Dataset Information

Topic2features: a novel framework to classify noisy and sparse textual data using LDA topic distributions.

ABSTRACT: In supervised machine learning, specifically in classification tasks, selecting and analyzing the feature vector to achieve better results is one of the most important tasks. Traditional methods such as comparing the features' cosine similarity and exploring the datasets manually to check which feature vector is suitable is relatively time consuming. Many classification tasks failed to achieve better classification results because of poor feature vector selection and sparseness of data. In this paper, we proposed a novel framework, topic2features (T2F), to deal with short and sparse data using the topic distributions of hidden topics gathered from dataset and converting into feature vectors to build supervised classifier. For this we leveraged the unsupervised topic modelling LDA (latent dirichlet allocation) approach to retrieve the topic distributions employed in supervised learning algorithms. We made use of labelled data and topic distributions of hidden topics that were generated from that data. We explored how the representation based on topics affect the classification performance by applying supervised classification algorithms. Additionally, we did careful evaluation on two types of datasets and compared them with baseline approaches without topic distributions and other comparable methods. The results show that our framework performs significantly better in terms of classification performance compared to the baseline(without T2F) approaches and also yields improvement in terms of F1 score compared to other compared approaches.

SUBMITTER: Wahid JA

PROVIDER: S-EPMC8372003 | biostudies-literature |

REPOSITORIES: biostudies-literature

ACCESS DATA

Similar Datasets

Project description:To perform parametric identification of mathematical models of biological events, experimental data are rare to be sufficient to estimate target behaviors produced by complex non-linear systems. We performed parameter fitting to a cell cycle model with experimental data as an in silico experiment. We calibrated model parameters with the generalized least squares method with randomized initial values and checked local and global sensitivity of the model. Sensitivity analyses showed that parameter optimization induced less sensitivity except for those related to the metabolism of the transcription factors c-Myc and E2F, which are required to overcome a restriction point (R-point). We performed bifurcation analyses with the optimized parameters and found the bimodality was lost. This result suggests that accumulation of c-Myc and E2F induced dysfunction of R-point. We performed a second parameter optimization based on the results of sensitivity analyses and incorporating additional derived from recent in vivo data. This optimization returned the bimodal characteristics of the model with a narrower range of hysteresis than the original. This result suggests that the optimized model can more easily go through R-point and come back to the gap phase after once having overcome it. Two parameter space analyses showed metabolism of c-Myc is transformed as it can allow cell bimodal behavior with weak stimuli of growth factors. This result is compatible with the character of the cell line used in our experiments. At the same time, Rb, an inhibitor of E2F, can allow cell bimodal behavior with only a limited range of stimuli when it is activated, but with a wider range of stimuli when it is inactive. These results provide two insights; biologically, the two transcription factors play an essential role in malignant cells to overcome R-point with weaker growth factor stimuli, and theoretically, sparse time-course data can be used to change a model to a biologically expected state.

Project description:Tuning curves are the functions that relate the responses of sensory neurons to various values within one continuous stimulus dimension (such as the orientation of a bar in the visual domain or the frequency of a tone in the auditory domain). They are commonly determined by fitting a model e.g. a Gaussian or other bell-shaped curves to the measured responses to a small subset of discrete stimuli in the relevant dimension. However, as neuronal responses are irregular and experimental measurements noisy, it is often difficult to determine reliably the appropriate model from the data. We illustrate this general problem by fitting diverse models to representative recordings from area MT in rhesus monkey visual cortex during multiple attentional tasks involving complex composite stimuli. We find that all models can be well-fitted, that the best model generally varies between neurons and that statistical comparisons between neuronal responses across different experimental conditions are affected quantitatively and qualitatively by specific model choices. As a robust alternative to an often arbitrary model selection, we introduce a model-free approach, in which features of interest are extracted directly from the measured response data without the need of fitting any model. In our attentional datasets, we demonstrate that data-driven methods provide descriptions of tuning curve features such as preferred stimulus direction or attentional gain modulations which are in agreement with fit-based approaches when a good fit exists. Furthermore, these methods naturally extend to the frequent cases of uncertain model selection. We show that model-free approaches can identify attentional modulation patterns, such as general alterations of the irregular shape of tuning curves, which cannot be captured by fitting stereotyped conventional models. Finally, by comparing datasets across different conditions, we demonstrate effects of attention that are cell- and even stimulus-specific. Based on these proofs-of-concept, we conclude that our data-driven methods can reliably extract relevant tuning information from neuronal recordings, including cells whose seemingly haphazard response curves defy conventional fitting approaches.

Project description:BackgroundHigh throughput data are complex and methods that reveal structure underlying the data are most useful. Principal component analysis, frequently implemented as a singular value decomposition, is a popular technique in this respect. Nowadays often the challenge is to reveal structure in several sources of information (e.g., transcriptomics, proteomics) that are available for the same biological entities under study. Simultaneous component methods are most promising in this respect. However, the interpretation of the principal and simultaneous components is often daunting because contributions of each of the biomolecules (transcripts, proteins) have to be taken into account.ResultsWe propose a sparse simultaneous component method that makes many of the parameters redundant by shrinking them to zero. It includes principal component analysis, sparse principal component analysis, and ordinary simultaneous component analysis as special cases. Several penalties can be tuned that account in different ways for the block structure present in the integrated data. This yields known sparse approaches as the lasso, the ridge penalty, the elastic net, the group lasso, sparse group lasso, and elitist lasso. In addition, the algorithmic results can be easily transposed to the context of regression. Metabolomics data obtained with two measurement platforms for the same set of Escherichia coli samples are used to illustrate the proposed methodology and the properties of different penalties with respect to sparseness across and within data blocks.ConclusionSparse simultaneous component analysis is a useful method for data integration: First, simultaneous analyses of multiple blocks offer advantages over sequential and separate analyses and second, interpretation of the results is highly facilitated by their sparseness. The approach offered is flexible and allows to take the block structure in different ways into account. As such, structures can be found that are exclusively tied to one data platform (group lasso approach) as well as structures that involve all data platforms (Elitist lasso approach).AvailabilityThe additional file contains a MATLAB implementation of the sparse simultaneous component method.

Dataset Information

Topic2features: a novel framework to classify noisy and sparse textual data using LDA topic distributions.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets