls) and services (water, sanitation, electricity) as well as demographic data (birth, deaths, migration). Data on participation in interventions, on food security, household assets, and women’s self-rated health were collected in 2014. A K-means algorithm were used to cluster the household data (54 variables). The poverty ranking of household clusters using the unsatisfied basic needs index (UBN) variables changed when including variables describing basic capabilities. The households in the fairly rich cluster, having assets as motorbikes and computers were described as modern. Those in the fairly poor cluster, having different degrees of food insecurity were labeled vulnerable. Poor and poorest clusters of households were traditional e.g. in using horses for transport. Results displayed a society transforming from traditional to a modern, where the forerunners were not the richest but educated, had more working members of household, fewer children and were food secure. Those lagging were the poor, traditional and food insecure. The approach and results may be useful for an improved understanding of poverty and to direct local interventions.