A Princeton University research team reports in Nature Machine Intelligence this week a new framework to facilitate the use of convolutional neural networks (CNNs) in genomics. Artificial neural networks such as CNNs represent powerful tools for analyzing biological sequences, but the need to tune network architectures, which is time-consuming and requires machine learning expertise, limits their application. To address this, the scientists developed Automated Modelling for Biological Evidence-based Research, or AMBER, which automates the design and application of optimal CNNs for genomic sequences through the state-of-the-art neural architecture search. Its developers show that AMBER-designed models outperform equivalent non-neural architecture search models, even published ones designed by experts, and demonstrate it using established benchmarks.
A single-cell atlas of early-stage human brain development is published in Nature Neuroscience this week, providing insights into the first trimester of human brain development and the subpopulations of progenitor cells that form the basis for creating the human cortex. To build the atlas, researchers from the University of California, San Francisco performed single-cell RNA sequencing across regions of the developing human brain including the telencephalon, diencephalon, midbrain, hindbrain, and cerebellum. They uncovered progenitor populations located near the telencephalon, suggesting more heterogeneity than previously known. A comparison of human and mouse progenitor populations at similar developmental timepoints, meanwhile, revealed two progenitor clusters that are enriched in the early stages of human cortical development.