Leveraging information in spatial transcriptomics to predict super-resolution gene expression from histology images in tumors

Recent developments in spatial transcriptomics (ST) technologies have enabled the profiling of transcriptome-wide gene expression while retaining the location information of measured genes within tissues. Moreover, the corresponding high-resolution hematoxylin and eosin-stained histology images are readily available for the ST tissue sections. Since histology images are easy to obtain, it is desirable to leverage information learned from ST to predict gene expression for tissue sections where only histology images are available. Here we present HisToGene, a deep learning model for gene expression prediction from histology images. To account for the spatial dependency of measured spots, HisToGene adopts Vision Transformer, a state-of-the-art method for image recognition. The well-trained HisToGene model can also predict super-resolution gene expression. Through evaluations on 32 HER2+ breast cancer samples with 9,612 spots and 785 genes, we show that HisToGene accurately predicts gene expression and outperforms ST-Net both in gene expression prediction and clustering tissue regions using the predicted expression. We further show that the predicted super-resolution gene expression also leads to higher clustering accuracy than observed gene expression. Gene expression predicted from HisToGene enables researchers to generate virtual transcriptomics data at scale and can help elucidate the molecular signatures of tissues.

Evaluation of Genotype-Based Gene Expression Model Performance: A Cross-Framework and Cross-Dataset Study

Predicting gene expression from genotyped data is valuable for studying inaccessible tissues such as the brain. Herein we present eGenScore, a polygenic/poly-variation method, and compare it with PrediXcan, a method based on regularized linear regression using elastic nets. While both methods have the same purpose of predicting gene expression based on genotype, they carry important methodological differences. We compared the performance of expression quantitative trait loci (eQTL) models to predict gene expression in the frontal cortex, comparing across these frameworks (eGenScore vs. PrediXcan) and training datasets (BrainEAC, which is brain-specific, vs. GTEx, which has data across multiple tissues). In addition to internal five-fold cross-validation, we externally validated the gene expression models using the CommonMind Consortium database. Our results showed that (1) PrediXcan outperforms eGenScore regardless of the training database used; and (2) when using PrediXcan, the performance of the eQTL models in frontal cortex is higher when trained with GTEx than with BrainEAC.

Statistical Transfer Learning with Generative Encoding for Spatial Transcriptomics

A new method proposes to align single cell reference datasets to spatial tran- scriptomic genes and tissue images. The technique can transfer single cells to unmeasured histology tissue by first aligning a single cell reference dataset to known Spatial Transcriptomic tissue, and learn from the alignment to predict gene expression on new histology. The model can invert the alignment transformation to generate new histology images from gene expression vectors, allowing for in-silico perturbation analyses through dynamically altering the levels of gene expression. Leveraging the cell atlas can lead to annotation of pathology and clinical specimens, enabling a mapping from the cellular and transcriptomic level to imaging tissue.

Chromatin states shaped by an epigenetic code confer regenerative potential to the mouse liver

We hypothesized that the highly controlled pattern of gene expression that is essential for liver regeneration is encoded by an epigenetic code set in quiescent hepatocytes. Here we report that epigenetic and transcriptomic profiling of quiescent and regenerating mouse livers define chromatin states that dictate gene expression and transposon repression. We integrate ATACseq and DNA methylation profiling with ChIPseq for the histone marks H3K4me3, H3K27me3 and H3K9me3 and the histone variant H2AZ to identify 6 chromatin states with distinct functional characteristics. We show that genes involved in proliferation reside in active states, but are marked with H3K27me3 and silenced in quiescent livers. We find that during regeneration, H3K27me3 is depleted from their promoters, facilitating their dynamic expression. These findings demonstrate that hepatic chromatin states in quiescent livers predict gene expression and that pro-regenerative genes are maintained in active chromatin states, but are restrained by H3K27me3, permitting a rapid and synchronized response during regeneration.

Applying cis-regulatory codes to predict conserved and variable heat and cold stress response in maize

Changes in gene expression are important for response to abiotic stress. Transcriptome profiling performed on maize inbred and hybrid genotypes subjected to heat or cold stress identifies many transcript abundance changes in response to these environmental conditions. Motifs that are enriched near differentially expressed genes were used to develop machine learning models to predict gene expression responses to heat or cold. The best performing models utilize the sequences both upstream and downstream of the transcription start site. Prediction accuracies could be improved using models developed for specific co-expression clusters compared to using all up- or down-regulated genes or by only using motifs within unmethylated regions. Comparisons of expression responses in multiple genotypes were used to identify genes with variable response and to identify cis- or trans-regulatory variation. Models trained on B73 data have lower performance when applied to Mo17 or W22, this could be improved by using models trained on data from all genotypes. However, the models have low accuracy for correctly predicting genes with variable responses to abiotic stress. This study provides insights into cis-regulatory motifs for heat- and cold-responsive gene expression and provides a framework for developing models to predict expression response to abiotic stress across multiple genotypes.One sentence summaryTranscriptome profiling of maize inbred and hybrid seedlings subjected to heat or cold stress was used to identify key cis-regulatory elements and develop models to predict gene expression responses.

Integrating long-range regulatory interactions to predict gene expression using graph convolutional neural networks

Long-range spatial interactions among genomic regions are critical for regulating gene expression and their disruption has been associated with a host of diseases. However, when modeling the effects of regulatory factors on gene expression, most deep learning models either neglect long-range interactions or fail to capture the inherent 3D structure of the underlying biological system. This prevents the field from obtaining a more comprehensive understanding of gene regulation and from fully leveraging the structural information present in the data sets. Here, we propose a graph convolutional neural network (GCNN) framework to integrate measurements probing spatial genomic organization and measurements of local regulatory factors, specifically histone modifications, to predict gene expression. This formulation enables the model to incorporate crucial information about long-range interactions via a natural encoding of spatial interaction relationships into a graph representation. Furthermore, we show that our model is interpretable in terms of the observed biological regulatory factors, highlighting both the histone modifications and the interacting genomic regions that contribute to a gene's predicted expression. We apply our GCNN model to datasets for GM12878 (lymphoblastoid) and K562 (myelogenous leukemia) cell lines and demonstrate its state-of-the-art prediction performance. We also obtain importance scores corresponding to the histone mark features and interacting regions for some exemplar genes and validate them with evidence from the literature. Our model presents a novel setup for predicting gene expression by integrating multimodal datasets.

Epigenome Prediction of Gene Expression using a Dynamical System Approach

Gene regulation is an important fundamental biological process. The regulation of gene expression is managed through a variety of methods including epigentic processes (e.g., DNA methylation). Understanding the role of epigenetic changes in gene expression is a fundamental question of molecular biology. Predictions of gene expression values from epigenetic data have tremendous research and clinical potential. Dynamical systems can be used to generate a model to predict gene expression using epigenetic data and a gene regulatory network (GRN). Here we present a novel stochastic dynamical systems model that predicts gene expression levels from methylation data of genes in a given GRN.

Predicting Gene Expression from DNA Sequence using Residual Neural Network

It is known that cis-acting DNA motifs play an important role in regulating gene expression. The genome in a cell thus contains the information that not only encodes for the synthesis of proteins but also is necessary for regulating expression of genes. Therefore, the mRNA level of a gene may be predictable from the DNA sequence. Indeed, three deep neural network models were developed recently to predict the mRNA level of a gene directly or indirectly from the DNA sequence around the transcription start side of the gene. In this work, we develop a deep residual network model, named ExpResNet, to predict gene expression directly from DNA sequence. Applying ExpResNet to the GTEx data, we demonstrate that ExpResNet outperforms the three existing models across four tissues tested. Our model may be useful in the investigation of gene regulation.