This article was accepted into the corpus but its outbound wikilinks were never NER-processed — typical at the deepest BFS hop or when the run's entity cap was reached. No expansion funnel to show.
| Tensor QTL | |
|---|---|
| Name | Tensor QTL |
| Developed by | Google Research, University of California, Los Angeles, Broad Institute |
| Initial release | 2018 |
| Programming language | Python, CUDA |
| License | MIT |
Tensor QTL
Tensor QTL is a computational tool for fast discovery of quantitative trait loci using tensor operations and GPU acceleration. It integrates statistical association testing, permutation strategies, and mixed-model corrections to map genetic variants to molecular phenotypes. Tensor QTL has been adopted by large consortia and research groups for high-throughput mapping in population-scale cohorts.
Tensor QTL was introduced to address scalability in quantitative trait loci mapping for projects such as the Genotype-Tissue Expression Project and the UK Biobank. The software combines ideas from genome-wide association studies applied in contexts like the 1000 Genomes Project, GTEx Consortium, and analyses by groups at the Broad Institute and Stanford University. Early adopters included researchers associated with the National Institutes of Health, Wellcome Trust, and academic centers such as Harvard University and University of California, Los Angeles.
Tensor QTL is implemented primarily in Python (programming language) with compute kernels accelerated via CUDA for NVIDIA GPUs and optional CPU fallbacks. The codebase interoperates with common genomics libraries and formats produced by tools from the Broad Institute like GATK and Picard (software), and with data structures used by Hail (software) and pandas (software). Packaging and distribution have used platforms like PyPI and continuous integration systems powered by Travis CI and GitHub workflows. Development and reproducibility practices draw from standards promoted by the Open Science Framework and groups such as the Allen Institute.
The core method uses matrix and tensor algebra to compute association statistics between genotypes and phenotypes, leveraging linear regression models similar to those used in PLINK and mixed-model approaches seen in BOLT-LMM. Permutation-based false discovery rate control follows strategies employed in large-scale studies like the ENCODE Project and the Roadmap Epigenomics Project. Covariate adjustment procedures reflect best practices from analyses by the GTEx Consortium and methods described in literature from research groups at Massachusetts Institute of Technology and Johns Hopkins University. For population structure correction, Tensor QTL can incorporate principal components derived with tools influenced by the Human Genome Project and methods used by the Framingham Heart Study.
Tensor QTL has been used in molecular QTL mapping for expression QTL studies inside the Genotype-Tissue Expression Project, methylation QTL analyses in cohorts like those assembled by the Wellcome Trust Case Control Consortium, and proteomic QTL discovery in initiatives associated with the European Molecular Biology Laboratory and the Protein Atlas. Clinical genetics groups at institutions such as Mayo Clinic, Mount Sinai Health System, and UCLA Health have employed Tensor QTL in translational studies. Large population cohorts that benefited include the UK Biobank, All of Us Research Program, and regional biobanks like the Estonian Biobank and deCODE genetics.
Benchmarks reported by the original authors compared runtime and memory usage against tools such as Matrix eQTL and FastQTL, demonstrating orders-of-magnitude speedups on NVIDIA GPUs used also in projects at Google Research and NVIDIA Research. Performance evaluations referenced hardware from vendors like Intel and NVIDIA and drew on benchmarking practices from high-performance computing centers such as the Lawrence Berkeley National Laboratory and Argonne National Laboratory. Reproducibility assessments mirrored protocols from the National Center for Biotechnology Information and community-driven benchmarking efforts like those organized by the Global Alliance for Genomics and Health.
Limitations include dependency on specialized hardware such as NVIDIA GPUs, challenges with integration in cloud environments run by providers like Amazon Web Services, Google Cloud Platform, and Microsoft Azure, and potential issues in very small sample sizes similar to concerns raised in studies from Institut Pasteur and Karolinska Institutet. Statistical caveats echo those from the GTEx Consortium about multiple testing and covariate confounding, and computational trade-offs parallel discussions in literature from Stanford University and the Max Planck Society.
Related and comparative tools include Matrix eQTL, FastQTL, BOLT-LMM, PLINK, Hail (software), and pipelines developed at the Broad Institute and University of California, San Diego. Extensions and wrappers have been developed in environments used by the European Bioinformatics Institute and the Wellcome Sanger Institute, and integrations into workflow managers such as Snakemake and Nextflow facilitate scaling on platforms supported by the European Grid Infrastructure and the Open Science Grid.
Category:Bioinformatics software