Sahely Bhadra
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Sahely Bhadra has a total of 23 co-authors. Here are the top co-authors by number of shared publications:
- Juho Rousu
Department of Computer Science, Aalto University, Espoo, Finland and Helsinki Institute for Information Technology, University of Helsinki, Helsinki, Finland
4 shared publications - Deepak Padmanabhan
Queen’s University Belfast, Belfast, UK
3 shared publications Juho Rousu
Helsinki Institute for Information Technology HIIT, Department of Computer Science, Aalto University, Espoo, Finland
2 shared publications- S. Abilasha
Computer Science and Engineering Department, Indian Institute of Technology Palakkad, Palakkad, India
2 shared publications - Samuel Kaski
Department of Computer Science, Aalto University, Espoo, Finland
2 shared publications Anish Mathew
Indian Institute of Technology,Palakkad,India
1 shared publication
Latest Publications
- Open Access
- 1 July 2025
- Published by Oxford University Press (OUP) in Bioinformatics
- Vol. 41 (Supplement) , i58-i67
Efficient 3D kernels for molecular property prediction
Motivation: This paper addresses the challenge of incorporating 3-dimensional (3D) structural information in graph kernels for machine learning-based virtual screening, a crucial task in drug discovery. Existing kernels that capture 3D information often suffer from high computational complexity, which limits their scalability. Results: To overcome this, we propose the 3D chain motif graph kernel, which effectively integrates essential 3D structural properties—bond length, bond angle, and torsion angle—within the three-hop neighborhood of each atom in a molecule. In addition, we introduce a more computationally efficient variant, the 3D graph hopper kernel (3DGHK), which reduces the complexity from the state-of-the-art O(n6) (for the 3D pharmacophore kernel) to O(n2(m+log(n)+δ2+dT6)). Here, n is the number of nodes, T is the highest degree of the node, m is the number of edges, δ is the diameter of the graph, and d is the dimension of the attributes of the nodes. We conducted experiments on 21 datasets, demonstrating that 3DGHK outperforms other state-of-the-art 2D and 3D graph kernels, but it also surpasses deep learning models in classification accuracy, offering a powerful and scalable solution for virtual screening tasks. Availability and implementation: Our code is publicly available at https://github.com/SantaAnkit/Efficient-3D-kernels-for-molecular-property-prediction.git.
- 1 July 2025
- Published by Oxford University Press (OUP) in Bioinformatics
- Vol. 41 (Supplement) , i58-i67
Open AccessEfficient 3D kernels for molecular property prediction
Motivation: This paper addresses the challenge of incorporating 3-dimensional (3D) structural information in graph kernels for machine learning-based virtual screening, a crucial task in drug discovery. Existing kernels that capture 3D information often suffer from high computational complexity, which limits their scalability. Results: To overcome this, we propose the 3D chain motif graph kernel, which effectively integrates essential 3D structural properties—bond length, bond angle, and torsion angle—within the three-hop neighborhood of each atom in a molecule. In addition, we introduce a more computationally efficient variant, the 3D graph hopper kernel (3DGHK), which reduces the complexity from the state-of-the-art O(n6) (for the 3D pharmacophore kernel) to O(n2(m+log(n)+δ2+dT6)). Here, n is the number of nodes, T is the highest degree of the node, m is the number of edges, δ is the diameter of the graph, and d is the dimension of the attributes of the nodes. We conducted experiments on 21 datasets, demonstrating that 3DGHK outperforms other state-of-the-art 2D and 3D graph kernels, but it also surpasses deep learning models in classification accuracy, offering a powerful and scalable solution for virtual screening tasks. Availability and implementation: Our code is publicly available at https://github.com/SantaAnkit/Efficient-3D-kernels-for-molecular-property-prediction.git.
- 27 January 2025
- Published by Springer Nature in Machine Learning
- Vol. 114 (2) , 1-20
Warping resilient robust anomaly detection for multivariate time series
- 27 January 2025
- Published by Springer Nature in Machine Learning
- Vol. 114 (2) , 1-20
Warping resilient robust anomaly detection for multivariate time series
- Open Access
- 27 May 2024
- Published by Springer Nature in AI and Ethics
- Vol. 5 (2) , 1865-1870
Geo-political bias in fake news detection AI: the case of affect
There have been massive advances in AI technologies towards addressing the contemporary challenge of fake news identification. However, these technologies, as observed widely, have not had the same kind or depth in impact across global societies. In particular, the AI scholarship in fake news detection arguably has not been as beneficial or appropriate for Global South, bringing geo-political bias into the picture. While it is often natural to think of data bias as the potential reason for geo-political bias, other factors could be much more important in being more latent, and thus less visible. In this commentary, we investigate as to how the facet of affect, comprising emotions and sentiments, could be a potent vehicle for geo-political biases in AI. We highlight, through assembling and interpreting insights from literature, the overarching neglect of affect across methods for fake news detection AI, and how this could be a potentially important factor for geo-political bias within them. This exposition, we believe, also serves as a first effort in understanding how geo-political biases work within AI pipelines beyond the data collection stage.
- 27 May 2024
- Published by Springer Nature in AI and Ethics
- Vol. 5 (2) , 1865-1870
Open AccessGeo-political bias in fake news detection AI: the case of affect
There have been massive advances in AI technologies towards addressing the contemporary challenge of fake news identification. However, these technologies, as observed widely, have not had the same kind or depth in impact across global societies. In particular, the AI scholarship in fake news detection arguably has not been as beneficial or appropriate for Global South, bringing geo-political bias into the picture. While it is often natural to think of data bias as the potential reason for geo-political bias, other factors could be much more important in being more latent, and thus less visible. In this commentary, we investigate as to how the facet of affect, comprising emotions and sentiments, could be a potent vehicle for geo-political biases in AI. We highlight, through assembling and interpreting insights from literature, the overarching neglect of affect across methods for fake news detection AI, and how this could be a potentially important factor for geo-political bias within them. This exposition, we believe, also serves as a first effort in understanding how geo-political biases work within AI pipelines beyond the data collection stage.
- Open Access
- 5 March 2024
- Published by Elsevier in Natural Language Processing Journal
- Vol. 7
Understanding latent affective bias in large pre-trained neural language models
- 5 March 2024
- Published by Elsevier in Natural Language Processing Journal
- Vol. 7
Open AccessUnderstanding latent affective bias in large pre-trained neural language models
- 3 April 2023
- Published by Springer Nature
- p. 104-119
CDGCN: Conditional de novo Drug Generative Model Using Graph Convolution Networks
- 3 April 2023
- Published by Springer Nature
- p. 104-119
CDGCN: Conditional de novo Drug Generative Model Using Graph Convolution Networks
- Open Access
Warping Resilient Scalable Anomaly Detection in Time Series
Time series data is ubiquitous in the real-world problems across various domains including healthcare, social media, and crime surveillance. Detecting anomalies, or irregular and rare events, in time series data, can enable us to find abnormal events in any natural phenomena, which may require special treatment. Moreover, labeled instances of anomaly are hard to get in time series data. On the other hand, time series data, due to its nature, often exhibits localized expansions and compressions in the time dimension which is called warping. These two challenges make it hard to detect anomalies in time series as often such warpings could get detected as anomalies erroneously. Our objective is to build an anomaly detection model that is robust to such warping variations. In this paper, we propose a novel unsupervised time series anomaly detection method, WaRTEm-AD, that operates in two stages. Within the key stage of representation learning, we employ data augmentation through bespoke time series operators which are passed through a twin autoencoder architecture to learn warping-robust representations for time series data. Second, adaptations of state-of-the-art anomaly detection methods are employed on the learnt representations to identify anomalies. We will illustrate that WaRTEm-AD is designed to detect two types of time series anomalies: point and sequence anomalies. We compare WaRTEm-AD with the state-of-the-art baselines and establish the effectiveness of our method both in terms of anomaly detection performance and computational efficiency.
Open AccessWarping Resilient Scalable Anomaly Detection in Time Series
Time series data is ubiquitous in the real-world problems across various domains including healthcare, social media, and crime surveillance. Detecting anomalies, or irregular and rare events, in time series data, can enable us to find abnormal events in any natural phenomena, which may require special treatment. Moreover, labeled instances of anomaly are hard to get in time series data. On the other hand, time series data, due to its nature, often exhibits localized expansions and compressions in the time dimension which is called warping. These two challenges make it hard to detect anomalies in time series as often such warpings could get detected as anomalies erroneously. Our objective is to build an anomaly detection model that is robust to such warping variations. In this paper, we propose a novel unsupervised time series anomaly detection method, WaRTEm-AD, that operates in two stages. Within the key stage of representation learning, we employ data augmentation through bespoke time series operators which are passed through a twin autoencoder architecture to learn warping-robust representations for time series data. Second, adaptations of state-of-the-art anomaly detection methods are employed on the learnt representations to identify anomalies. We will illustrate that WaRTEm-AD is designed to detect two types of time series anomalies: point and sequence anomalies. We compare WaRTEm-AD with the state-of-the-art baselines and establish the effectiveness of our method both in terms of anomaly detection performance and computational efficiency.
- Open Access
- 1 December 2019
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-5
Time Series Representation Learning Applications for Power Analytics
The uptake of solar power generation is on the rise. This necessitates more research into developing data-driven intelligent methods that can perform effective analytics over power generation data to inform strategies to improve solar power generation systems. In this paper, we consider the utility of time series representation learning for analytics over power generation data. WaRTEm, a representation learning method that focuses on learning time series representations that are invariant to local phase shifts, is the focus of our investigations in this paper. We identify two metadata attributes for power generation sequences, month and CellID, as attributes that embed useful notions of semantic similarity between time series sequences. We evaluate the effectiveness of WaRTEm representations, as against using the raw time series sequences, in alignment to the month and CellID labellings, using accuracy over 1NN retrieval as an evaluation framework. Through empirical evaluations, we identify that WaRTEm embeddings are consistently able to achieve better representations when evaluated on 1NN accuracy. We also identify some features of WaRTEm that are more suited for time series representation learning, which provides promising directions for future work.
- 1 December 2019
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-5
Open AccessTime Series Representation Learning Applications for Power Analytics
The uptake of solar power generation is on the rise. This necessitates more research into developing data-driven intelligent methods that can perform effective analytics over power generation data to inform strategies to improve solar power generation systems. In this paper, we consider the utility of time series representation learning for analytics over power generation data. WaRTEm, a representation learning method that focuses on learning time series representations that are invariant to local phase shifts, is the focus of our investigations in this paper. We identify two metadata attributes for power generation sequences, month and CellID, as attributes that embed useful notions of semantic similarity between time series sequences. We evaluate the effectiveness of WaRTEm representations, as against using the raw time series sequences, in alignment to the month and CellID labellings, using accuracy over 1NN retrieval as an evaluation framework. Through empirical evaluations, we identify that WaRTEm embeddings are consistently able to achieve better representations when evaluated on 1NN accuracy. We also identify some features of WaRTEm that are more suited for time series representation learning, which provides promising directions for future work.
- 21 July 2018
- Published by Springer Nature
- Vol. 1807, 141-161
Analysis of Fluxomic Experiments with Principal Metabolic Flux Mode Analysis
In the analysis of metabolism, two distinct and complementary approaches are frequently used: Principal component analysis (PCA) and stoichiometric flux analysis. PCA is able to capture the main modes of variability in a set of experiments and does not make many prior assumptions about the data, but does not inherently take into account the flux mode structure of metabolism. Stoichiometric flux analysis methods, such as Flux Balance Analysis (FBA) and Elementary Mode Analysis, on the other hand, are able to capture the metabolic flux modes, however, they are primarily designed for the analysis of single samples at a time, and assume the stoichiometric steady state of the metabolic network. We will discuss a new methodology for the analysis of metabolism, called Principal Metabolic Flux Mode Analysis (PMFA), which marries the PCA and stoichiometric flux analysis approaches in an elegant regularized optimization framework. In short, the method incorporates a variance maximization objective form PCA coupled with a stoichiometric regularizer, which penalizes projections that are far from any flux modes of the network. For interpretability, we also discuss a sparse variant of PMFA that favors flux modes that contain a small number of reactions. PMFA has several benefits: (1) it can be applied to large metabolic network in efficient way as PMFA does not enumerate elementary modes, (2) the method is more robust to the steady-state violations than competing approaches, and (3) can compactly capture the variation in the data by a few factors. This chapter will describe the detailed steps how to do the above task on experimental data from fluxomic and gene expression measurements.
- 21 July 2018
- Published by Springer Nature
- Vol. 1807, 141-161
Analysis of Fluxomic Experiments with Principal Metabolic Flux Mode Analysis
In the analysis of metabolism, two distinct and complementary approaches are frequently used: Principal component analysis (PCA) and stoichiometric flux analysis. PCA is able to capture the main modes of variability in a set of experiments and does not make many prior assumptions about the data, but does not inherently take into account the flux mode structure of metabolism. Stoichiometric flux analysis methods, such as Flux Balance Analysis (FBA) and Elementary Mode Analysis, on the other hand, are able to capture the metabolic flux modes, however, they are primarily designed for the analysis of single samples at a time, and assume the stoichiometric steady state of the metabolic network. We will discuss a new methodology for the analysis of metabolism, called Principal Metabolic Flux Mode Analysis (PMFA), which marries the PCA and stoichiometric flux analysis approaches in an elegant regularized optimization framework. In short, the method incorporates a variance maximization objective form PCA coupled with a stoichiometric regularizer, which penalizes projections that are far from any flux modes of the network. For interpretability, we also discuss a sparse variant of PMFA that favors flux modes that contain a small number of reactions. PMFA has several benefits: (1) it can be applied to large metabolic network in efficient way as PMFA does not enumerate elementary modes, (2) the method is more robust to the steady-state violations than competing approaches, and (3) can compactly capture the variation in the data by a few factors. This chapter will describe the detailed steps how to do the above task on experimental data from fluxomic and gene expression measurements.
- Open Access
- 6 February 2018
- Published by Oxford University Press (OUP) in Bioinformatics
- Vol. 34 (14) , 2409-2417
Principal metabolic flux mode analysis
In the analysis of metabolism, two distinct and complementary approaches are frequently used: Principal component analysis (PCA) and stoichiometric flux analysis. PCA is able to capture the main modes of variability in a set of experiments and does not make many prior assumptions about the data, but does not inherently take into account the flux mode structure of metabolism. Stoichiometric flux analysis methods, such as Flux Balance Analysis (FBA) and Elementary Mode Analysis, on the other hand, are able to capture the metabolic flux modes, however, they are primarily designed for the analysis of single samples at a time, and not best suited for exploratory analysis on a large sets of samples. We propose a new methodology for the analysis of metabolism, called Principal Metabolic Flux Mode Analysis (PMFA), which marries the PCA and stoichiometric flux analysis approaches in an elegant regularized optimization framework. In short, the method incorporates a variance maximization objective form PCA coupled with a stoichiometric regularizer, which penalizes projections that are far from any flux modes of the network. For interpretability, we also introduce a sparse variant of PMFA that favours flux modes that contain a small number of reactions. Our experiments demonstrate the versatility and capabilities of our methodology. The proposed method can be applied to genome-scale metabolic network in efficient way as PMFA does not enumerate elementary modes. In addition, the method is more robust on out-of-steady steady-state experimental data than competing flux mode analysis approaches. Matlab software for PMFA and SPMFA and dataset used for experiments are available in https://github.com/aalto-ics-kepaco/PMFA. Supplementary data are available at Bioinformatics online.
- 6 February 2018
- Published by Oxford University Press (OUP) in Bioinformatics
- Vol. 34 (14) , 2409-2417
Open AccessPrincipal metabolic flux mode analysis
In the analysis of metabolism, two distinct and complementary approaches are frequently used: Principal component analysis (PCA) and stoichiometric flux analysis. PCA is able to capture the main modes of variability in a set of experiments and does not make many prior assumptions about the data, but does not inherently take into account the flux mode structure of metabolism. Stoichiometric flux analysis methods, such as Flux Balance Analysis (FBA) and Elementary Mode Analysis, on the other hand, are able to capture the metabolic flux modes, however, they are primarily designed for the analysis of single samples at a time, and not best suited for exploratory analysis on a large sets of samples. We propose a new methodology for the analysis of metabolism, called Principal Metabolic Flux Mode Analysis (PMFA), which marries the PCA and stoichiometric flux analysis approaches in an elegant regularized optimization framework. In short, the method incorporates a variance maximization objective form PCA coupled with a stoichiometric regularizer, which penalizes projections that are far from any flux modes of the network. For interpretability, we also introduce a sparse variant of PMFA that favours flux modes that contain a small number of reactions. Our experiments demonstrate the versatility and capabilities of our methodology. The proposed method can be applied to genome-scale metabolic network in efficient way as PMFA does not enumerate elementary modes. In addition, the method is more robust on out-of-steady steady-state experimental data than competing flux mode analysis approaches. Matlab software for PMFA and SPMFA and dataset used for experiments are available in https://github.com/aalto-ics-kepaco/PMFA. Supplementary data are available at Bioinformatics online.