Cell Painting is an established community-based microscopy-assay platform that provides high-throughput, high-content data for biological readouts. In November 2022, the JUMP-Cell Painting Consortium released the largest publicly available Cell Painting dataset with CellProfiler features, comprising more than 2 billion cell images. This dataset is designed for predicting the activity and toxicity of 115k drug compounds, with the aim to make cell images as computable as genomes and transcriptomes. In this context, our paper introduces a scalable and computationally efficient data analytics workflow created to meet the needs of researchers. This data-driven workflow facilitates the comparison of drug treatment effects through significant and biologically relevant insights. The workflow consists of two parts: first, the Equivalence score (Eq. score), a straightforward yet sophisticated metric highlighting relevant deviations from negative controls based on cell image morphology; second, the scalability of the workflow, by utilizing the Eq. scores on a large scale to predict and classify the subtle morphological changes in cell image profiles. By doing so, we show classification improvements compared to using the raw CellProfiler features on the CPJUMP1-pilot dataset on three types of perturbations.
We hope that our workflow’s contributions will enhance drug screening efficiency and streamline the drug development process. As this process is resource-intensive, every incremental improvement is valuable. Through our collective efforts in advancing the understanding of high-throughput image-based data, we aim to reduce both the time and cost of developing new, life-saving treatments.
Support vector machine (SVM) is known for its good generalization performance and wide application in various fields. Despite its success, the learning efficiency of SVM decreases significantly originating from the assumption that the number of training samples increases rapidly. Consequently, the traditional SVM with standard optimization methods faces challenges such as excessive memory requirements and slow training speed, especially for large-scale training sets. To address this issue, this paper draws inspiration from the fuzzy support vector machine (FSVM). Considering that each sample has varying contributions to the decision plane, we propose an effective SVM sample reduction method based on the fuzzy membership function (FMF). This method uses FMF to calculate the fuzzy membership of each training sample. Training samples with low fuzzy memberships are then deleted. Specifically, we propose SVM sample reduction algorithms based on class center distance, kernel target alignment, centered kernel alignment, slack factor, entropy, and bilateral weighted FMF, respectively. Comprehensive experiments on UCI and KEEL datasets demonstrate that our proposed algorithms outperform other comparative methods in terms of accuracy, F-measure, and hinge-loss measures.
Drug discovery is the process by which a drug is discovered. Drug-target interactions prediction is a major part of drug discovery. Unfortunately, producing new drugs is time-consuming and expensive; Because it requires a lot of human and laboratory resources. Recently, predictions have been made using computational methods to solve these problems and prevent blindly examining all interactions. Various experiences using computational methods show that no single algorithm can be suitable for all applications; Hence, ensemble learning is expressed. Although various ensemble methods have been proposed, it is still not easy to find a suitable ensemble method for a particular dataset. In general, the existing algorithms in aggregation and combination method are selected manually based on experience. Reinforcement learning can be one way to meet this challenge. High-dimensional feature space and class imbalance are among the challenges of drug-target interactions prediction. This paper proposes HEnsem_DTIs, a heterogeneous ensemble model, for predicting drug-target interactions using dimensionality reduction and concepts of recommender systems to address these challenges. HEnsem_DTIs is configured with reinforcement learning. Dimensionality reduction is applied to handle the challenge of high-dimensional feature space and recommender systems to improve under-sampling and solve the class imbalance challenge. Six datasets are used to evaluate the proposed model; Results of the evaluation on datasets show that HEnsem_DTIs works better than other models in this field. Results of evaluation of the proposed model on the first dataset using 10-fold cross-validation experiments show the amount of sensitivity 0.896, specificity 0.954, GM 0.924, AUC 0.930 and AUPR 0.935.
In classification problems, many models with superior performance fail to provide confidence estimates or intervals for each prediction. This lack of reliability poses risks in real-world applications, making these models difficult to trust. Conformal prediction, as distribution-free and model-free approaches with finite-sample coverage guarantee, have recently been widely used to construct prediction sets for classification models. However, traditional conformal prediction methods only produce set-valued results without specifying a definitive predicted class. Particularly in complex settings, these methods fail to assist models in effectively addressing challenges such as high dimensionality, resulting in ambiguous prediction sets with low statistical efficiency, i.e. the prediction sets contain many false classes. In this study, a novel Ensemble Conformal Prediction algorithm based on Random Projection and a designed voting strategy, RPECP, is developed to tackle these challenges. Initially, a procedure for selecting the approximately oracle random projections and classifiers is executed to best leverage the internal information and structure of the data. Subsequently, based on the approximately oracle random projections and underlying classifiers, conformal prediction is performed on new test samples in a lower-dimensional space, resulting in multiple independent prediction sets. Finally, an accurate predicted class and a precise prediction set with high coverage and statistical efficiency are produced through a designed voting strategy. Compared to several base classifiers, RPECP obtain higher classification accuracy; against other conformal prediction algorithms, it achieves less ambiguous prediction sets with fewer false classes while guaranteeing high coverage. For illustration, this paper demonstrates RPECP's superiority over other methods in four cases: two high-dimensional settings and two real-world datasets.
Dimensionality reduction is an essential step in the processing of analytical chemistry data. When this reduction is carried out by variable selection, it can enable the identification of biochemical pathways. CovSel has been developed to meet this requirement, through a parsimonious selection of non-redundant variables. This article presents the g-CovSel method, which modifies the CovSel algorithm to produce highly complementary groups containing highly correlated variables. This modification requires the theoretical definition of the groups' construction and of the deflation of the data with respect to the selected groups. Two applications, on two extreme case studies, are presented. The first, based on near-infrared spectra related to four chemicals, demonstrates the relevance of the selected groups and the method's ability to handle highly correlated variables. The second, based on genomic data, demonstrates the method's ability to handle very highly multivariate data. Most of the groups formed can be interpreted from a functional point of view, making g-CovSel a tool of choice for biomarker identification in omics. Further work will be carried out to generalize g-CovSel to multi-block and multi-way data.
Guar gum is a non-ionic polysaccharide found in abundance in nature. It may be used as a thickening agent, stabilizer, or emulsifier in pharmaceutical formulations, food products, or cosmetics. Its ability to form viscous solutions makes it useful in drug delivery systems, controlled release formulations, and as a matrix for oral drug delivery. The investigation of chemical structures through graph invariants is of great concern. Topological descriptors are numerical numbers associated with the molecular structure and have the ability to predict certain physical and chemical properties of the underlying structure. In this paper, we have calculated the harmonic index, the inverse sum indeg index, the third Zagreb index, the Hyper Zagreb index, the sigma index, the reformulated first Zagreb index, the reformulated multiplicative first Zagreb index, the Harmonic–arithmetic index, and the Atom Bond sum connectivity indices of guar gum and its chemical derivatives. Finally, the chemical applicability of these topological descriptors is checked for different carbohydrates (monosaccharides, disaccharides, and polysaccharides) by using straight-line, parabolic and logarithmic regression models. It has been observed that these topological descriptors are useful to predict two physical properties, namely density and molecular weight.
Definitive screening design (DSD) has become a widely used type of Design of Experiments for chemical, pharmaceutical and biopharmaceutical processes and product development due to its optimization properties with an estimation of main, interaction, and squared variable effects with a minimum number of experiments. These high dimensional DOEs with more variables than samples, and with partly correlated variables, make the statistical interpretation frequently challenging. The purpose of the study was to test bootstrap PLSR using a heredity procedure to select the variable subset to be finally evaluated by MLR. The heredity selection was used on bootstrap T values given by original PLSR coefficients (B) divided on the bootstrap estimated standard deviation. The investigated fractional weighted and non-parametric bootstrap PLSR resulted in same variable selection outcome and final models in this study.
A simulation study with 7 main variables and 12 tested literature real data DSDs with 4, 5, 7 and 8 main variables showed improved model performance for small and particularly for large DSDs for the bootstrap PLSR MLR methods compared to two common DSD reference methods; DSD fit definitive screening and AICc forward stepwise regression (AICc FSR). Variable selection accuracy and predictive ability were significantly improved by the investigated method in 6 out of 13 DSDs compared to the best model from either of the two reference methods. The remaining 7 DSDs gave the same model as best reference model. Strong heredity was found to provide the best models for all real data in this study. The use of the heredity procedure on the percent non-zero SVEM FSR variable effects followed by MLR showed promising results. AICc Lasso regression was among other methods partially tested and was found to set almost all variables to zero effect when tested on three large minimum DSDs. While the DSD fit definitive screening method may often be the first choice for DSD, the heredity bootstrap PLSR MLR and heredity SVEM FSR MLR may be alternative methods to improve the variable selection and model precision.
The determination of total nicotine, total sugar, reducing sugar and total nitrogen contents in tobacco is of great significance to tobacco quality evaluation and formulation design. To quickly detect the content of 4 components of tobacco, using near-infrared (NIR) and mid-infrared (MIR) spectral data from 129 solid samples of tobacco powder provided by Shanghai Tobacco Group Co., Ltd., Two NIR-MIR spectral fusion techniques are studied, that is, fusion technology 1 is to establish a model by fusing feature variables after variable selection of each spectrum. The fusion technology 2 is to first fuse the NIR-MIR spectral data and then select the variables to establish the model. Both fusion technologies use successive projections algorithm (SPA), competitive adaptive reweighted sampling (CARS), backward interval PLS (biPLS), forward interval PLS (fiPLS), synergy interval PLS (siPLS), and interval interaction moving window partial least squares (iMWPLS) algorithms to filter wavelength variables. The results showed that for total nicotine and total sugar, the PLSR model established by fusion technology method 2 combined with iMWPLS algorithm is the best, and its RMSEP decreases from 0.2314 to 1.3225 to 0.0821 and 0.8079 respectively compared with the full spectrum fusion method, which is superior to the single NIR and MIR models and NIR-MIR fusion technology 1. For reducing sugars, the simple full-spectrum fusion model has the best analytical ability and the lowest RMSEP, which is superior to the single NIR-MIR models and all models established by two spectral fusion techniques combined with six wavelength selection algorithms. For total nitrogen, the prediction effect of fusion technology 1 combined with iMWPLS algorithm model was significantly improved compared with single NIR and MIR models and NIR-MIR fusion technology 2, and its RMSEP was 0.0634. The results showed that the two NIR-MIR spectral fusion techniques made full use of the complementary information provided by NIR and MIR spectroscopy, and successfully applied them to the rapid detection of total nicotine, total sugar, reducing sugar and total nitrogen content in tobacco, which provided a new method and idea for the rapid detection of tobacco components.