Integrating histopathology with spatial transcriptomics (ST) provides a powerful opportunity to link tissue morphology with molecular function. Yet most existing multimodal approaches rely on a small set of highly variable genes, which limits predictive scope and overlooks the coordinated biological programs that shape tissue phenotypes. We present PEaRL (Pathway Enhanced Representation Learning), a multimodal framework that represents transcriptomics through pathway activation scores computed with ssGSEA. By encoding biologically coherent pathway signals with a transformer and aligning them with histology features via contrastive learning, PEaRL reduces-dimensionality, improves interpretability, and strengthens cross-modal correspondence. Across three cancer ST datasets—breast, skin, and lymph node—PEaRL consistently outperforms SOTA methods, yielding higher accuracy for both gene- and pathway-level expression prediction (up to 58.9% and 20.4% increase in Pearson correlation coefficient compared to SOTA). These results demonstrate that grounding transcriptomic representation in pathways produces more biologically faithful and interpretable multi-modal models, advancing computational pathology beyond gene-level embeddings.

Recent advances in Spatial Transcriptomics (ST) pair histology images with spatially resolved gene expression profiles, enabling predictions of gene expression across different tissue locations based on image patches. This opens up new possibilities for enhancing whole slide image (WSI) prediction tasks with localized gene expression. However, existing methods fail to fully leverage the interactions between different tissue locations, which are crucial for accurate joint prediction. To address this, we introduce MERGE (Multi-faceted hiErarchical gRaph for Gene Expressions), which combines a multi-faceted hierarchical graph construction strategy with graph neural networks (GNN) to improve gene expression predictions from WSIs. By clustering tissue image patches based on both spatial and morphological features, and incorporating intra- and inter-cluster edges, our approach fosters interactions between distant tissue locations during GNN learning. As an additional contribution, we evaluate different data smoothing techniques that are necessary to mitigate artifacts in ST data, often caused by technical imperfections. We advocate for adopting gene-aware smoothing methods that are more biologically justified. Experimental results on gene expression prediction show that our GNN method outperforms state-of-the-art techniques across multiple metrics.

Gene expression analysis of samples with mixed cell types only provides limited insight to the characteristics of specific tissues. In silico deconvolution can be applied to extract cell type specific expression, thus avoiding prohibitively expensive techniques such as cell sorting or single-cell sequencing. Non-negative matrix factorization (NMF) is a deconvolution method shown to be useful for gene expression data, in part due to its constraint of non-negativity. Unlike other methods, NMF provides the capability to deconvolve without prior knowledge of the components of the model. However, NMF is not guaranteed to provide a globally unique solution. In this work, we present FaStaNMF, a method that balances achieving global stability of the NMF results, which is essential for inter-experiment and inter-lab reproducibility, with accuracy and speed. Results: FaStaNMF was applied to four datasets with known ground truth, created based on publicly available data or by using our simulation infrastructure, RNAGinesis. We assessed FaStaNMF on three criteria - speed, accuracy, and stability, and it favorably compared to the standard approach of achieving reproduceable results with NMF. We expect that FaStaNMF can be applied successfully to a wide array of biological data, such as different tumor/immune and other disease microenvironments.

MOTIVATION: High-throughput sequencing technologies enable the genome-wide analysis of the impact of genetic variation on molecular phenotypes at unprecedented resolution. However, although powerful, these technologies can also introduce unexpected artifacts. RESULTS: We investigated the impact of library amplification bias on the identification of allele-specific (AS) molecular events from high-throughput sequencing data derived from chromatin immunoprecipitation assays (ChIP-seq). Putative AS DNA binding activity for RNA polymerase II was determined using ChIP-seq data derived from lymphoblastoid cell lines of two parent-daughter trios. We found that, at high-sequencing depth, many significant AS binding sites suffered from an amplification bias, as evidenced by a larger number of clonal reads representing one of the two alleles. To alleviate this bias, we devised an amplification bias detection strategy, which filters out sites with low read complexity and sites featuring a significant excess of clonal reads. This method will be useful for AS analyses involving ChIP-seq and other functional sequencing assays. AVAILABILITY: The R package abs filter for library clonality simulations and detection of amplification-biased sites is available from http://updepla1srv1.epfl.ch/waszaks/absfilter

Reassortments are events in the evolution of the genome of influenza (flu). As reassortments have been implicated in major human pandemics of the last century, their identification has become a health priority. While such identification can be done on a small dataset by a biologist (using phylogenetic trees), databases of flu sequences are growing exponentially, so that it is imperative to develop automated identification methods. However, current methods are limited to pairwise segment comparisons. We present FluRF, a fully automated flu virus reassortment finder. FluRF is inspired by the visual approach to reassortment identification and thus uses the reconstructed phylogenetic trees of the individual segments and the full genome. We also present a simple flu evolution simulator, used to produce synthetic datasets to tune the FluRF parameters. On synthetic datasets produced by our flu evolution simulator, FluRF, tuned for a 0% false positive rate, yielded false negative rates of 10% or less. FluRF corroborated two new reassortments identified by visual analysis of 75 Human H3N2 New York flu strains from 2005-2008 and gave partial verification of reassortments found using another bioinformatics method. FluRF finds reassortments by a bottom-up search of the full-genome and segment-based phylogenetic trees for candidate clades-groups of one or more sampled viruses that are separated from the other variants from the same season. Candidate clades in each tree are tested to guarantee confidence values, using the lengths of key edges as well as other tree parameters; clades with reassortments must have validated incongruencies among segment trees. FluRF demonstrates robustness of prediction for geographically and temporally expanded datasets, and is not limited to finding reassortments with previously collected sequences. The complete source code is available from lcbb.epfl.ch/software.html.