In today's scientific landscape, research software has evolved from being a supportive tool to becoming a fundamental driver of discovery, particularly in life sciences. Beyond its roots in software engineering, research software now plays a crucial role in facilitating efficient data analysis and enabling the exploration of complex natural phenomena. The advancements in simulations and modeling through research software have significantly accelerated the pace of scientific research while reducing associated costs. This growing reliance underscores the importance of software in ensuring reproducibility - a cornerstone of scientific rigor and trustworthiness. Although verifying reproducibility presents challenges, well-developed and openly accessible research software enhances transparency and aids in the early detection of errors. Although verifying reproducibility can be challenging, well-developed and accessible research software improves transparency and facilitates error detection. This mini-review examines the characteristics of research software and summarizes the key events that have shaped its development, alongside changes in requirements and guidelines. Moreover, we propose two additional principles - reviewability and supportability - complementing the widely accepted FAIR principles (Findability, Accessibility, Interoperability, and Reusability). These new principles aim to improve the efficiency and effectiveness of software evaluation during the peer review process. Through this review, we aim to assist scientists, especially those without extensive software development expertise, in understanding best practices for developing research software and the underlying motivations driving these practices.

Predicting the transcriptional response of chemical perturbations is crucial to understanding gene function and developing drug candidates, promising a streamlined drug development process. Single-cell sequencing has provided an ideal data basis for training machine learning models for this task. Recent advances in deep learning have led to significant improvements in predictions of chemical as well as genetic perturbations at the single cell level. Experiments have shown that different cell types exhibit distinct transcriptional patterns and responses to perturbation. This poses a fundamental problem for predicting transcriptional responses of drugs or cell types outside the training data. Accordingly, existing methods lack cell-type-specific modeling or do not explicitly provide an interpretable mechanism for the gene features. In this study, we introduce a novel approach that employs a network representation of various cell types as an inductive bias, improving prediction performance in scenarios with limited data while acknowledging cellular differences. We applied our framework to four small-scale single-cell perturbation datasets and one large-scale screening experiment, demonstrating that this representation can inherently generalize to previously unseen cell types. Furthermore, our method outperforms the state-of-the-art methods in predicting the post-perturbation response in unobserved cell types. ### Competing Interest Statement The authors have declared no competing interest.

Spatial Transcriptomics (ST) allows deep characterization of the 2D organization of expression data within tissue slices. The ST technology provides a tissue contextualization of deep single-cell profiles. Recently, numerous computational and machine learning methods have addressed challenges such as data quality, augmentation, annotation, and the development of integrative platforms for data analysis. In contrast, here we ask whether unseen spatial transcriptomics data can be predicted and if we can interpolate novel transcriptomic slices. To this end, we adopt a denoising diffusion probabilistic-based model (DDPM) to demonstrate the learning of generative ST models for several tissues. Furthermore, our generative diffusion model interpolates (predicts) unseen slices located “between” the collected finite number of ST slices. This methodology set the stage for learning predictive deep 3D models of tissues from a finite number of spatial transcriptomics slices, thus harboring the advent of AI-augmented spatial transcriptomics.### Competing Interest StatementThe authors have declared no competing interest.

The unproductive time is very high due to drill string jamming. So the main objective of this research is to determine the influences of the parameters in the accidents of stuck pipes using the construction of XGBoost models. To develop the model, drilling parameters are taken from daily drilling reports of the construction of 30 wells in the West Qurna field, Iraq. The data includes well drilling reports from 2013 to 2020. The results show that the factors such as Measured depth (MD), Rate of penetration (ROP), 10-sec gel (GEL1), Plastic Viscosity (PV), Mud weight (MW), Yield point (YP) contribute positively to the model predictions. In contrast, the factors such as Flow rate (FR), Rotation per minute (RPM), Filtrate (API/HPHT-FILTR and Bottom hole assembly (BHA) length has a negative contribution to the final model predictions. This research concluded that pipe sticking in the borehole is primarily due to inclination, penetration rate, and flow rate. This study is useful in the drilling of any field.

With the emergence of single-cell foundation models, an important question arises: how do these models perform when trained on datasets having an imbalance in cell type distribution due to rare cell types or biased sampling? We benchmark three foundation models, scGPT, scBERT, and Geneformer, using skewed single-cell cell-type distribution for cell-type annotation. While all models had reduced performance when challenged with rare cell types, scGPT and scBERT, performed better than Geneformer. Notably, in contrast to scGPT and scBERT, Geneformer uses ordinal positions of the tokenized genes rather than actual raw gene expression values. To mitigate the effect of a skewed distribution, we find that random oversampling, but not random undersampling, improved the performance for all three foundation models. Finally, scGPT, using FlashAttention, has the fastest computational speed, whereas scBERT is more memory-efficient. We conclude that tokenization and data representation are essential areas of research, and new strategies are needed to mitigate the effects of imbalanced learning in single-cell foundation models. Code and data for reproducibility are available at https://github.com/SabbaghCodes/ImbalancedLearningForSingleCellFoundationModels .

In this work, we explore Parameter-Efficient-Learning (PEL) techniques to repurpose a General-Purpose-Speech (GSM) model for Arabic dialect identification (ADI). Specifically, we investigate different setups to incorporate trainable features into a multi-layer encoder-decoder GSM formulation under frozen pre-trained settings. Our architecture includes residual adapter and model reprogramming (input-prompting). We design a token-level label mapping to condition the GSM for Arabic Dialect Identification (ADI). This is challenging due to the high variation in vocabulary and pronunciation among the numerous regional dialects. We achieve new state-of-the-art accuracy on the ADI-17 dataset by vanilla fine-tuning. We further reduce the training budgets with the PEL method, which performs within 1.86% accuracy to fine-tuning using only 2.5% of (extra) network trainable parameters. Our study demonstrates how to identify Arabic dialects using a small dataset and limited computation with open source code and pre-trained models.

We introduce a new cross-modal fusion technique designed for generative error correction in automatic speech recognition (ASR). Our methodology leverages both acoustic information and external linguistic representations to generate accurate speech transcription contexts. This marks a step towards a fresh paradigm in generative error correction within the realm of n-best hypotheses. Unlike the existing ranking-based rescoring methods, our approach adeptly uses distinct initialization techniques and parameter-efficient algorithms to boost ASR performance derived from pre-trained speech and text models. Through evaluation across diverse ASR datasets, we evaluate the stability and reproducibility of our fusion technique, demonstrating its improved word error rate relative (WERR) performance in comparison to n-best hypotheses by relatively 37.66%. To encourage future research, we have made our code and pre-trained models open source at https://github.com/Srijith-rkr/Whispering-LLaMA.

Predicting drug-target interactions is a tremendous challenge for drug development and lead optimization. Recent advances include training algorithms to learn drug-target interactions from data and molecular simulations. Here we utilize Evolutionary Scale Modeling (ESM-2) models to establish a Transformer protein language model for drug-target interaction predictions. Our architecture, LEP- AD, combines pre-trained ESM-2 and Transformer-GCN models predicting bind-ing affinity values. We report new best-in-class state-of-the-art results compared to competing methods such as SimBoost, DeepCPI, Attention-DTA, GraphDTA, and more using multiple datasets, including Davis, KIBA, DTC, Metz, ToxCast, and STITCH. Finally, we find that a pre-trained model with embedding of proteins (the LED-AD) outperforms a model using an explicit alpha-fold 3D representation of proteins (e.g., LEP-AD supervised by Alphafold). The LEP-AD model scales favorably in performance with the size of training data. Code available at https://github.com/adaga06/LEP-AD.

Background: Single-cell multi-omics technologies allow a profound system-level biology understanding of cells and tissues. However, an integrative and possibly systems-based analysis capturing the different modalities is challenging. In response, bioinformatics and machine learning methodologies are being developed for multi-omics single-cell analysis. It is unclear whether current tools can address the dual aspect of modality integration and prediction across modalities without requiring extensive parameter fine-tuning. Methods: We designed LIBRA, a neural network based framework, to learn translation between paired multi-omics profiles so that a shared latent space is constructed. Additionally, we implemented a variation, aLIBRA, that allows automatic fine-tuning by identifying parameter combinations that optimize both the integrative and predictive tasks. All model parameters and evaluation metrics are made available to users with minimal user iteration. Furthermore, aLIBRA allows experienced users to implement custom configurations. The LIBRA toolbox is freely available as R and Python libraries at GitHub (TranslationalBioinformaticsUnit/LIBRA). Results: LIBRA was evaluated in eight multi-omic single-cell data-sets, including three combinations of omics. We observed that LIBRA is a state-of-the-art tool when evaluating the ability to increase cell-type (clustering) resolution in the integrated latent space. Furthermore, when assessing the predictive power across data modalities, such as predictive chromatin accessibility from gene expression, LIBRA outperforms existing tools. As expected, adaptive parameter optimization (aLIBRA) significantly boosted the performance of learning predictive models from paired data-sets. Conclusion: LIBRA is a versatile tool that performs competitively in both “integration” and “prediction” tasks based on single-cell multi-omics data. LIBRA is a data-driven robust platform that includes an adaptive learning scheme.

The rise of single-cell genomics is an attractive opportunity for data-hungry machine learning algorithms. The scBERT method, inspired by the success of BERT (‘bidirectional encoder representations from transformers’) in natural language processing, was recently introduced by Yang et al. as a data-driven tool to annotate cell types in single-cell genomics data. Analogous to contextual embedding in BERT, scBERT leverages pretraining and self-attention mechanisms to learn the ‘transcriptional grammar’ of cells. Here we investigate the reusability beyond the original datasets, assessing the generalizability of natural language techniques in single-cell genomics. The degree of imbalance in the cell-type distribution substantially influences the performance of scBERT. Anticipating an increased utilization of transformers, we highlight the necessity to consider data distribution carefully and introduce a subsampling technique to mitigate the influence of an imbalanced distribution. Our analysis serves as a stepping stone towards understanding and optimizing the use of transformers in single-cell genomics.

The study of cellular decision-making can be approached comprehensively using multimodal single-cell omics technology. Recent advances in multimodal single-cell technology have enabled simultaneous profiling of more than one modality from the same cell, providing more significant insights into cell characteristics. However, learning the joint representation of multimodal single-cell data is challenging due to batch effects. Here we present a novel method, scJVAE (single-cell Joint Variational AutoEncoder), for batch effect removal and joint representation of multimodal single-cell data. The scJVAE integrates and learns joint embedding of paired scRNA-seq and scATAC-seq data modalities. We evaluate and demonstrate the ability of scJVAE to remove batch effects using various datasets with paired gene expression and open chromatin. We also consider scJVAE for downstream analysis, such as lower dimensional representation, cell-type clustering, and time and memory requirement. We find scJVAE a robust and scalable method outperforming existing state-of-the-art batch effect removal and integration methods.

Recent progress in Single-Cell Genomics has produced different library protocols and techniques for molecular profiling. We formulate a unifying, data-driven, integrative, and predictive methodology for different libraries, samples, and paired-unpaired data modalities. Our design of scAEGAN includes an autoencoder (AE) network integrated with adversarial learning by a cycleGAN (cGAN) network. The AE learns a low-dimensional embedding of each condition, whereas the cGAN learns a non-linear mapping between the AE representations. We evaluate scAEGAN using simulated data and real scRNA-seq datasets, different library preparations (Fluidigm C1, CelSeq, CelSeq2, SmartSeq), and several data modalities as paired scRNA-seq and scATAC-seq. The scAEGAN outperforms Seurat3 in library integration, is more robust against data sparsity, and beats Seurat 4 in integrating paired data from the same cell. Furthermore, in predicting one data modality from another, scAEGAN outperforms Babel. We conclude that scAEGAN surpasses current state-of-the-art methods and unifies integration and prediction challenges.

Melanoma skin cancer is considered as one of the most common diseases in the world. Detecting such diseases at early stage is important to saving lives. During medical examinations, it is not an easy task to visually inspect such lesions, as there are similarities between lesions. Technological advances in the form of deep learning methods have been used for diagnosing skin lesions. Over the last decade, deep learning, especially CNN (convolutional neural networks), has been found one of the promising methods to achieve state-of-art results in a variety of medical imaging applications. However, ConvNets' capabilities are considered limited due to the lack of understanding of long-range spatial relations in images. The recently proposed Vision Transformer (ViT) for image classification employs a purely self-attention-based model that learns long-range spatial relations to focus on the image's relevant parts. To achieve better performance, existing transformer-based network architectures require large-scale datasets. However, because medical imaging datasets are small, applying pure transformers to medical image analysis is difficult. ViT emphasizes the low-resolution features, claiming that the successive downsampling results in a lack of detailed localization information, rendering it unsuitable for skin lesion image classification. To improve the recovery of detailed localization information, several ViT-based image segmentation methods have recently been combined with ConvNets in the natural image domain. This study provides a comprehensive comparative study of U-Net and attention-based methods for skin lesion image segmentation, which will assist in the diagnosis of skin lesions. The results show that the hybrid TransUNet, with an accuracy of 92.11% and dice coefficient of 89.84%, outperforms other benchmarking methods.

Background Single-cell multi-omics technologies allow the profiling of different data modalities from the same cell. However, while isolated modalities only capture one view of the total information of a biological cell, an integrative analysis capturing the different modalities is challenging. In response, bioinformatics and machine learning methodologies have been developed for multi-omics single-cell analysis. Nevertheless, it is unclear if current tools can address the dual aspect of modality integration and prediction across modalities without requiring extensive parameter finetuning. Results We designed LIBRA, a Neural Network based framework, to learn a translation between paired multi-omics profiles such that a shared latent space is constructed. LIBRA is a state-of-the-art tool when evaluating the ability to increase cell-type (clustering) resolution in the latent space. When assessing the predictive power across data modalities, LIBRA outperforms existing tools. Finally, considering the importance of hyperparameters, we implemented an adaptative-tuning strategy, labelled aLIBRA, in the LIBRA package. As expected, adaptive parameter optimization significantly boosts the performance of learning predictive models from paired datasets. Additionally, aLIBRA provides parameter combinations balancing the integrative and predictive tasks. Conclusions LIBRA is a versatile tool, uniquely targeting both integration and prediction tasks of Single-cell multi-omics data. LIBRA is a data-driven robust platform that includes an adaptive learning scheme. Furthermore, LIBRA is freely available as R and Python libraries ( https://github.com/TranslationalBioinformaticsUnit/LIBRA ).

Modeling low level features to high level semantics in medical imaging is an important aspect in filtering anatomy objects. Bag of Visual Words (BOVW) representations have been proven effective to model these low level features to mid level representations. Convolutional neural nets are learning systems that can automatically extract high-quality representations from raw images. However, their deployment in the medical field is still a bit challenging due to the lack of training data. In this paper, learned features that are obtained by training convolutional neural networks are compared with our proposed hand-crafted HSIFT features. The HSIFT feature is a symmetric fusion of a Harris corner detector and the Scale Invariance Transform process (SIFT) with BOVW representation. The SIFT process is enhanced as well as the classification technique by adopting bagging with a surrogate split method. Quantitative evaluation shows that our proposed hand-crafted HSIFT feature outperforms the learned features from convolutional neural networks in discriminating anatomy image classes.

Educational data mining (EDM) evaluates and predicts students' performance that assists to discover important factors affecting students' academic performance and also guides educational managers to make appropriate decisions accordingly. The most common technique for discovering meaningful information from the educational database is classification. The accuracy of classification algorithms on educational data can be increased by applying feature selection algorithms. Feature selection algorithms help in selecting robots and meaningful features for predicting students' performance with high accuracy. This paper presents different EDM approaches for forecasting students' performance using different data mining techniques. In addition, this paper also presents an evaluation of recent classification algorithms and feature selection algorithms used in educational data mining. Furthermore, the paper will guide the researchers on new and possible dimensions in building a prediction model in EDM.

Biology has become a data-intensive science. Recent technological advances in single-cell genomics have enabled the measurement of multiple facets of cellular state, producing datasets with millions of single-cell observations. While these data hold great promise for understanding molecular mechanisms in health and disease, analysis challenges arising from sparsity, technical and biological variability, and high dimensionality of the data hinder the derivation of such mechanistic insights. To promote the innovation of algorithms for analysis of multimodal single-cell data, we organized a competition at NeurIPS 2021 applying the Common Task Framework to multimodal single-cell data integration. For this competition we generated the first multimodal benchmarking dataset for single-cell biology and defined three tasks in this domain: prediction of missing modalities, aligning modalities, and learning a joint representation across modalities. We further specified evaluation metrics and developed a cloud-based algorithm evaluation pipeline. Using this setup, 280 competitors submitted over 2600 proposed solutions within a 3 month period, showcasing substantial innovation especially in the modality alignment task. Here, we present the results, describe trends of well performing approaches, and discuss challenges associated with running the competition.

Abstract Unconventional resources have made a significant contribution to fossil energy supply to date, and some specific stimulation techniques have been used in their exploitation. For example, the use of scCO2 foam as a hydraulic fracturing stimulation fluid has sparked considerable interest due to its numerous advantages in terms of fracturing and production performance. The strength of scCO2 foam, an indicator of foam performance, highly depends on the formulation design, foaming properties and operating conditions. Due to complex nature of foam, the quantification of foam strength at downhole conditions is challenging. Specific screening and optimization processes are required to design high performance foam. Although the flow behavior (apparent viscosity) of foam has been extensively studied with empirical models, integrating some essential process parameters into the foam flow behavior evaluation remains challenging. In this study, we present an effective model that incorporates the benefits of a deep learning (DL) approach while taking into account the integration of specific process variables. Several input parameters such as surfactant types and concentration, salinity, polymer concentration, temperature and pressure were used in conjunction with foam quality and shear rate. To predict foam strength while taking the aforementioned parameters into account, a deep neural network (DNN) with optimized hyperparameters was developed. The experimental data for this purpose were obtained using a pressurized foam rheometer. An improved deep learning framework was developed and designed to learn the intrinsic relation among various parameters. The predictive study concludes that, the developed optimized DNN algorithm can provide a reliable and robust prediction with significantly high accuracy. When compared to a shallow network with a standard deviation of less than 5%, the developed optimal deep neural network increased average predictive accuracy to 95.64%. The regression coefficient in the optimized case was found to be nearly one with a low mean square error. The developed DNN algorithm is considered as an improved framework which encompasses several process variables and provides reliable and accurate prediction thus makes it suitable for further integration with fracturing simulator. It would also be helpful for optimizing fracturing process and improving foam formulations.

CO2 foam as a fracturing fluid for unconventional reservoir has been of huge interest due to its potential in solving various challenges related to conventional water-based fracturing. The rheological property of CO2 foam is a key factor to control the efficiency of fracturing process that is strongly influenced by different process parameters such as foam quality, temperature, pressure, and shear rate. The quantification of these parameters under reservoir conditions leads to the design of optimum injection strategy. However, the traditional modeling approaches are unable to provide fast and accurate prediction while considering combined effect of all these parameters. Here, we proposed a data driven approach based on supervised deep learning to estimate rheological property of CO2 foam as a function of foam quality, temperature, pressure, and shear rate. We exploit deep neural networks (DNNs) that are trained to learn the complex nonlinear aspects of the data. For the data generation, we performed a series of experiments for CO2 foams by varying different process variables. CO2 foams at different qualities were generated using conventional surfactant in a flow loop system and foam viscosity measurements were performed at HPHT under wide range of shear rate. The architecture of DNN was optimized to accurately estimate the foam apparent viscosity for given foam quality, temperature, pressure, and shear rate. The predictive capability of designed network is found to be significantly high, analyzed by regression coefficient approaching unity, low mean squared error, and low average absolute relative deviation (<1.8%). The designed neural network allows robust and accurate prediction of foam apparent viscosity at different foam qualities under various reservoir condition, which demonstrates its practicality for CO2 foam projects for fracturing unconventional reservoirs.

CO2 foam as a fracturing fluid for unconventional reservoir has been of huge interest due to its potential in solving various challenges related to conventional water-based fracturing. The rheological property of CO2 foam is a key factor controlling the efficiency of fracturing process and it is strongly influenced by different parameters such as foam quality, temperature, pressure and shear rate. The quantification of these parameters under reservoir conditions leads to the design of optimum injection strategy. However, the traditional modeling approaches are unable to provide fast and accurate prediction while considering combined effect of all these parameters. Here, we proposed a data driven approach based on supervised deep learning to estimate rheological property of CO2 foam as a function of foam quality, temperature, pressure, and shear rate. We exploit deep neural networks (DNNs) that are trained to learn the complex nonlinear aspects of the data. For the data generation, we performed a series of experiments for CO2 foams by varying different process variables. CO2 foams at different qualities were generated using conventional surfactant in a flow loop system and foam viscosity measurements were performed at HPHT under wide range of shear rate. The architecture of DNN was optimized to accurately estimate the foam apparent viscosity for given foam quality, temperature, pressure, and shear rate. The predictive capability of designed network is found to be significantly high, analyzed by regression coefficient approaching unity, low mean squared error, and low average absolute relative deviation (≪ 2.5%). The designed neural network allows robust and accurate prediction of foam apparent viscosity at different foam qualities under various reservoir condition, which demonstrates its practicality for CO2 foam projects for fracturing unconventional reservoirs.