Institute for
Biomedical Informatics

Plants make up the biggest biotic component of the biosphere and play essential roles in all ecosystems. Our survival and well-being depend on plants and this dependence will increase as the climate changes rapidly. To improve how we obtain food, energy, and materials from plants and steward the health of our environment for future generations, we need to understand how plants work at multiple scales from molecules to cells to ecosystems. A major challenge to achieving this goal is a limited understanding of functions of plant genes. The majority of genes in plant genomes are uncharacterized and many of them are found only in plant lineages. Traditional sequence-similarity based biochemical function inference cannot address this challenge. Another aspect of gene function that is critical but generally lacking is the spatial and temporal context under which gene products operate. These challenges have, in part, driven the spectacular advances and inventions in genomics, imaging, mass spectrometry and we are now capable of high-throughput, high-content, and high-resolution measurements of gene and protein function parameters. Along with these technologies and emerging datasets, we need advances in computational biology and bioinformatics tools, concepts, and methods. In this talk, I will describe these challenges and some of the efforts we are making in addressing them.

Artificial intelligence is now recognized as an on-going paradigm shift, with an emphasis on machine learning especially deep learning. Computer vision and image analysis are two major applications of deep learning. While computer vision and image analysis deal with existing images and produce features of these images (images to features), tomographic imaging produces images of multi-dimensional structures from experimentally measured data (line integrals, harmonic components, and so on, of underlying images) which are tomographic features (features to images). Since 2016, deep learning is being actively developed worldwide for tomographic imaging, forming a new area of imaging research. In this presentation, we present a perspective on deep imaging involving data processing, image reconstruction, radiomics, and beyond. We show deeptomographic results and also explore network innovations.

Slides from presentations:

Perspective on Deep Imaging

Medical Imaging in the Deep Learning Framework

The biological sciences are currently experiencing an increase in data size and complexity. Traditional methods are now reaching the point they can no longer keep up with the data output. Larger datasets, particularly when combined with other data sources, open opportunities to apply methods like deep learning on these complex data. NVIDIA is contributing to these needs by developing algorithms that can leverage graphics processing units (GPUs). These problems benefit from high memory bandwidth and very high parallelism inherent of the GPU architecture. This talk will explore some of the collaborations we have with industry and academia and share some of what our research team is currently working on.

GPUs for Genomics