Genomics Analysis

Although cost-effective and scalable, short-read sequencing can have limited accuracy when mapping complex genetic variants and exploring large structural variants, copy number variations, or epigenetics. Since long-read sequencing produces significantly longer sequences, it can provide higher accuracy in these use cases, with applications to oncology and other diseases.

By generating tens of thousands of base pairs, long reads provide higher accuracy for identifying longer polymorphisms, including indels and structural variants. Plus, they enable higher-accuracy genome assembly without a reference genome.

With NVIDIA Parabricks, researchers and developers can conduct long-read analysis to:

• Power high-throughput analysis, decrease speed, and enhance accuracy.  
• Enhance accuracy and ensure transparency by reproducing results from trusted open-source tools—including Minimap2 and DeepVariant.
• Address computational challenges of long-read sequencing by accelerating basecalling, alignment, and variant calling. 
• Use data from Oxford Nanopore and PacBio long-read sequencers.

2D and 3D visualizations of a simulated abdominal CT scan.

For well over a decade, scientists have used single-cell omics to better understand biology and disease. By looking at the individual-cell level, researchers can gain visibility into a wide spectrum of cellular states and how they interact with each other. This helps researchers understand gene expressions and identify unique states and rare cell types that may be linked to specific diseases. 

Bulk RNA-sequencing approaches typically pool RNA from cells or tissues to analyze in aggregate. Unlike bulk RNA-sequencing, which provides an average of cell expression across a sample, single-cell approaches provide granularity on a cellular level. As a result, single-cell omics provide more precise analysis between what’s happening to individual cells in control and disease samples.

With NVIDIA’s accelerated computing and AI platform for single-cell omics, researchers and developers can:

• Reduce analysis time for processing increasingly large single-cell datasets.
• Accelerate data processing, clustering, dimensionality, reduction, and regression with NVIDIA RAPIDS™ and RAPIDS-singlecell, developed by scverse.
• Accurately predict gene behavior and disease mechanisms through foundation models in BioNeMo™.

Before getting started with RAPIDS-singlecell, bioinformaticians and data scientists can try the NVIDIA AI Blueprint for Single-Cell Analysis to:

• Easily deploy and run single-cell analysis with NVIDIA RAPIDS without requiring local GPUs or self-managed cloud provisions.
• Test near-real-time data analysis—completing in minutes, not hours on GPU compared to CPU.
• Load and analyze an 11 million-cell dataset, leveraging Dask.

Image courtesy of Bruker Spatial Biology.

Although single-cell techniques have helped researchers understand diseases by evaluating cells on an individual level, they lack spatial context within the tissue surrounding these cells. With the introduction of spatial transcriptomics, researchers can use everything from relational data to imaging data to better understand gene expression and cell dynamics.  

Relational data provides context into where cells are located in relation to one another and makes it possible to overlay imaging data with molecular data. The localization of cells and how they interact within their environment is critical for research, particularly when looking at rare cell types. However, spatial omics provide more context than local cell interactions, showcasing how a disease progresses within a tissue’s architecture. As a result, scientists are able to gain previously unknown spatial context for rare cell types and disease progression.

With NVIDIA’s accelerated computing and AI platform for spatial transcriptomics, researchers and developers can:

• Enable novel methods of analysis by accelerating bottlenecks and increasing accuracy.
• Use generative AI for high-accuracy cell segmentation with VISTA-2D, an NVIDIA AI foundation model.
• Utilize cuCIM for accelerated image processing and data loading for the spatial industry and RAPIDS-singlecell to analyze the transcriptomics side of spatial. 
• Leverage NVIDIA GPUs in spatial analysis to reduce analysis time for processing large amounts of spatial data.

Image of a human brain, the hippocampus, using the CosMx Whole Transcriptome panel (commercially available in 2025). Image provided by Bruker Spatial Biology.