We introduce PR-XAI, a feature attribution method for transformer models based on the PageRank algorithm.The proposed PR-XAI models the attention mechanism as a directed graph, with weights derived from attention weights and their gradients.Evaluations across five well-known text classification datasets and three different architectures show that PR-AG, one variant of PR-XAI, outperforms state-ofthe-art attribution methods in faithfulness and classification metrics, with significant gains on long-form text.

This paper introduces Generalized Attention Flow (GAF), a novel feature attribution method for Transformer-based models to address the limitations of current approaches. By extending Attention Flow and replacing attention weights with the generalized Information Tensor, GAF integrates attention weights, their gradients, the maximum flow problem, and the barrier method to enhance the performance of feature attributions. The proposed method exhibits key theoretical properties and mitigates the shortcomings of prior techniques that rely solely on simple aggregation of attention weights. Our comprehensive benchmarking on sequence classification tasks demonstrates that a specific variant of GAF consistently outperforms state-of-the-art feature attribution methods in most evaluation settings, providing a more reliable interpretation of Transformer model outputs.

MOTIVATION: The genome-wide chromosome conformation capture assay Hi-C is widely used to study chromatin 3D structures and their functional implications. Read counts from Hi-C indicate the strength of chromatin contact between each pair of genomic loci. These read counts are heteroskedastic: that is, a difference between the interaction frequency of 0 and 100 is much more significant than a difference between the interaction frequency of 1000 and 1100. This property impedes visualization and downstream analysis because it violates the Gaussian variable assumption of many computational tools. Thus heuristic transformations aimed at stabilizing the variance of signals like the shifted-log transformation are typically applied to data before its visualization and inputting to models with Gaussian assumption. However, such heuristic transformations cannot fully stabilize the variance because of their restrictive assumptions about the mean-variance relationship in the data. RESULTS: Here, we present VSS-Hi-C, a data-driven variance stabilization method for Hi-C data. We show that VSS-Hi-C signals have a unit variance improving visualization of Hi-C, for example in heatmap contact maps. VSS-Hi-C signals also improve the performance of subcompartment callers relying on Gaussian observations. VSS-Hi-C is implemented as an R package and can be used for variance stabilization of different genomic and epigenomic data types with two replicates available. AVAILABILITY AND IMPLEMENTATION: https://github.com/nedashokraneh/vssHiC.

MOTIVATION: The organization of the genome into domains plays a central role in gene expression and other cellular activities. Researchers identify genomic domains mainly through two views: 1D functional assays such as ChIP-seq, and chromatin conformation assays such as Hi-C. Fully understanding domains requires integrative modeling that combines these two views. However, the predominant form of integrative modeling uses segmentation and genome annotation (SAGA) along with the rigid assumption that loci in contact are more likely to share the same domain type, which is not necessarily true for epigenomic domain types and genome-wide chromatin interactions. RESULTS: Here, we present an integrative approach that annotates domains using both 1D functional genomic signals and Hi-C measurements of genome-wide 3D interactions without the use of a pairwise prior. We do so by using a graph embedding to learn structural features corresponding to each genomic region, then inputting learned structural features along with functional genomic signals to a SAGA algorithm. We show that our domain types recapitulate well-known subcompartments with an additional granularity that distinguishes a combination of the spatial and functional states of the genomic regions. In particular, we identified a division of the previously identified A2 subcompartment such that the divided domain types have significantly varying expression levels. AVAILABILITY AND IMPLEMENTATION: https://github.com/nedashokraneh/IChDA. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

MOTIVATION: Segmentation and genome annotation (SAGA) algorithms are widely used to understand genome activity and gene regulation. These methods take as input a set of sequencing-based assays of epigenomic activity, such as ChIP-seq measurements of histone modification and transcription factor binding. They output an annotation of the genome that assigns a chromatin state label to each genomic position. Existing SAGA methods have several limitations caused by the discrete annotation framework: such annotations cannot easily represent varying strengths of genomic elements, and they cannot easily represent combinatorial elements that simultaneously exhibit multiple types of activity. To remedy these limitations, we propose an annotation strategy that instead outputs a vector of chromatin state features at each position rather than a single discrete label. Continuous modeling is common in other fields, such as in topic modeling of text documents. We propose a method, epigenome-ssm-nonneg, that uses a non-negative state space model to efficiently annotate the genome with chromatin state features. We also propose several measures of the quality of a chromatin state feature annotation and we compare the performance of several alternative methods according to these quality measures. RESULTS: We show that chromatin state features from epigenome-ssm-nonneg are more useful for several downstream applications than both continuous and discrete alternatives, including their ability to identify expressed genes and enhancers. Therefore, we expect that these continuous chromatin state features will be valuable reference annotations to be used in visualization and downstream analysis. AVAILABILITY AND IMPLEMENTATION: Source code for epigenome-ssm is available at https://github.com/habibdanesh/epigenome-ssm and Zenodo (DOI: 10.5281/zenodo.6507585). SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

MOTIVATION: With the advancement of sequencing technologies, genomic data sets are constantly being expanded by high volumes of different data types. One recently introduced data type in genomic science is genomic signals, which are usually short-read coverage measurements over the genome. To understand and evaluate the results of such studies, one needs to understand and analyze the characteristics of the input data. RESULTS: SigTools is an R-based genomic signals visualization package developed with two objectives: (i) to facilitate genomic signals exploration in order to uncover insights for later model training, refinement and development by including distribution and autocorrelation plots; (ii) to enable genomic signals interpretation by including correlation and aggregation plots. In addition, our corresponding web application, SigTools-Shiny, extends the accessibility scope of these modules to people who are more comfortable working with graphical user interfaces instead of command-line tools. AVAILABILITY AND IMPLEMENTATION: SigTools source code, installation guide and manual is freely available on http://github.com/shohre73.

The availability of thousands of assays of epigenetic activity necessitates compressed representations of these data sets that summarize the epigenetic landscape of the genome. Until recently, most such representations were cell type-specific, applying to a single tissue or cell state. Recently, neural networks have made it possible to summarize data across tissues to produce a pan-cell type representation. In this work, we propose Epi-LSTM, a deep long short-term memory (LSTM) recurrent neural network autoencoder to capture the long-term dependencies in the epigenomic data. The latent representations from Epi-LSTM capture a variety of genomic phenomena, including gene-expression, promoter-enhancer interactions, replication timing, frequently interacting regions, and evolutionary conservation. These representations outperform existing methods in a majority of cell types while yielding smoother representations along the genomic axis due to their sequential nature.

Drug repurposing can accelerate the identification of effective compounds for clinical use against SARS-CoV-2, with the advantage of pre-existing clinical safety data and an established supply chain. RNA viruses such as SARS-CoV-2 manipulate cellular pathways and induce reorganization of subcellular structures to support their life cycle. These morphological changes can be quantified using bioimaging techniques. In this work, we developed DEEMD: a computational pipeline using deep neural network models within a multiple instance learning framework, to identify putative treatments effective against SARS-CoV-2 based on morphological analysis of the publicly available RxRx19a dataset. This dataset consists of fluorescence microscopy images of SARS-CoV-2 non-infected cells and infected cells, with and without drug treatment. DEEMD first extracts discriminative morphological features to generate cell morphological profiles from the non-infected and infected cells. These morphological profiles are then used in a statistical model to estimate the applied treatment efficacy on infected cells based on similarities to non-infected cells. DEEMD is capable of localizing infected cells via weak supervision without any expensive pixel-level annotations. DEEMD identifies known SARS-CoV-2 inhibitors, such as Remdesivir and Aloxistatin, supporting the validity of our approach. DEEMD can be explored for use on other emerging viruses and datasets to rapidly identify candidate antiviral treatments in the future. Our implementation is available online athttps://www.github.com/Sadegh-Saberian/DEEMD.

MOTIVATION: A sequencing-based genomic assay such as ChIP-seq outputs a real-valued signal for each position in the genome that measures the strength of activity at that position. Most genomic signals lack the property of variance stabilization. That is, a difference between 0 and 100 reads usually has a very different statistical importance from a difference between 1000 and 1100 reads. A statistical model such as a negative binomial distribution can account for this pattern, but learning these models is computationally challenging. Therefore, many applications-including imputation and segmentation and genome annotation (SAGA)-instead use Gaussian models and use a transformation such as log or inverse hyperbolic sine (asinh) to stabilize variance. RESULTS: We show here that existing transformations do not fully stabilize variance in genomic datasets. To solve this issue, we propose VSS, a method that produces variance-stabilized signals for sequencing-based genomic signals. VSS learns the empirical relationship between the mean and variance of a given signal dataset and produces transformed signals that normalize for this dependence. We show that VSS successfully stabilizes variance and that doing so improves downstream applications such as SAGA. VSS will eliminate the need for downstream methods to implement complex mean-variance relationship models, and will enable genomic signals to be easily understood by eye. AVAILABILITY AND IMPLEMENTATION: https://github.com/faezeh-bayat/VSS. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Segmentation and genome annotation (SAGA) algorithms are widely used to understand genome activity and gene regulation. These algorithms take as input epigenomic datasets, such as chromatin immunoprecipitation-sequencing (ChIP-seq) measurements of histone modifications or transcription factor binding. They partition the genome and assign a label to each segment such that positions with the same label exhibit similar patterns of input data. SAGA algorithms discover categories of activity such as promoters, enhancers, or parts of genes without prior knowledge of known genomic elements. In this sense, they generally act in an unsupervised fashion like clustering algorithms, but with the additional simultaneous function of segmenting the genome. Here, we review the common methodological framework that underlies these methods, review variants of and improvements upon this basic framework, and discuss the outlook for future work. This review is intended for those interested in applying SAGA methods and for computational researchers interested in improving upon them.

Motivation: The prediction of drug resistance and the identification of its mechanisms in bacteria such as Mycobacterium tuberculosis, the etiological agent of tuberculosis, is a challenging problem. Modern methods based on testing against a catalogue of previously identified mutations often yield poor predictive performance. On the other hand, machine learning techniques have demonstrated high predictive accuracy, but many of them lack interpretability to aid in identifying specific mutations which lead to resistance. We propose a novel technique, inspired by the group testing problem and Boolean compressed sensing, which yields highly accurate predictions and interpretable results at the same time. Results: We develop a modified version of the Boolean compressed sensing problem for identifying drug resistance, and implement its formulation as an integer linear program. This allows us to characterize the predictive accuracy of the technique and select an appropriate metric to optimize. A simple adaptation of the problem also allows us to quantify the sensitivity-specificity trade-off of our model under different regimes. We test the predictive accuracy of our approach on a variety of commonly used antibiotics in treating tuberculosis and find that it has accuracy comparable to that of standard machine learning models and points to several genes with previously identified association to drug resistance.

Summary: Segway performs semi-automated genome annotation, discovering joint patterns across multiple genomic signal datasets. We discuss a major new version of Segway and highlight its ability to model data with substantially greater accuracy. Major enhancements in Segway 2.0 include the ability to model data with a mixture of Gaussians, enabling capture of arbitrarily complex signal distributions, and minibatch training, leading to better learned parameters. Availability and implementation: Segway and its source code are freely available for download at http://segway.hoffmanlab.org. We have made available scripts (https://doi.org/10.5281/zenodo.802939) and datasets (https://doi.org/10.5281/zenodo.802906) for this paper's analysis. Contact: [email protected]. Supplementary information: Supplementary data are available at Bioinformatics online.

Graph smoothness objectives have achieved great success in semi-supervised learning but have not yet been applied extensively to unsupervised generative models. We define a new class of entropic graph-based posterior regularizers that augment a probabilistic model by encouraging pairs of nearby variables in a regularization graph to have similar posterior distributions. We present a three-way alternating optimization algorithm with closed-form updates for performing inference on this joint model and learning its parameters. This method admits updates linear in the degree of the regularization graph, exhibits monotone convergence and is easily parallelizable. We are motivated by applications in computational biology in which temporal models such as hidden Markov models are used to learn a human-interpretable representation of genomic data. On a synthetic problem, we show that our method outperforms existing methods for graph-based regularization and a comparable strategy for incorporating long-range interactions using existing methods for approximate inference. Using genome-scale functional genomics data, we integrate genome 3D interaction data into existing models for genome annotation and demonstrate significant improvements in predicting genomic activity.