This transcript captures a series of twelve lightning talks from a bioinformatics conference, showcasing a diverse range of new computational tools, software packages, and analytical workflows. The presentations cover topics including the inference of genetic ancestry from challenging molecular data (RAID), the characterization of microbial peptides in immunopeptidomics (MetaIPE), and the allele-specific analysis of DNA methylation using long-read sequencing (plasma). Several talks introduce new R/Bioconductor packages for specialized analyses, such as applying gene expression signatures (SigneR), predicting transcription factor binding (simple R), and efficiently annotating large-scale genomic variant data (GDS annotator). Other presentations focus on network biology, with applications in understanding sex-specific viral responses (demand) and deciphering protein regulatory mechanisms (CLEC12A network analysis). The session also highlights advancements in handling large-scale data, including a memory-efficient algorithm for comparative genomics (ExoLabel), multiscale analysis of spatial transcriptomics data (Voyager), and new workflows within the Galaxy platform for soil metagenomics (BioDigs). Finally, a new interactive tool for visualizing differential transcript usage from long-read RNA-seq data is presented (explain).
A Collection of Bioinformatics Lightning Talks on New Tools and Analyses
00:00:05 - Inferring Genetic Ancestry with the RAID Tool: This presentation provides an update on RAID (Reverse Ancestry Inference from challenging human molecular Data), a tool designed to infer genetic ancestry from various data types beyond standard genotypes.
Key Challenge: Standard ancestry inference requires high-quality, cancer-free genomic data. RAID aims to unlock vast public data resources by working with cancer-derived genomes and low-coverage data from protocols like RNA-seq, ATAC-seq, and ChIP-seq.
Method: RAID uses a "data synthesis" approach to assess the accuracy of inference for a specific molecular profile before making the final call. It leverages a reference population (e.g., 1000 Genomes) and can use methods like PCA or admixture analysis.
Results: The tool shows high accuracy (near 1.0) for whole-genome, whole-exome, RNA-seq, and ATAC-seq data, with good performance on ChIP-seq and bisulfite sequencing data as well.
Takeaway: RAID is an accurate and versatile tool for determining genetic ancestry from diverse and challenging biological samples, with a Galaxy version in development.
00:09:21 - MetaIPE: A Galaxy Pipeline for Meta-Immunopeptidomics: This talk introduces MetaIPE, a pipeline developed within the Galaxy platform to identify and characterize microbially-derived peptides that are bound to HLA molecules and presented to the immune system.
Key Challenge: Identifying non-host peptides in immunopeptidomics is difficult due to the massive search space and lack of enzymatic specificity, which lowers sensitivity.
Method: The MetaIPE pipeline uses a multi-step approach: 1) Initial peptide identification from mass spectrometry data using FragPipe. 2) A verification step using PepQuery against a human-only database to filter out false positives. 3) Taxonomic and functional annotation (UniProt, BLASTp). 4) Prediction of HLA binding.
Takeaway: The pipeline provides a rigorous, multi-step verification process to increase confidence in the identification of microbial peptides, prioritizing quality over quantity.
00:18:50 - Allele-Specific DNA Methylation Analysis with "plasma": A professor from Johns Hopkins discusses a new pipeline, "plasma," for analyzing allele-specific DNA methylation using long-read nanopore sequencing data from mouse crosses.
Key Challenge: Accurately analyzing methylation requires phasing (assigning reads to a parental chromosome) and comparing two different genomes, which involves complex bookkeeping that existing tools handle poorly.
Method: The "plasma" pipeline takes reads and two parental genomes as input, performs multiple alignments and phasing, and places all data into a common intermediate genome for analysis. This rescues about 10% more CpG sites for analysis compared to single-reference methods.
Results: The talk showcases how this method can distinguish between cis-regulatory effects (where methylation patterns are preserved in F1/F2 mice) and trans-acting effects (where methylation patterns are intermediate in F1 and variable in F2 mice). It also visualizes the difference between incomplete dominance and co-dominance at the single-read level.
00:28:32 - SigneR: A Package for Applying Gene Expression Signatures: This presentation introduces SigneR, an R package created to make it easy to compute and explore hundreds of published gene expression signatures.
Key Challenge: Re-implementing gene expression signatures from the literature is often difficult due to hidden steps or missing files.
Method: SigneR is a collection of over 70 pre-implemented signatures related to cancer and immune activity. It takes summarized experiment, single-cell, or spatial data as input and provides functions for quality control, visualization, and comparing signatures.
Results: A case study on a spatial breast cancer dataset demonstrates how SigneR can reveal biological patterns, such as high hypoxia scores in the tumor core and an epithelial-to-mesenchymal transition (EMT) signature at the tumor border.
00:37:41 - Network Analysis of Sex-Specific Viral Responses: This talk describes the use of a network disregulation algorithm called "demand" to study sex-specific host responses to influenza A virus infection, particularly the effects of estradiol.
Key Challenge: Identifying the molecular mechanisms behind sex differences in immune response.
Method: The "demand" algorithm analyzes disregulation in a protein-protein interaction network, propagating significance scores from gene expression changes. The speaker argues this is more powerful than performing network analysis on a list of differentially expressed genes.
Takeaway: Network disregulation analysis identified more relevant viral host factors than standard differential expression. The speaker is also developing a companion package, "TopGene," for rapid, broad enrichment analysis.
00:47:01 - simple R: Predicting Transcription Factor Binding Affinity: This presentation introduces simple R, an R package for predicting transcription factor (TF) binding and the impact of genetic variants on binding.
Key Challenge: There is a need for tools to prioritize and investigate the functional impact of the many SNPs that fall within TF binding sites on a large scale.
Method: The package uses pre-computed "Snip Effect Matrices" (SEMs), which are adjusted position-weighted matrices that better reflect binding affinity. Simple R provides functions to score binding of genomic regions, perform enrichment analysis for TF binding, and score the differential binding of SNP alleles.
Takeaway: simple R provides a user-friendly toolkit with built-in data for 223 common TFs, enabling large-scale analysis of TF binding and the functional effects of variants without needing complex command-line tools.
00:55:43 - GDS Annotator for Large-Scale Variant Analysis: This talk introduces GDS Annotator, an R package designed for rapid variant annotation in the Genomic Data Structure (GDS) format, targeting large biobank-scale datasets.
Key Challenge: Standard text-based formats like VCF are not optimized for the scale and complexity of modern whole-genome sequencing datasets from biobanks (e.g., UK Biobank).
Method: The GDS format is highly efficient for storing and querying genomic data, offering significant file size reduction (e.g., 3x smaller than pgen for UKB). GDS Annotator leverages this by providing fast, memory-efficient functions to annotate variants in GDS files using pre-formatted databases like dbNSFP, NOMAD, and outputs from VEP or SnpEff.
Takeaway: The package enables a streamlined, R-based workflow for performing large-scale rare variant association studies by integrating efficient storage (GDS) and rapid annotation.
01:04:35 - ExoLabel: A Scalable Algorithm for Comparative Genomics: This talk presents ExoLabel, a new, highly scalable community detection algorithm for building gene families from massive sequence similarity networks.
Key Challenge: Standard comparative genomics tools (e.g., OrthoFinder using MCL) require loading the entire sequence similarity network into RAM, which scales quadratically and becomes infeasible for thousands of genomes.
Method: Implemented in the Sinext R package, ExoLabel uses a disk-based approach, building an on-the-fly database to store network data. This results in memory usage that scales linearly with the number of genes (nodes) and is constant with the number of connections (edges).
Results: ExoLabel is not only extremely memory-efficient (processing 31 million nodes with <5 GB RAM) but is also more accurate and significantly faster (around 3x) than industry-standard alternatives like MCL.
01:15:14 - Multiscale Spatial Transcriptomics with Voyager: This presentation showcases a multiscale exploratory spatial data analysis (ESDA) using the Voyager R package on a lung adenocarcinoma dataset containing both Visium and Xenium data.
Key Challenge: Biological patterns exist at different spatial scales, and the results of spatial analysis can depend heavily on the resolution of the technology used.
Method: The speaker aggregated high-resolution Xenium data to various coarser resolutions, including pseudo-Visium spots, and computed spatial metrics (Moran's I for autocorrelation, local spatial correlation) at each scale.
Results: The analysis shows that spatial patterns can strengthen, weaken, or change entirely depending on the scale of observation. For example, a gene might show no pattern at the single-cell level due to heterogeneity but a strong pattern when aggregated into neighborhoods.
01:25:07 - Deciphering CLEC12A Regulation with Network Biology: This talk details a bioinformatic workflow used to investigate the molecular mechanism of CLEC12A, a receptor that negatively regulates neutrophil activation in arthritis.
Method: A protein-protein interaction network was built starting from CLEC12A and its first- and second-degree interactors, using data from BioGRID with specific filters. This network was then annotated with an internal phosphoproteomics dataset to identify proteins under CLEC12A's control.
Results: The analysis of first-degree interactors pointed strongly to vesicular trafficking and SNARE complexes. The analysis of the phosphoproteomics-annotated second-degree network highlighted the MAP Kinase pathway.
Takeaway: The integrated network analysis successfully generated the hypothesis, confirmed by experiments, that CLEC12A regulates neutrophil activation by downregulating the MAPK/ERK pathway.
01:34:37 - The BioDigs Project: Soil Metagenomics in Galaxy: This presentation describes the BioDigs project, a large-scale effort to characterize microbial diversity in soil across the U.S. while providing research opportunities for under-resourced institutions.
Key Challenge: A vast majority of microbial diversity in soil is uncharacterized ("microbial dark matter"), and standard tools often fail to classify most sequencing reads.
Method: The project collects soil samples, performs multi-platform sequencing (Illumina, PacBio, Nanopore), and uses the Galaxy platform for collaborative analysis. New tools like Silo, a more sensitive taxonomic classifier, were integrated into Galaxy to improve results. A long-read metagenome-assembled genome (MAG) pipeline is also being implemented.
Takeaway: The BioDigs project is expanding the toolkit within Galaxy for metagenomics to better explore complex environmental samples and enable broad scientific participation.
01:44:08 - explain: A Shiny App for Visualizing Isoform Switching: This final talk introduces explain, an interactive Shiny app for exploring the results of differential transcript usage (DTU) analysis, particularly from long-read RNA-seq.
Key Challenge: Understanding the functional consequences of isoform switches requires user-friendly tools to visualize changes at the transcript and exon levels.
Method: The app takes DTU results and visualizes the structures of differentially used isoforms, their relative proportions, and automatically detects events like skipped exons. It can analyze the sequence content of regulated regions to infer the binding motifs of RNA-binding proteins.
Results: Using a public dataset of PTBP1 knockdown, the app correctly shows that exons downregulated by PTBP1 have a strong upstream enrichment of C and U nucleotides, matching the known binding preference of the protein.